WO2022253288A1 - Methylation sequencing method and device - Google Patents

Abstract

Provided are a deep methylation sequencing method and device assisted by a machine learning classifier of a methylation pattern. The method and the device allow detection of a tumor-derived signal under the condition that the dilution factor is reduced to 10-4. The methylation sequencing method provided can detect the presence and/or content of low-frequency ctDNA in a sample, and can provide advantages in cancer screening and therapeutic effect evaluation.

Description

A method and device for methylation sequencing technical field

The present application relates to the field of biomedicine, in particular to a methylation sequencing method and device.

Background technique

In 2018, human cancers were responsible for 9.6 million deaths worldwide, most of which were diagnosed at an advanced stage. Intervention before distant metastases offers by far the greatest chance of improving prognosis, so the development of sensitive, reliable and minimally invasive tests to detect cancer before symptoms appear is highly desirable. Unfortunately, serum-based biomarkers are often limited to cancer surveillance (eg, carbohydrate antigen 19-9), but are not sensitive and/or specific enough for screening purposes.

Cell-free DNA (cfDNA) refers to degraded DNA fragments in the bloodstream, most of which are derived from normal white blood cells (WBC). In cancer patients, a fraction of cfDNA is of tumor origin (ctDNA), providing a snapshot of the cancer genome in real time. Mutational characterization of ctDNA has achieved exciting success for cancer diagnosis, prognosis, and monitoring. However, these methods are largely limited to advanced patients due to limited sensitivity or the need for biopsy to guide downstream analysis.

With the rapid development of next-generation sequencing (NGS) technologies, epigenetic analysis has gained significant attention. Compared with hunting for rare somatic mutations in blood, methylome analysis revealed several advantages: (i) aberrant methylation often occurs widely during cancer initiation, reflecting early changes in tumors; Specific genomic regions such as "CpG islands" are frequently found to be methylated, which provides a great opportunity to analyze large numbers of alterations by targeted sequencing; (iii) methylation status is cell-type specific and thus can be used to infer Tissue source of ctDNA.

Currently, NGS-based DNA methylation analysis techniques can be classified into two categories: bisulfite conversion-based methods (WGBS, RRBS) and enrichment-based methods (MeDIP, MBD-seq). Among them, bisulfite sequencing (BS-seq) is considered the gold standard for DNA methylation analysis because it provides quantification at single base resolution. Unfortunately, the harsh conditions of bisulfite conversion (BC) can cause enormous damage to DNA, limiting its use in blood-based applications. Furthermore, transformed DNA is often poor in sequence diversity, thus issues such as bias-prone target enrichment, and high sequencing errors further complicate scoring analysis.

To overcome these problems, the present applicant developed the MERMAID methylation detection method, which can be based on a sequencing method that maximizes the use of cfDNA and greatly reduces the artifacts of methylation sequencing. MERMAID, aided by a robust machine learning classifier, substantially outperformed other nonviable tissues in a proof-of-principle study for low-frequency ctDNA detection by ELSA-seq (see International Patent Publications WO2019/191900A1 and WO2019/192489A1) The method of examination provides new opportunities for advancing clinical applications.

Contents of the invention

To overcome these problems, the present applicants developed a MERMAID sequencing method. Aided by a robust machine learning classifier, MERMAID substantially outperformed other biopsy-free methods in a proof-of-principle study for low-frequency ctDNA detection, providing new opportunities to advance clinical applications.

In one aspect, the present application provides a method for detecting methylation modification of a target nucleic acid, the method comprising the following steps: Step (a-1) is based on the correlation coefficient of the CpG site in the target nucleic acid, the CpG site The methylation level of the point and the positional information of the CpG site determine the co-methylation block; and/or step (a-2) is based on the correlation coefficient of the CpG site in the target nucleic acid, the candidate co-methylation The amount of information in the methylation block and the division balance degree of the candidate co-methylation block, determining the co-methylation block, and step (b) based on the methylation of the co-methylation block The degree determines the presence and/or amount of the target nucleic acid in the sample to be tested.

In another aspect, the present application provides an analysis device for detecting the methylation modification of a target nucleic acid, the device comprising: a block division module (a-1), based on the correlation coefficient of the CpG site in the target nucleic acid, The methylation level of the CpG site and the position information of the CpG site determine a co-methylation block; and/or the block division module (a-2), based on the CpG site in the target nucleic acid The corrected correlation coefficient of the point, the information amount of the candidate co-methylation block and the division balance degree of the candidate co-methylation block, determine the co-methylation block, and the judgment module (b) based on the The methylation degree of the co-methylation block determines the presence and/or content of the target nucleic acid in the sample to be tested.

Methylation sequencing has attracted enormous interest because of its great potential to improve current ctDNA assays. Herein the present application presents MERMAID as a novel epigenetic analysis method characterized by well-conserved molecular diversity, robust noise suppression, and robust high-dimensional modeling. In addition to these properties, MERMAID may be particularly useful for blood-based applications: (i) a portion of cfDNA can be in single-stranded form, so this ssDNA-compatible approach can maximize the use of limited starting material, increasing the potential for rare Opportunities for ctDNA testing. (ii) Capture panels are designed with an excess of long RNA probes (>100 nucleotides long) complementary to various methylation patterns. This strategy is more tolerant to sequence-related biases and polymorphisms than amplicon-based target approaches (~20 nucleotides long). (iii) MERMAID does not require prior knowledge of the assay (eg, biopsied tissue), thus providing a solution for patients without surgically resected samples. Although this method has only been validated on LC, it could be customized to other types of cancer (e.g., CRC) or body fluids (e.g., urine). It can be extended to answer fundamental questions, such as tumor heterogeneity, or applied to other clinical scenarios, such as evaluating treatment effects.

Those skilled in the art can easily perceive other aspects and advantages of the present application from the following detailed description. In the following detailed description, only exemplary embodiments of the present application are shown and described. As those skilled in the art will appreciate, the content of the present application enables those skilled in the art to make changes to the specific embodiments which are disclosed without departing from the spirit and scope of the invention to which this application relates. Correspondingly, the drawings and descriptions in the specification of the present application are only exemplary rather than restrictive.

Description of drawings

The particular features of the invention to which this application relates are set forth in the appended claims. The features and advantages of the invention to which this application relates can be better understood with reference to the exemplary embodiments described in detail hereinafter and the accompanying drawings. A brief description of the accompanying drawings is as follows:

Fig. 1 illustrates the overview of the MERMAID method of the present application, including (D) to (E) in (A) to (E), wherein,

(A) Schematic illustration of the sample preparation procedure.

(B-C) Schematic representation of the library construction and targeted sequencing workflow. First, cfDNA fragments were denatured and transformed with sodium bisulfite (Lightning). A "tailing" (Tail&Tag-1) step is then performed by TdT (terminal deoxynucleotidyl transferase) to add an extra nucleotide ( Mainly dC, dark purple, just below Tail&Tag-1 on the right end). Splint junction 1 (pink and yellow, complementary double strands at the right end of the small fragment at the lower right of Tail&Tag-1) protrudes through a 5' protruding "arm" (mainly dG, light purple, at the left end of the small fragment at the lower right of Tail&Tag-1 Part) The ligation process is facilitated in the presence of E. coli ligase (dotted arc). A copy strand of the original template is then generated from the common anchor site (yellow, the right end of the lower two strands between Pre-amp and Tail&Tag-2) by a uracil-tolerant polymerase, followed by linker 2 (green and Blue, the ligation mediated by the left end of the small fragment at the lower left of Tail&Tag-2 (complementary double strand). Whole methylome sequencing libraries were generated by PCR amplification (PCR-1) and a dual indexing (barcoding) system was implemented to differentiate up to 384 samples. Use a library of biotin-labeled RNA probes (purple, short dark gray fragments below the strand directly below Hyb&Cap) and streptavidin beads (solid gray, spheres with dendrites above PCR-2) to bind Both strands of the DNA target are pulled down, and the captured molecule is then subjected to PCR amplification (PCR-2) and NGS sequencing.

(D) Schematic of deep sequencing-driven noise suppression and single-molecule-based methylation pattern recognition.

(E) Schematic illustration of machine learning-assisted classification.

Figure 2 schematically illustrates an exemplary WGBS library construction by the present application, showing a comparison of different WGBS protocols. Partially methylated E. coli (DH5α) DNA was sheared to ~200 bp to mimic cfDNA. A no-template control (NTC) was performed to assess the background readout (eg, residual adapter dimer) for each protocol, which was then subtracted from the library yield measured in each experiment. ELSA: An exemplary sequencing method of this application ELSA-seq, SWT: Accel-NGS Methyl-SEQ, NEB: NEBNext ultra II. Unless noted, experiments were performed with two technical replicates and two biological replicates, error bars represent one S.D. Fig. 2 includes (A) to (F), wherein,

(A) Final library yields obtained with 12, 12 and 16 PCR cycles for ELSA, SWT and NEB, respectively. NOTE: 12 cycles of NEB did not provide enough DNA for quantification.

(B) Unique read depths (with PCR duplicates removed) were observed with different input amounts (0.5, 1, 5 ng) when sequenced to ~2,000X median depth.

(C) Unique read depth observed by shallow sequencing libraries constructed with 500 pg of E. coli DNA at different depths. The genome size of DH5α is about 4.5 Mbp and 0.5 ng refers to about 100,000 haploid genome copies before BC. Normalization was performed by subtracting the unique read depths observed in NTCs.

(D) Graph showing overlap (depth >0) of methylated sites detected between 500 pg and 5 ng DNA input.

(E-F) Mutation spectrum and frequency observed without (E) or with (F) deep sequencing-induced error suppression. The same library was sequenced on Illumina Hiseq 2500, Novaseq 6000 and MiniSeq, respectively.

Figure 3 illustrates the design and performance of an exemplary target panel of the present application, including (A) to (E), wherein,

(A) Heatmap of methylation levels of 2,765 probe regions (Illumina 450K) in WBC (n = 656), cancer tissue (n = 4,539), and normal tissue (n = 521) based on GSE and TCGA databases.

(B-D) Performance of an exemplary target panel of the present application by using 10 ng cfDNA from a representative donor, wherein,

(B) Unique depth across all genomic regions in the target panel.

(C) Fragment sizes obtained using one of the exemplary library construction methods of the present application or the Truseq library construction method.

(D) Methylation level of each CpG site (iAF) calculated with "+" or "-" strand.

(E) Panel-wide technical background assessed with CHH methylation levels.

Figure 4 illustrates methylation block definition and pattern recognition, including (A) to (D), wherein,

(A) Representative genomic regions (CHR14:26674253-26674534) showing co-methylation status measured with Poisson correlation (left panel) or "block index" (right panel).

(B) Box plot showing statistical distribution of AUC for classifying malignant (n=48) and normal lung tissues (n=20) using the methylation status (mAF or iAF) of SHOX2 (CHR3: 157813800-157822158). All samples were sequenced at unique read depths of 500X (left panel) and 100X (right panel). Unless noted, graphs are marked with one asterisk (*), two asterisks (**) if P<0.05, three asterisks (***) if P<0.01, and P<0.001 . In this figure, P-values were calculated by using the Wilcoxon Sign test.

(C) Equation of MBS and examples of genomic loci with various methylation patterns.

(D) Density curves of mAF (left panel) and MBS (right panel) observed in methyltransferase-treated lambda spike-in experiments. Colored lines represent samples with different dilution ratios (technical replicates are colored the same), while log-transformed mAF or MBS values are shown on the x-axis.

Figure 5 illustrates the analytical validation of MERMAID, including (A) to (F), where,

(A) Poisson correlation (ρ) matrix of MBS values among different replicates, individuals, and diseases. Data were generated from 4 different healthy donors (H1-4) and 3 LC patients (C1-3), and were performed in triplicate for H1 (H1R1-R3). Computations are performed on the entire subgroup by using 8,312 blocks.

(B) Methylation signal detection with serial dilutions (0.0001-0.05) of CRC cell DNA. Error bars depict 95% CIs of estimated tumor fractions, while dotted lines indicate y=x.

(C) FDR measured by sequencing normal leukocytes at different depths and with different input amounts (X-axis). The Y axis represents the percentage of false positive calls.

(D) Bioinformatics (in sillico) sensitivity measured by mixing sequencing reads from CRC into normal cfDNA data at different ratios (0.00001–0.05). X-axis represents expected tumor fraction. The Y axis represents the percentage of positive calls. P values were calculated by t-test.

(E-F) Assay sensitivity of MERMAID (E), ddPCR (F, left panel) and HS-UMI (right panel) measured by sequential dilution of cancer cell (LC, CRC) DNA with normal leukocytes at a ratio of 0.0001–0.1. The Y-axis represents the percentage (E) or allele frequency (F) of positive markers observed. Dashed lines represent detection thresholds (95% CI) of ddPCR or HS-UMI for the indicated mutations.

Figure 6 shows an exemplary sequencing method of the present application and the adapter ligation efficiency of TruSeq, including (A) to (E), wherein,

(A) Schematic illustration of droplet digital PCR (ddPCR) used to estimate ligation efficiency. The synthesized DNA template was ligated to Adapter 1 or Adapter 2 by one of the exemplary sequencing methods of the present application. Primers were designed to amplify "total" (For.1-Rev.1) or "ligated" fragments (For.1-Rev.2). Reactions were arranged in separate wells with the same reporter fluorescent probe.

(B) Representative fluorescence curves of ddPCR reactions. Each dot represents a single PCR reaction (droplet) with one DNA molecule. Gray dots show background fluorescence and green dots are single droplets with amplified signal.

(C-E) ddPCR copy number curves (left panel) and ligation efficiency table (right panel) using ds TruSeq adapters (C), adapter 1 (D) and adapter 2 (E). For copy number curves, the Y-axis represents copies/μL, while the X-axis represents 3 technical replicates (T1, 2, 3) and 3 biological replicates (B1, 2, 3). Notably, the curves show counts per well, while the tables reflect counts normalized by reaction volume (Tail-Tag.1 vs. Tail-Tag.2).

Figure 7 illustrates schematically the principle of the library preparation method, including (A) to (B), wherein,

(A) Library construction by NEB (NEBNext Ultra II), SWT (Accel-NGS Methyl-Seq), TELP, ELSA (an exemplary sequencing method in this application), SPLAT, SALP (SALP-seq) and Padlock Schematic diagram of the workflow.

(B) Summary of library preparation methods for whole genome bisulfite sequencing (WGBS). Data were derived from the current study (NEB, SWT, ELSA) or previous studies (TELP, SPLAT, SALP). ds: double-stranded linker; ss: single-stranded linker; ds-TA: TA cloning-mediated linker attachment; ss-Tailing: synthetic tailing-mediated linker attachment; ss-guided: random guide-mediated linker attachment concatenation; chimera: A pair of reads combining two or more different partial sequences. Note: Padlock can only be used with targeted bisulfite sequencing.

Figure 8 illustrates the free capture performance of an exemplary sequencing method of the present application, including (A) to (G), wherein,

(A) Fragment size density curves using λ DNA with different input amounts.

(B) Genome-wide Poisson correlation (sliding window: 20,400 bp) of methylation levels using 0.5, 1 and 5 ng of E. coli (DH5α) DNA.

(C) ELSA by using 500pg (red, dark red, lower peak), 1 ng (green, blue-purple, middle peak) and 2ng cfDNA (yellow, aqua, upper peak) from two patients A representative library constructed. Each sample was prepared with technical replicates and a no-template control (NTC, blue line, flatter bottom line) was included.

(D) Representative libraries constructed by NEB, SWT and ELSA using 500 pg of E. coli DNA. Each sample was prepared with technical replicates and a no-template control (NTC, dark red line, line with higher peak) was included.

(E) Mutation spectrum and frequency of libraries constructed with NEB.

(F) Coverage distribution over the entire lambda genome when sequenced at ~2000X with different input amounts. Gray shading indicates unique depth across the genome, while black bars highlight coverage over CpG sites.

(G) The methylation signal of each sequencing cycle estimated by C/C+T (R1) and G/G+A (R2) on Illumina Novaseq 6000 with different amounts of λDNA (each group of histograms from left to The right order is: Hiseq, Miniseq, Novaseq).

Figure 9 illustrates the targeting performance of an exemplary sequencing method of the present application, including (A) to (G), wherein,

(A) Overlap of selected DMLs (n=80,672) with known human genome regions.

(B) Distribution of targeted CpG sites relative to classes of known gene signatures.

(C) Coverage uniformity curve showing the fraction of targeted bases (y-axis) with a unique depth equal to or greater than normalized coverage (x-axis). Here, normalized coverage was calculated as the observed unique read depth for each base divided by the average unique read depth for all targeted bases.

(D) Density curve showing GC content versus normalized read depth. The Y-axis represents the unique depth over the panel region, while the X-axis indicates the GC content of the reference genome.

(E) Sequencing quality (Phred score) of reads 1 and 2 measured at each sequencing cycle on an Illumina Novaseq 6000.

(F) Original cfDNA (blue, dotted line with peak marked to the left), amplified cfDNA without capture (Pre-Lib, red, dotted line with peak marked to the center) and post-capture amplification from two representative donors Size distribution of cfDNA (post-lib, dark red, dotted line marking peaks to the right). The shift of the peak (dotted line) to higher molecular weight reflects the attachment of adapter/linker sequences to the template during PCR-1 (80bp, left dashed line) and PCR-2 (69bp, middle dashed line).

(G) Unique depth of each CpG site on the "+" or "-" strand.

Figure 10 illustrates a depiction of methylation blocks and patterns, including (A) to (E), wherein,

(A-B) Determination of penalty coefficients α2(A) and α1(B) for block separation.

(C) Distribution of block lengths (in base pairs) across capture panels.

(D) Distribution of the number of CpG sites per block on the capture panel.

(E) Methylation status (gray) and unmethylation status (black) illustrated by each unique sequencing read (vertical line) within a representative SHOX2 region (chr3:157821291-157821596).

Figure 11 illustrates the reproducibility of MERMAID, including (A) to (C), where,

(A) Scatterplot showing intra-assay consistency of iAF (left panel) and MBS (right panel) by using the same NA12878 DNA sample (10 ng).

(B) Scatterplot of MBS values observed in two within-batch replicates (left panel) and between-batch replicates (right panel) by using the same cfDNA sample (10 ng).

(C) Density curves of MBS values by using different NA12878 DNA input amounts (2, 5, 10, 30 ng) or at different sequencing depths (1000-5000X).

Figure 12 illustrates the accuracy of MERMAID, including (A) to (D), where,

(A) IGV (Integrated Genomics Viewer) visualization of MERMAID data for H2122, H2228 and NA12878 cells at the SHOX2 and SEPT9 loci. For forward reads, a red "C" indicates a methylated "C" (not converted), while a blue "T" indicates an unmethylated C (C->T converted). For reverse reads, the interpretation is reversed.

(B) The methylation of SHOX2 and SEPT9 was verified by qMSP (quantitative methylation-specific PCR). The Y-axis represents the relative fold change in methylation level (msp_PCR on the left of each group of histograms and NGS on the right).

(C) Density curve showing iAF differences measured in NA12878 cells with MERMAID and Illumina Epic TruSeq (public data). Only overlapping CpG sites (n=32,383) were used. The dashed red line indicates the median of platform-dependent iAF differences.

(D) Scatter plot of log-transformed iAF generated with Illumina Epic TruSeq Methyl (Y-axis) and MERMAID (X-axis).

Figure 13 illustrates the FDR and LoD of MERMAID, including (A) to (F), where,

(A) Box plot showing the dependence of FDR (Y-axis) and sequencing depth (Y-axis) by using 10 ng of cfDNA.

(B) Numerical simulation of sampling noise via a Poisson distribution (unique segments = 500, noise = 0.002). The X-axis represents the number of observations (depth), while the Y-axis represents the variance.

(C) Detection of tumor-derived methylation counts as a function of tumor burden (0.000001-0.001), unique depth (left panel, markers=1000) and number of markers (right panel, depth=500). The X-axis represents the pre-defined tumor fraction θ _i in each simulated "tumor" sample and the Y-axis represents the average detection of that sample over 10,000 replicates. The probability of observed success was defined as P<0.05, and P values were determined by testing the null hypothesis of θ _i =0 using a likelihood ratio test.

(D) Evaluation of batch effects for bioinformatics LoD assays. Samples A (left) and B (right) came from two different healthy donors and were sequenced in two different rounds. Read spiking and analysis were performed as described in the Methods section of Example 1.

(E-F) Fluorescence curves of ddPCR showing detection of EGFR p.G719S mutation (A) and EML4-ALK fusion (B) in spike experiments in SW48 and NCI-H2228 cell lines, respectively. In each panel, gating thresholds are indicated by solid horizontal lines (eg: 6000, 2400, 3000, and 2760).

Figure 14 illustrates tissue-based selection and classification of LC-specific markers, including (A) to (D), wherein,

(A) Volcano plots of MBS measured in LC tissue (n=48) versus normal lung tissue (n=20, left panel) and LC tissue versus healthy cfDNA (n=30, right panel). Dark dots indicate important markers and light gray dots are unimportant markers.

(B) Violin plot showing MBS over representative specifiers in LC, normal lung, and control cfDNA samples. Each point represents a sample.

(C) Supervised classification of tumor and normal tissue samples with different numbers of specifiers. The Y-axis represents prediction accuracy across 100 replicates and the X-axis represents the number of specifiers.

(D) PCA (principal component analysis) clustering of tumor and normal tissue samples with different numbers of specifiers.

Figure 15 graphically illustrates the clinical characteristics of the validation cohort (plasma). The table shows 308 LC patients and 261 non-cancer controls stratified by clinical characteristics. UNK: Unknown. LUAD: lung adenocarcinoma; LUSC: lung squamous cell carcinoma.

Figure 16 illustrates the side-by-side comparison of MERMAID, HS-UMI and patient-specific ddPCR, including (A) to (F), wherein,

(A) Study designs for "double," "triple," and "quadruple" comparisons.

(B-D) ddPCR fluorescence curves showing the detection of EGFR P.S746_I750DEL for patient P65 (B), KRAS p.G12D for patient P64 (C) and EGFR P.L858R for patient P66 (D). The DNA input for each ddPCR reaction was 38, 62 and 54 ng, respectively. NTC: no template control; NC: normal WBC; PC: positive control with the desired mutation at 0.1% or 0.5% AF (Multiple I cfDNA reference standard set, Horizon Discovery). Notably, 0.5% PC was generated by mixing WT and 1% reference DNA in a 1:1 ratio. Details are provided in Table 7.

(E) maxAF for all participants in Group 1 and Group 2 by disease status (n=115).

(F) Predicted scores for all participants in Group 1 and Group 2 by maxAF (n=115). The dotted line represents the threshold (positive or negative) at 96% training specificity.

Figure 17 graphically illustrates the performance results for detecting cancer-associated changes based on the methylation level iAF of a single site, and based on the average methylation level mAF (mean methylation allele frequency) per block.

Figure 18 graphically illustrates the results of the relationship between the "regional median length" and "regional length variation coefficient" for the values of α ₂ and α ₁ , respectively.

Figure 19 graphically illustrates the density curves of the MBS statistic and the mAF statistic in samples of different blends.

Detailed ways

The implementation of the invention of the present application will be described in the following specific examples, and those skilled in the art can easily understand other advantages and effects of the invention of the present application from the content disclosed in this specification.

Definition of Terms

In this application, the terms "next-generation gene sequencing (NGS)", high-throughput sequencing" or "next-generation sequencing" generally refer to the second-generation high-throughput sequencing technology and higher-throughput sequencing methods developed thereafter. Next-generation sequencing platforms include but are not limited to existing sequencing platforms such as Illumina. With the continuous development of sequencing technology, those skilled in the art can understand that other sequencing methods and devices can also be used for this method. For example, two Generation gene sequencing can have the advantages of high sensitivity, high throughput, high sequencing depth, or low cost. According to the development history, influence, sequencing principles and technologies, there are mainly the following types: Massively Parallel Signature Sequencing (Massively Parallel Signature Sequencing, MPSS), Polony Sequencing, 454pyro sequencing, Illumina (Solexa) sequencing, Ion semi conductor sequencing, DNA nano-ball sequencing , Complete Genomics' DNA nanoarray and combined probe anchor ligation sequencing method, etc. The second-generation gene sequencing can make it possible to analyze the transcriptome and genome of a species in detail, so it is also called deep sequencing ( deep sequencing). For example, the method of the present application can also be applied to first-generation gene sequencing, second-generation gene sequencing, third-generation gene sequencing or single molecule sequencing (SMS).

In this application, the term "sample to be tested" generally refers to a sample that needs to be tested. For example, it can be detected whether one or more gene regions on the sample to be tested are modified.

In this application, the term "complementary region" generally refers to a region that is complementary to a reference nucleotide sequence. For example, a complementary nucleic acid can be a nucleic acid molecule that optionally has an opposite orientation. For example, the complementary may refer to having the following complementary associations: guanine and cytosine; adenine and thymine; adenine and uracil.

In this application, the term "hybridization" generally refers to a reaction in which one or more polynucleotides react to form a complex stabilized by hydrogen bonds between the bases of the nucleotide residues. Hydrogen bonding can occur through Watson-Crick base pairing, Hoogstein binding, or in any other sequence-specific manner based on base complementarity. The complex may comprise two strands forming a double helix, three or more strands forming a multi-strand complex, self-hybridizing single strands, or any combination of these. The hybridization reaction may constitute a step in a wider method, such as the initiation of PCR or the enzymatic cleavage of polynucleotides by endonucleases. A second sequence that is completely complementary to a first sequence or that is polymerized by a polymerase using the first sequence as a template is said to be "complementary" to said first sequence. The term "hybridizable" as applied to a polynucleotide refers to the ability of a polynucleotide to form complexes that are stabilized by hydrogen bonds between the bases of the nucleotide residues in a hybridization reaction. In some embodiments, a hybridizable nucleotide sequence is at least about 50%, 60%, 70%, 75%, 80%, 85%, 90%, 95%, or 100% complementary to the sequence to which it hybridizes.

Terminology In this application, the terms "polynucleotide", "nucleotide", "nucleic acid" and "oligonucleotide" are used interchangeably. They represent polymeric forms of nucleotides (deoxyribonucleotides or ribonucleotides) of any length, or analogs thereof. A polynucleotide can have any three-dimensional structure and can perform any function, whether known or unknown. The following are non-limiting examples of polynucleotides: coding or non-coding regions of genes or gene segments, loci (loci) defined by linkage analysis, exons, introns, messenger RNA (mRNA), translocation RNA (tRNA), ribosomal RNA (rRNA), short interfering RNA (siRNA), short-hairpin RNA (shRNA), microRNA (miRNA), ribozyme, cDNA, recombinant polynucleotide, branched polynucleotide, Plasmids, vectors, isolated DNA of any sequence, isolated RNA of any sequence, nucleic acid probes, primers and linkers. A polynucleotide may include one or more modified nucleotides, such as methylated nucleotides and nucleotide analogs.

In the present application, the term "modification state" generally refers to the modification state of the gene fragment, nucleotide or its base in the present application. For example, the modification state in the present application may refer to the modification state of cytosine. For example, a gene segment of the present application having a modified state may have altered gene expression activity. For example, the modification status of the present application may refer to the methylation modification of a base. For example, the modified state in this application may refer to the covalent bonding of a methyl group at the 5' carbon position of cytosine in the CpG region of genomic DNA, for example, it may become 5-methylcytosine (5mC). For example, a modification state can refer to the presence or absence of 5-methylcytosine ("5-mCyt") within the DNA sequence.

In this application, the term "methylation" generally refers to the methylation state of a gene fragment, nucleotide or its base in this application. For example, the DNA fragment where the gene in this application is located may have methylation on one strand or multiple strands. For example, the DNA fragment where the gene in this application is located may have methylation at one site or multiple sites.

In this application, the term "transformation" generally refers to the transformation of one or more structures into another structure. For example, the transformations of the present application can be specific. For example, cytosine without methylation modification can be converted into other structures (such as uracil), and cytosine with methylation modification can be converted substantially unchanged. For example, cytosine without methylation modification can be cleaved after conversion, and cytosine with methylation modification can be substantially unchanged after conversion.

In this application, the term "bisulfite", or "bisulfite" generally refers to a reagent that can distinguish DNA regions with and without modification states. For example, the bisulfite may include bisulfite, or an analog thereof, or a combination thereof. For example, bisulfite can deaminate the amino group of unmodified cytosine to distinguish it from modified cytosine. In this application, the term "analogue" generally refers to a substance having a similar structure and/or function. For example, analogs of bisulfite may have a similar structure to bisulfite. For example, an analog of bisulfite may refer to a reagent that can also distinguish between DNA regions that have a modified state and those that do not.

In this application, the term "comprising" generally means including specifically specified features, but not excluding other elements.

In this application, the term "about" generally refers to a range of 0.5%-10% above or below the specified value, such as 0.5%, 1%, 1.5%, 2%, 2.5%, above or below the specified value. 3%, 3.5%, 4%, 4.5%, 5%, 5.5%, 6%, 6.5%, 7%, 7.5%, 8%, 8.5%, 9%, 9.5%, or 10%.

Detailed description of the invention

In one aspect, the present application provides a method for detecting methylation modification of a target nucleic acid, the method comprising the following steps: Step (a-1) is based on the correlation coefficient of the CpG site in the target nucleic acid, the CpG site The methylation level of the target nucleic acid and the position information of the CpG site determine the co-methylation block; and/or step (a-2) is based on the correlation coefficient of the CpG site in the target nucleic acid, the candidate co-methylation The amount of information in the methylation block and the division balance degree of the candidate co-methylation block, determining the co-methylation block, and step (b) based on the degree of methylation of the co-methylation block Determine the presence and/or content of the target nucleic acid in the sample to be tested.

For example, the method detects the presence and/or content of the target nucleic acid with methylation modification in the test sample. As used herein, the "subject" from which the sample to be tested can be a mammal, such as a non-primate (for example, cow, pig, horse, cat, dog, rat, etc.) or a primate (for example, monkey or person). In some embodiments, the subject is a human. In some embodiments, the subject is a mammal (eg, a human) suffering from or potentially suffering from a disease, disorder or condition, examples of which are described herein. In some embodiments, the subject is a mammal (eg, a human) at risk of developing a disease, disorder or condition, examples of which are described herein.

For example, the correlation coefficient of the CpG sites comprises a Pearson correlation coefficient between two or more of the CpG sites in the target nucleic acid.

For example, said level of methylation of said CpG sites comprises a difference in mean methylation allele frequency (mAF) between two or more of said CpG sites in said target nucleic acid. For example, the methylation level of the CpG sites comprises the ratio of the difference in mAF to the sum of mAF between two or more of the CpG sites in the target nucleic acid.

For example, the location information of the CpG sites comprises differences in genomic locations between two or more of the CpG sites in the target nucleic acid. For example, the position information of the CpG sites comprises the ratio of the genomic position distance between two or more of the CpG sites in the target nucleic acid to the length of the target nucleic acid.

其中，ρ _ij表示皮尔逊相关系数，E(y _i)表示位点i处所有样品的平均甲基化等位基因频率(mAF)，E(y _j)表示位置j处所有样品的平均甲基化等位基因频率(mAF)，pos _i表示位点i的基因组位置，pos _j表示位点j的基因组位置，L表示所述目标核酸区域的长度，而λ ₁和λ ₂相互独立地选自0或更大的数。例如，所述λ ₁的取值范围为0至1。例如，所述λ ₂的取值范围为0至1。例如，本申请中λ ₁和λ ₂相互独立地选自0。例如，例如，本申请中λ ₁选自0；例如，本申请中λ ₂选自0。 For example, the step (a-1) includes: determining the corrected correlation coefficient between every two CpG sites of the target nucleic acid, the corrected correlation coefficient d _ij between site i and site j Calculated by the following formula:

where _ρij represents the Pearson correlation coefficient, E(y _i ) represents the average methylated allele frequency (mAF) of all samples at site i, and E(y _j ) represents the average methylated allele frequency (mAF) of all samples at position j. λ allele frequency (mAF), pos _i represents the genomic position of locus i, pos _j represents the genomic position of locus j, L represents the length of the target nucleic acid region, and λ ₁ and λ ₂ are independently selected from 0 or greater number. For example, the value range of _λ1 is 0-1. For example, the value range of λ ₂ is 0 to 1. For example, in the present application, _λ1 and _λ2 are independently selected from 0. For example, for example, _λ1 is selected from 0 in the present application; for example, _λ2 is selected from 0 in the present application.

For example, the correlation coefficient based on the CpG site in the target nucleic acid in the step (a-2) comprises a corrected correlation coefficient of the CpG site in the target nucleic acid, the CpG site in the target nucleic acid The corrected correlation coefficient comprises between two or more of the CpG sites in the target nucleic acid corrected based on the methylation level of the CpG sites and/or the position information of the CpG sites Pearson's correlation coefficient.

For example, the corrected correlation coefficient of the CpG site in the target nucleic acid comprises the corrected correlation coefficient in the above step (a-1) method.

For example, the information amount of the candidate co-methylation block includes the quantity information of the CpG sites of the candidate co-methylation block.

For example, the degree of partition balance of the candidate co-methylation blocks includes differences in the number of CpG sites of different candidate co-methylation blocks. For example, the degree of partition balance of the candidate co-methylation blocks includes different coefficients of variation of the numbers of the CpG sites of the candidate co-methylation blocks.

的所述区块指数

通过以下公式计算：

For example, the step (a-2) includes: maximizing the block index of the candidate co-methylation block, determining the co-methylation block, the candidate co-methylation in the target nucleic acid block

The block index of

Calculated by the following formula:

B _i表示第i个候选共甲基化区块的所述CpG位点的数量，α ₁和α ₂相互独立地选自0或更大的数。例如，通过唯一断点的迭代法确定使得所述共甲基化区块的区块断点。例如，所述α ₁的取值范围为0至10。例如，所述α ₂的取值范围为0至10。例如，本申请中，α ₁和α ₂的取值范围相互独立地选自0至10的有理数。例如，本申请中α ₁和α ₂相互独立地选自0。例如，例如，本申请中α ₁选自0、1、2、3、4、5、6、7、8、9或10；例如，本申请中α ₂选自0、1、2、3、4、5、6、7、8、9或10。 in,

B _i represents the number of the CpG sites of the ith candidate co-methylation block, and α ₁ and α ₂ are independently selected from 0 or greater numbers. For example, the block breakpoints for the co-methylated blocks are determined by an iterative method of unique breakpoints. For example, the value range of _α1 is 0-10. For example, the value of _α2 ranges from 0 to 10. For example, in the present application, the value ranges of _α1 and _α2 are independently selected from rational numbers from 0 to 10. For example, in the present application, _α1 and _α2 are independently selected from 0. For example, for example, in the present application, α ₁ is selected from 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 or 10; For example, in the present application, α ₂ is selected from 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 or 10.

For example, the step (b) comprises: the length of the continuous CpG of each sequencing read (read) based on the co-methylation block, the number of CpGs on the read, and the co-methylation region The total number of reads of the block determines the presence and/or amount of the target nucleic acid.

For example, the step (b) includes: determining the methylation block score of the co-methylation block, and the methylation block score MBS is calculated by the following formula:

Wherein, the n is the total number of reads covering all CpG sites of the co-methylation block, L _i is the number of CpG sites contained on the i-th read, and l _ij is the continuous methylation CpG on the i-th read The length of the site, m is the sequencing depth on the i-th read.

For example, UMI correction is applied to the summary of the sequencing data of the samples to be tested.

For example, the method further includes: extracting the feature value of the MBS of the co-methylation block of the tumor sample and the healthy sample through a machine learning model, and determining the MBS based on the MBS of the co-methylation block in the sample to be tested. The presence and/or amount of said target nucleic acid.

An analysis device for detecting the methylation modification of a target nucleic acid, the device comprising: a block division module (a-1), based on the correlation coefficient of the CpG site in the target nucleic acid, the methyl group of the CpG site The methylation level and the position information of the CpG site determine the co-methylation block; and/or the block division module (a-2), based on the corrected correlation coefficient of the CpG site in the target nucleic acid, the candidate The amount of information of the co-methylation block and the division balance degree of the candidate co-methylation block, determining the co-methylation block, and the determination module (b) based on the co-methylation block The degree of methylation determines the presence and/or amount of the target nucleic acid in the sample to be tested.

For example, the analytical device for detecting the methylation modification of the target nucleic acid of the present application may comprise the steps of implementing the method for detecting the methylation modification of the target nucleic acid of the present application.

As an aspect of the present disclosure, the present application provides a method for detecting ctDNA using methylation sequencing, which includes: selecting differentially methylated CpG sites; based on the similarity of the methylation status of the CpG sites, CpG sites are separated into multiple co-methylation blocks; samples are sequenced to obtain methylation sequencing reads; the average methylation level of each co-methylation block of samples is detected for further DNA analysis Methylation analysis.

In a preferred embodiment of the present disclosure, differentially methylated CpG sites are selected from the TCGA database generated by the Infinium HumanMethylation 450K array.

In another preferred embodiment of the present disclosure, the average methylation level is the average methylation allele frequency.

In another preferred embodiment of the present disclosure, the number of co-methylation blocks is between 1/30 and 1/5 of the number of CpG sites.

In another preferred embodiment of the present disclosure, co-methylated blocks are further restricted by comparing a specific tumor sample with a normal tissue sample, eg, a primary lung tumor with a normal lung tissue sample.

In another preferred embodiment of the present disclosure, co-methylated blocks of insufficient depth (<100) on most samples (>80%) are excluded from downstream analysis.

In another preferred embodiment of the present disclosure, said co-methylation blocks are separated based on a modified correlation matrix called "block index".

In another preferred embodiment of the present disclosure, the method further comprises normalizing the depth difference of each methylation block using the Methylation Block Score (MBS), the depth difference being used to distinguish between very small Tumor signal of , such as 0.1%, 0.2%, 0.5% and 1%.

In another preferred embodiment of the present disclosure, the sequencing read length to be analyzed is trimmed by any one or more of the following criteria: (i) query length of G bases longer than a fixed value m; (ii) non-G - Fraction of bases less than fixed number p; next base is high quality A/T/C (Phred score > 30)

In another preferred embodiment of the present disclosure, duplicate removal is applied with a tolerance of +/- 3 bp at both the start points of Rl and R2 to minimize artifacts associated with inappropriately assigned fragment end positions.

In another preferred embodiment of the present disclosure, UMI is applied in the correction.

In another preferred embodiment of the present disclosure, the average methylation level of each co-methylated block of unmethylated phage lambda DNA is detected to measure genome-wide "technical noise" (read length 1 in C/C+T in read length 2, G/G+A in read length 2).

In another preferred embodiment of the present disclosure, a machine learning classifier of methylation patterns is applied to assess tumor levels, preferably in early tumor screening.

As another aspect of the present disclosure, it provides the use of the above-mentioned method in assessing the general tumor level during early tumor screening. In a preferred embodiment of the present disclosure, the method described above is used in the assessment of tumor levels during early screening of tumors from homogenous tumors, heterogeneous tumors, hematologic cancers and/or solid tumor; preferably, said tumor is from one or more of the cancers of the following group: brain cancer, lung cancer, skin cancer, nasopharyngeal cancer, throat cancer, liver cancer, bone cancer, lymphoma, pancreatic cancer, skin cancer , bowel cancer, rectal cancer, thyroid cancer, bladder cancer, kidney cancer, oral cancer, stomach cancer, solid tumors, ovarian cancer, esophageal cancer, gallbladder cancer, biliary tract cancer, breast cancer, cervical cancer, uterine cancer, prostate cancer, head and neck cancer , sarcomas, thoracic malignancies (except lung), melanoma, and testicular cancer. In a preferred embodiment of the present disclosure, the above method is used in assessing the level of lung tumors during early tumor screening.

As another aspect of the present disclosure, it provides a kit for detecting ctDNA with methylation sequencing, wherein the kit can be used to capture at least 50, 100, 150, 200, 300, 500, 800, 1000, 1500 or 2000 co-methylation blocks as shown in Table 5-2; preferably, this can be used to capture at least all co-methylation blocks as shown in Table 5-1.

As another aspect of the present disclosure, it provides a device for performing the above method in assessing general tumor level during early tumor screening.

As another aspect of the present disclosure, it provides a non-volatile memory storing a program, which can be used to perform the above method in assessing general tumor level during early tumor screening.

In one aspect, the present application provides a method for detecting the level of base modification, comprising providing the nucleic acid molecule combination of the present application and/or the kit of the present application. For example, the base modification includes methylation modification.

In one aspect, the present application provides a storage medium, which records a program capable of running the method of the present application. For example, the non-transitory computer readable storage medium may include a floppy disk, a flexible disk, a hard disk, a solid state storage (SSS) (such as a solid state drive (SSD)), a solid state card (SSC), a solid state module (SSM)), an enterprise high-grade flash drives, tape, or any other non-transitory magnetic media, etc. Non-transitory computer readable storage media may also include punched cards, paper tape, cursor sheets (or any other physical media having a pattern of holes or other optically identifiable markings), compact disc read only memory (CD-ROM) , Rewritable Disc (CD-RW), Digital Versatile Disc (DVD), Blu-ray Disc (BD) and/or any other non-transitory optical media.

In one aspect, the present application provides a device, and the device includes the storage medium of the present application. For example, the device further includes a processor coupled to the storage medium, and the processor is configured to execute based on a program stored in the storage medium to implement the method of the present application.

Not intending to be limited by any theory, the following examples are only for explaining the methods and uses of the present application, and are not intended to limit the scope of the invention of the present application.

Example

Embodiment 1 The detection method of the present application

method part

Marker Discovery and Validation

Initially, differentially methylated sites were initially screened from the TCGA database generated by the Infinium HumanMethylation 450K array. A total of 4539 tumor samples and 521 normal tissue samples were analyzed. Data from a GEO dataset (GSE40279) for 656 normal WBC samples was used to remove hypermethylated CpG sites (>0.1 ) in the hematopoietic lineage. CpG sites located on the X or Y chromosomes were also excluded. DML selection was performed using "limma (V2.0)" software, and the cutoff value (cutoff) was set to B-H corrected FDR<0.05. In addition, CpG sites associated with common cancers in previous studies were also included. This resulted in a total of 80,672 CpG sites in the marker discovery phase.

The CpG sites were then partitioned into 8,312 blocks (described in "Co-methylation block partitioning") and compared using in-house sequencing of 48 primary lung tumor and 20 normal lung tissue samples The data is verified. Blocks with insufficient depth (<100) on most samples (>80%) were excluded from downstream analysis. Linear regression was used to select differentially methylated blocks, and cutoffs were set at log(fold change) >0.05, and B-H corrected FDR<0.05. A total of 2473 blocks were selected as classification features, while the genome coordinates are listed in Table 5.

Data Processing Framework

FASTQ files were generated from raw BCL data by using bcl2fastq (V2.19.1). Illumina-specific adapters and low-quality sequences (SLIDINGWINDOW:4:15TRAILING:20) were trimmed with trimmomatic (V0.36). For Accel-NGS Methyl-Seq (Swift Biosciences), additional trimming was performed as described in previous work. For ELSA-seq, an exemplary sequencing method of the present application, the present application tested parameters at different stringency levels to remove low-complexity tail sequences. The final pruning criteria are set as follows: (i) query the segment number of G bases longer than a fixed value m; (ii) the fraction of non-G-bases is less than a fixed number p; (iii) the next base is a high-quality A/T /C (Phred score>30). On the basis of analysis based on simulations and experiments, M=10 and P=0.1 were used in this study. The trimmed reads were then checked for quality with FastQC (V0.11.5), and a minimum read length of 50 bases was required.

Second, use the software bwa-meth (V0.2.0) to compare the read lengths with the bioinformatics-converted hg 19 human genome (for R1, C->T, for R2, G->A) with default parameters. Repeated reads were marked with Samblaster (V0.1.24), and bam files were sorted with Picard (V1.138). Reads with alignment scores <20, mismatches >5, improper pairing, or multiple mapping positions were excluded from secondary analysis. To minimize artifacts associated with misassignment of fragment end positions, duplicate removal was applied with a +/- 3 bp tolerance at both the start points ("blur windows") of R1 and R2. An important note is that while this strategy can correctly identify the original molecule over 90% of the time, accuracy decreases when the library is severely over- or under-sequenced.

For context-based error correction and methylation metric calculations, an internal module was built to collapse forward and reverse reads with respect to base call quality, cigar string, and mutation orientation relative to a reference genome (eg C->A) single consensus sequence. Base calls in low-fidelity regions (eg, near the end of the read) are underestimated, and missing bases are recovered based on supporting reads with the nearest Hamming distance.

Co-methylation block separation

的改进的相关矩阵将捕获小组的设计区域分隔为共甲基化区块。 This application is first based on the "block index" based on

The improved correlation matrix of the capture panel separates the design regions into co-methylation blocks.

而位点i与位点j之间的相关系数计算为

(i) Express the iAF correlation matrix of each CpG site in the r region as

And the correlation coefficient between site i and site j is calculated as

ρ _ij represents the Poisson correlation coefficient,

E(y _i ) denotes the mAF over all samples at position i,

pos _i represents the genomic position,

L represents the length of the original region,

Whereas _λ1 and _λ2 are parameters estimated by using prior information.

的“区块指数”定义为以下函数， (ii) For the new split in the original region r

The "block index" of is defined as the following function,

where B _i represents the number of sites in the i-th block,

α ₁ and α ₂ are the penalty coefficients for unbalanced split and over-split, respectively. In this study, both _α1 and _α2 are set to 1.0 based on the desired block length and uniformity.

This application defines two indicators "regional median length" and "regional length variation coefficient": "regional median length" is used to measure the size of the defined region, the smaller the value, the less information contained in the region, The larger the value, the more information the region contains; the "variation coefficient of region length" (standard deviation/mean) measures the size difference between different regions, and the larger the value, the more unbalanced the region division (more independent points, The more information is missing), the smaller the value, the more balanced the regional division. Among them, the value of α ₁ mainly affects the "variation coefficient of regional length", and the value of α ₂ mainly affects the "median length of the region". The influence relationship is shown in Figure 18.

达到最大来选择第(k+1)个新断点。 (iii) Repeat the separation process by iterative division with unique breakpoints. Assuming k breakpoints, then by making

The maximum is reached to select the (k+1)th new breakpoint.

(iv) Terminate the interactive algorithm when the (k+1)th breakpoint coincides with the kth breakpoint.

Methylation Block Score (MBS) Definition

This application defines the measure of MBS as follows

For a given block,

n is the total number of reads covering multiple CpG sites.

Li is the number of CpG sites covered on the i-th read.

lij represents the length of consecutive methylated CpG sites (>1),

And m represents the total count on the i-th read length. The depth difference was normalized using the number of reads in each block.

Simulation of segment recognition based on UMI correction or shift correction

The current pipeline and UMI auxiliary method simulation of this application are performed as follows:

(i) Template generation: a library of 2 kb DNA segments was generated and computationally "cut" into various lengths (mean = 170, sd = 30);

(ii) Tail addition: Based on empirical observations, a tail sequence (90% G) was added to the 5'-end of R2 with different lengths;

(iii) UMI incorporation: a random 6-base UMI tag (corresponding to modified linker 1) was added at the 5'-end of R2 to label the original template;

(iv) Technical errors: Three datasets were generated with low/medium/high substitution error rates (0.005, 0.01, 0.02) typical for Illumina sequencers; all errors (e.g., A->T; T->A) Equal likelihood setting; considering homopolymer properties, introducing indels (+/- 1 base) in the tail region; by 10 rounds of PCR reactions with a repeat probability of 0.9 (1 means complete doubling)/round Accumulation of PCR errors was simulated; random positional fluctuations (<10 bases) at the 5'-end of R1 were used to simulate incomplete extensions.

(v) UMI and non-UMI schemes:

a) UMI: UMI extraction (5′-end of R2) and shift correction (5′-end of R1, fw=3) were applied separately for each end; to avoid overcounting of UMIs due to amplification or sequencing errors , allowing a minimum edit distance of 2.

b) Non-UMI: shift correction is performed on both ends (5'-end of R1 and 5'-end of R2, fw=3);

(vi) Stencil Count: Set the original (post-BC) fragment depth at 250-500X and the original fragment depth at 500-1000X.

The whole process was repeated 9 times and the average performance of the simulations is summarized in Table 2.

UMI-facilitated template counting experiments

A DNA mixture (10 ng) of lambda DNA and stuffer DNA with -5,000 haploid copies (before BC) was prepared. In order to incorporate UMIs into ELSA-seq, an exemplary sequencing method of the present application, the 3' adapter was modified with a 6-base random UMI inserted next to the overhang sequence (connector 1-UMI), which was inserted next to the overhang sequence, thus R2 was sequenced during its first six cycles (Table 1). To help locate UMIs, two nucleotides (eg, 5'-NNDDNN-3', D: A/T/G) were predefined to act as anchors. To minimize the risk of false UMI calls due to polymerase or base calling errors, a minimum edit distance of 2 was allowed for error correction. A region of ~2 kb in the lambda genome (4,500-6,539) was targeted and analyzed, resulting in ~50,000 raw reads. Unique fragments measured by our current pipeline (non-UMI) were compared to a UMI-facilitated counting strategy by random read downsampling. The results are summarized in Table 2.

Supplementary Methods Section

Determination of adapter ligation efficiency of an exemplary sequencing method ELSA-seq and TruSeq of the application by ddPCR

To evaluate ligation efficiency, set up the ddPCR reaction as follows:

1. ELSA-seq ligation reaction (20 μl)

10X CutSmart buffer 2 _CoCl2 (2.5mM) 2 NAD (0.5mM) 1 Template DNA 8.6 H ₂ O 2.1 dCTP: dNTP (2.5mM) 0.8 adapter (joint) (10uM) 2 TdT 0.5 E.coli Ligase (ligase) 1

2.ddPCR reaction (20μl):

Substrate:

ELSA-SEQ: Synthetic ssDNA (KRAS-10N)

TruseQ: dsDNA amplified using primers KRASF and KRASR (KRAS-177)

ELSA connection-1:

Total copies were detected by primers LEF and KRASR, probe KRAS-G13D.

Ligated copies were detected by primers LEF and LER, probe KRAS-G13D.

ELSA connection-2:

Total copies detected by primers LEF and KRASR, probe KRAS-G13D

Ligated copies were detected by primers LEF and LER-ATNR1, probe KRAS-G13D.

TruSeq connection:

Total copies were detected by primers LEF and KRASR, probe KRAS-G13D.

The concatenated copies were detected by primers LEF and LER-ATNR1, probe KRAS-G13D.

A design using 'N' in the KRAS-10N template was used to avoid sequence-dependent bias. All oligonucleotide sequences are listed in Table 1 - Summary of Oligonucleotide Sequences.

Construction and Analysis of Whole Genome Bisulfite Sequencing (WGBS) Libraries

WGBS library construction with ELSA-seq, NEBNext Ultra II (New England Biolabs) and Accel-NGS Methyl-Seq (Swift Biosciences) was performed as described in Methods or according to manufacturer's instructions. Genomic DNA was sheared to ~200 bp (peak) by sonication. Library quality was assessed using LabChip GXII touch 24 (Perkin Elmer). Paired-end sequencing (2×150bp) was then performed on the Illumina NovaSeq 6000 system.

Simulation of segment recognition based on UMI correction or shift correction

The simulation of the current pipeline and UMI auxiliary method of this application is carried out as follows:

(vii) Template generation: A library of 2kb DNA segments was generated and computationally "cut" into various lengths (mean = 170, sd = 30);

(viii) Tail addition: Based on empirical observations, tail sequences (90%G) were added to the 5'-end of R2 with different lengths;

(ix) UMI incorporation: a random 6-base UMI tag (corresponding to modified linker 1) was added at the 5'-end of R2 to label the original template;

(x) Technical errors: Three datasets were generated with low/medium/high substitution error rates (0.005, 0.01, 0.02) typical for Illumina sequencers; all errors (eg, A->T; T->A) Equal likelihood setting; considering homopolymer properties, introducing indels (+/- 1 base) in the tail region; by 10 rounds of PCR reactions with a repeat probability of 0.9 (1 means complete doubling)/round Accumulation of PCR errors was simulated; random positional fluctuations (<10 bases) at the 5'-end of R1 were used to simulate incomplete extensions.

(xi) UMI and non-UMI schemes:

a) UMI: UMI extraction (5′-end of R2) and shift correction (5′-end of R1, fw=3) were applied separately for each end;

b) Non-UMI: shift correction is performed on both ends (5'-end of R1 and 5'-end of R2, fw=3);

(xii) Template counting: To reflect the ELSA-seq library in a clinical setting, the original (post-BC) fragment depth was set at 250-500X, while the original fragment depth was set at 500-1000X.

The whole process was repeated 9 times and the average performance of the simulations is summarized in Table 2.

UMI-facilitated template counting experiments

A DNA mixture (10 ng) of lambda DNA and stuffer DNA with -5,000 haploid copies (before BC) was prepared. In order to incorporate UMIs into ELSA-seq, an exemplary sequencing method of the present application, the 3' adapter was modified with a 6-base random UMI inserted next to the overhang sequence (connector 1-UMI), which was inserted next to the overhang sequence, thus R2 was sequenced during its first six cycles (Table 1). To help locate UMIs, two nucleotides (eg, 5'-NNDDNN-3', D: A/T/G) were predefined to act as anchors. To minimize the risk of false UMI calls due to polymerase or base calling errors, a minimum edit distance of 2 was allowed for error correction. A region of ~2 kb in the lambda genome (4,500-6,539) was targeted and analyzed, resulting in ~50,000 raw reads. Unique fragments measured by our current pipeline (non-UMI) were compared to a UMI-facilitated counting strategy by random read downsampling. The results are summarized in Table 2.

Methyltransferase-mediated methylation in vitro

Set up the reaction as follows:

The in vitro methylation process was performed at 37°C for 1 hour and terminated by heating at 65°C for 20 minutes.

Sample size calculation with desired margin of error (<0.05)

The sample size (n) is calculated as follows:

并且

为标准累积正态分布的反函数 in

and

is the inverse function of the standard cumulative normal distribution

Analysis of group composition

To assess potential confounding effects (e.g. age, sex) in cohort designs, chi-square tests were used with the null hypothesis that categorical variables were comparable in sample or control classes (P>0.05). Additional data on patients and tumors are provided in Figure 15 and Table 6.

Multivariate analysis for predictors

Univariate and multivariate analyzes were performed by logistic regression to identify clinically important factors for MERMAID trial outcomes. All independent variables (clinical factors) were tested for logit linear correlation with the dependent variable (prediction score). The results are provided in Table 6.

Calculation of Potential Clinical Benefit in a Hypothetical Screening Population

MERMAID was evaluated for diagnostic yield for LC in a hypothetical screening population of 10,000 subjects. Based on the results of this study, the sensitivity was set at 63.0% (194/308), and the specificity was set at 96.2% (251/261). According to the Surveillance, Epidemiology, and End Results (SEER) program (SEER.cancer.gov/data/access, 2020), the prevalence of LC among average-risk older adults was assumed to be 0.53%. Among 10000 subjects, the number of individuals with true positive (TP), false positive (FP), true negative (TN) and false negative (FN) results was predicted. Positive predictive value (PPV) and negative predictive value (NPV) were calculated as follows:

Predicted FP per TP indicates the number of FP subjects observed when detecting TP subjects. One caveat is that although all patients in this application had early stage (I-III) LC and 65% of them (199/308) were diagnosed with stage I (Stages Ia and Ib) at study entry, this application still The stage distribution in the true screening setting is not known, so the PPV may be less than reported here.

For tests where a positive result leads to an action, performance can be evaluated as follows:

(1)

其中R指肺结节为恶性的风险，对此本申请可以不偏倚地选择监测或积极调查。在此基于来自国家肺部筛查试验(NLST)研究小组的报告，将R设置为1.1％。 (2)

Where R refers to the risk of pulmonary nodules being malignant, for which the applicant can choose to monitor or actively investigate without bias. Here R is set at 1.1% based on reports from the National Lung Screening Trials (NLST) study group.

In the assumed average risk population, the performance of MERMAID met the criteria based on equations (1) and (2) (MERMAID: 16.6 > 2.1).

Example 2 The results of the methylation detection method of the present application

Design of an ultrasensitive BS-seq assay in this application

The methylation detection method of the present application may be based on methylation sequencing data known in the art. For example, the methylation detection method of the present application can be based on an ultrasensitive BS-seq (bisulfite sequencing). For example, an available BS-seq can be ELSA-Seq described in WO2019192489A1. For example, the design principle of the sequencing method for the methylation sequencing data of this application involves two aspects of molecular and calculation ( FIG. 1 ). In order to improve the detection ability, the present application first focuses on increasing the templates that can be sequenced effectively: (i) DNA molecules need specific templates at both the 5' end and the 3' end to be "read" by the high-throughput sequencer. connector. Fragmentation of tagged molecules will lead to failure of "seeding" on the flow cell surface. In order to minimize loss of adapters by BC, the applicant arranges this denaturation step first, followed by a single-stranded DNA (ssDNA) compatible step to maximize retrieval of the original template. (ii) Adapter ligation is another common limiting factor, so this applicant devised a new strategy called "tail and tag" to improve efficiency. Briefly, bisulfite-treated DNA is denatured, dephosphorylated and extended with a cytosine-rich nucleotide tail by TdT (terminal deoxynucleotidyl transferase). Then, the splint adapter was annealed to the tail in the presence of E. coli ligase to facilitate a highly efficient ligation step (Tail-Tag. 1). (iii) Second, copy strands are generated from the common anchor site by a uracil-tolerant polymerase, thereby providing high molecular redundancy for the single-tag intermediate to enable the next round of adapter attachment (Tail-Tag.2). Template loss is minimized (Figure 1).

In order to estimate the template recovery rate of the sequencing method for the methylation sequencing data of the present application, the present application first compared its ligation efficiency with the traditional method (TruSeq) by ddPCR (Fig. 6A-B). The ratios of ligated/total DNA copies are 82% (Tail-Tag.1) and 86% (Tail-Tag.2) for the sequencing methods of the methylation sequencing data of this application, and 64% for TruSeq (Fig. 6C-E). Considering that two rounds of ligation are required for both methods, the recovery is almost doubled (0.64*0.64 vs. 0.82*0.86) just by applying step (ii). Previous work has demonstrated that the concepts employed in steps (i) and (iii) can increase template retention, so the combination of all these steps is expected to significantly increase library complexity. The comparison of the sequencing method of the methylation sequencing data of the present application with other library preparation methods is summarized in FIG. 7 .

In order to further evaluate its performance, the applicant compared the sequencing method of the methylation sequencing data of the application with two commercial kits: Accel Methyl-Seq (SWT) and NEBNext Ultra (NEB), and found that using the application's Whole-genome bisulfite sequencing (WGBS) libraries constructed with sequencing methods for methylation-sequencing data showed a 10-fold increase in yield and exhibited the highest number of unique molecules regardless of input amount or sequencing depth, (Fig. 2A-C). Furthermore, for an input as low as 500 pg, the present application's sequencing method for methyl-sequencing data showed the greatest methylome coverage, little amplification bias, and highly reproducible methylation levels (Fig. 2D , Figure 8A-E). It is worth noting that although reads are subjected to stringent adapter and tail removal, residual synthetic sequences and incomplete extension of copy strands may still alter where DNA fragments start or end, leading to incorrect estimates of library complexity . Therefore, the present application employs "conditional pruning" and "shift correction" in data processing to overcome these challenges (Methods section of Example 1). Through simulation- and experiment-based analyses, we found that >90% of fragments could be correctly identified by our current informatics strategy, approaching the level of UMI-aided measurements (Supplementary Methods section of Example 1, Table 2).

Traditional BS-seq introduces a significant amount of technical artifacts, which hinders its use for rare allele detection. To address this issue, the present application utilizes deep sequencing to suppress errors within PCR repeat families. The present application first applied the sequencing method of the methylation sequencing data of the present application to the unmethylated phage lambda DNA to measure the genome-wide "technical noise" (C/C+T in read length 1, C/C+T in read length 2 G/G+A). Compared with previous studies, the error rate (maximum 0.0025, average 0.0017) of the sequencing method for the methylation sequencing data of the present application was reduced by almost 10 times, regardless of the number of sequencing cycles (Fig. 8F). The non-reference allele frequencies (0.00003-0.00135) for each substitution type appeared to be comparable to those described for the Illumina platform (Fig. 2E-F, Fig. 8G). Recently, a method called EM-seq featuring mild enzyme-based transformations was reported, allowing libraries to be prepared from small amounts of input. Analysis of human methylome data reveals that both EM-seq and the sequencing method of this application's methylation-sequencing data have superior performance in terms of library complexity and coverage uniformity, with EM-seq showing fewer AT Bias and the sequencing method of the methylation-sequencing data of the present application exhibit lower patterning noise (two unconverted C in tandem, Table 2). The full capabilities of this new technology have yet to be tested.

Panel Design and Target Capture Performance

Because deep sequencing of the full human methylome would be prohibitively costly, we focused on epigenetic changes associated with common cancers by analyzing data from previous studies (Fig. 3A). Therefore, 80,672 CpG sites were selected, spanning a genomic region of approximately 1.05 MB. This panel demonstrated a significant enrichment in CpG islands, consistent with the reported role of DNA methylation in transcriptional regulation during tumorigenesis (Fig. 9A-B).

The performance of targeted sequencing was evaluated using human lymphocyte DNA (NA12878) and plasma samples. Fairly uniform capture of amplified DNA fragments was achieved with as little as 2 ng of cfDNA, where 60-80% of reads aligned on the bait region only (target ratio) and >90% of the bait region consisted of >200% reads Coverage (uniformity) (Figure 3B, Figure 9C, Table 2). Read coverage plotted against GC content revealed a typically unimodal distribution (Fig. 9D) and estimated base calling accuracy was above 99.9% (Phred score >30) for most (>0.94) bases (Fig. 9E, Table 2). The sequenced cfDNA fragments had a mononucleosome peak length of ∼160 bp, which was in close agreement with the results of the traditional method (TruSeq) (Fig. 3C). Furthermore, negligible amplification or capture bias was observed for fragments associated with mono- and di-nucleosomes (Fig. 9F). To assess common capture bias, we calculated the percentage of methylated cytosines at each individual CpG site (iAF, individual methylated allele frequency) and found high correlations on both the positive and negative strands (ρ=0.90, Figure 3D). This is confirmed by the nearly balanced read depth of the positive vs. negative strands (FIG. 9G). The typical technical error rate after target enrichment was 0.0012, similar to that observed using the no-capture method (Figure 3E).

The detection method of the present application, MERMAID, recognizes signals through single-molecule derivatization patterns

Traditional DNA methylation analysis is largely based on iAF, which is sensitive to sampling variance and technical noise. In response, MERMAID, the methylation detection method of the present application, devised a metric, the "block index" (BI), to separate CpG sites showing similar methylation status into different blocks (Fig. 4A). A total of 8312 blocks were defined, with a median block size of ~143 bp and an average of ~13 CpG sites/block (Fig. 10A-D). The present application defines the average methylation level in each block as mAF (mean methylated allele frequency) and compares its performance with iAF for detection of cancer-associated changes. By examining SHOX2, a gene frequently methylated in LC, we found that mAF showed significantly higher AUC values than iAF, showing that "blocks" are more distinguishable units than "sites" (Fig. 4B).

The advantage of MERMAID is the much improved separation of biological signal and technical noise. Analysis of each DNA fragment revealed distinct patterns for cancer cells and normal cells, while chemical or sequencing errors were often sporadic (Fig. 10E). To highlight the contrast, this application develops a new unit of measurement, the "Methylation Block Score" (MBS), which is defined as the ratio of the weighted occurrence of consecutive methylation patterns to the total CpG sites per read. ratio (Fig. 4C). In order to evaluate the performance of MBS, DNA treated with in vitro methyltransferase was mixed into λDNA at different ratios. As shown in Figure 4D. By MBS, even samples with small spikes (0.001) were clearly distinguishable from negative controls, whereas if mAF was used, a clear overlap of signals was observed, supporting the success of pattern recognition in increasing the signal-to-noise ratio.

As shown in Figure 17, traditional DNA methylation analysis is mainly based on the methylation level iAF of a single site, and iAF is very sensitive to sequencing depth and technical noise. Therefore, this application designs an index "block index" to separate CpG sites into different blocks according to the methylation status and genomic position information of CpG sites. The present application defines the average methylation level of each block as mAF (mean methylated allele frequency) and compares it with the performance of iAF to detect cancer-associated changes. By detecting SHOX2, a gene that frequently undergoes methylation mutations in lung cancer patients, the present application found that the area under the receiver operating characteristic curve (AUROC) of mAF was significantly higher than that of iAF, which indicated that "block" is more important than "site "The more discriminative units, and the lower the sequencing depth, the better the effect of mAF compared to iAF.

As shown in Figure 19, the present application also compares the advantages of the MBS statistic over the traditional mAF statistic. Using the blending ratio data between a set of negative standard and positive standard, the density curves of MBS statistic and mAF statistic in samples with different blending ratios were drawn. It can be seen from the figure that the ratio of MBS statistic to mAF statistic is in There is a larger difference between the "Negative Standard" and "0.1% Spiked Positive Standard", indicating that the MBS statistic is more capable of discriminating methylation differential signals.

Evaluation of MERMAID for Signal of Tumor Origin Detection

The analytical performance of MERMAID was examined at both assay and bioinformatics levels. Highly correlated MBS values (p = 0.94-0.97) were observed in healthy individuals (including repeats), suggesting excellent reproducibility (Fig. 5A). Similar results were also observed with different input amounts (2-30 ng), sequencing depth (1000-5000X) or DNA source (from cells or plasma) (Fig. 11). However, the differences between healthy and cancer groups and among different cancer patients were very significant (ρ=0.60-0.83), implying that methylation profiles were greatly affected by disease status (Fig. 5A).

The quantitative accuracy of MERMAID was evaluated by performing tumor cell spike experiments. On a dilution series of normal WBC DNA versus colorectal cancer (CRC) DNA, a near-perfect correlation was found between observed and expected tumor proportions at a dilution of 0.0005 (Fig. 5B, r ² =0.99) , implying fine quantitative accuracy. Highly consistent results were also observed when the present application compared MERMAID with quantitative methylation-specific PCR (qMSP) or different target methods (Figure 12).

The false discovery rate (FDR) of the assay was evaluated empirically by repeatedly sequencing normal leukocyte WBCs. As shown in Figure 5C and Table 2, FDR decreased steadily with increasing DNA input or sequencing depth, most likely because markers with low methylation counts generated the majority of false calls, and these markers were more Many benefit from the reduction of Poisson noise (Fig. 13A-B).

The applicant evaluated the limit of detection (LoD) of MERMAID at three different levels: (i) Numerical simulations: By simulating methylation counts from a binomial distribution, the applicant found that both increasing the number of markers and sequencing depth, The success rate (sensitivity) could be improved (FIG. 13C). Notably, the size of the markers swelled dramatically as the ctDNA fraction decreased from 1/10,000 to 1/100,000, revealing that tumor burden is a fundamental limiting factor for detection sensitivity. (ii) Bioinformatics simulations: The theory was further tested by computationally mixing sequencing reads from cancer cells with those from healthy cfDNA at different ratios. As shown in Figure 5D, the bioinformatic sensitivity of MERMAID reached 1/100,000 in simulated data from both LC and CRC (two-tailed t-test, P<0.001). This observation appeared to be batch-independent, as pooled control data from different rounds of sequencing did not yield significant detections (two-tailed t-test, P>0.05, Figure 13D). (iii) Experimental evaluation: LC and CRC DNA were spiked into normal WBC DNA at serial dilutions. A cancer signal (detection rate) quantified with MBS was detected at dilutions as low as 1/10000 for both cell lines (two-tailed t-test, P<0.01, FIG. 5E ).

Finally, the present application compares the LoD of MERMAID with ddPCR and ultra-deep mutation sequencing with unique molecular specifiers (HS-UMI), two methods that are exceptional in detecting variants at very low frequencies. For each diluted sample, approximately 9000 copies of the human haploid genome were used to mimic the average cfDNA amount in 10 ml of blood. For the pre-defined hotspot mutations, EGFR p.G719S was identified by ddPCR and HS-UMI at 0.03-0.11% AF in all 1/1,000 dilutions (Fig. 5F, Fig. 13E, Table 4). Detection of the oncogenic EML4-ALK fusion at 1/1,000 dilution was verified with ddPCR at an AF of 0.03%, but failed with HS-UMI (Fig. 5F, Fig. 13F, Table 4). Depending on read depth and background noise, including mutations of uncertain functional significance did not increase the maximum dilution ratio above 1/1000 (Table 4). In conclusion, MERMAID showed at least 10-fold greater power than mutation assays under the same conditions.

Validity of the MERMAID methylation sequencing method in this application

Methylation-sequencing has attracted enormous interest because it has great potential to improve current ctDNA assays. Herein the present application presents MERMAID as a novel epigenetic analysis method characterized by well-conserved molecular diversity, robust noise suppression and robust high-dimensional modeling. In addition to these properties, MERMAID can be particularly useful for blood-based applications: (i) a portion of cfDNA may be in single-stranded form, so this ssDNA-compatible approach can maximize the use of limited starting material, increasing the availability of rare ctDNA opportunity for detection. (ii) Capture panels were designed with an excess of long RNA probes (>100 nucleotides long) complementary to various methylation patterns. This strategy is more tolerant to sequence-related biases and polymorphisms than amplicon-based target approaches (~20 nucleotides long). (iii) MERMAID does not require prior knowledge of the assay (eg, biopsied tissue), thus providing a solution for patients without surgically resected samples. Although this method has only been validated on LC, it could be customized to other types of cancer (e.g., CRC) or body fluids (e.g., urine). It can be extended to answer fundamental questions, such as tumor heterogeneity, or applied to other clinical scenarios, such as evaluating treatment effects.

The disadvantage of introducing synthetic sequences to improve adapter labeling efficiency is the failure to protect the native ends of the DNA template, which may then result in incomplete repeat removal. The computational strategies employed by MERMAID can, at least in part, overcome this problem of incomplete duplicate removal. At the same time, MERMAID can adopt the bias processing method commonly used in the field to deal with the accompanying risk of C->T/G->A artifacts caused by the conversion of cytosine to uracil after oxidative stress. The MERMAIDs of the present application can employ the addition of tissue-specific markers to target panels for multiple cancer classification.

supplementary form

Oligonucleotide sequences used in the method of the present application in table 1-1

Table 1-2 Oligonucleotide sequences used in the method of this application + UMI correction

Table 1-3 is used for the oligonucleotide sequence of ligation efficiency measurement (ddPCR)

Table 2-1 Using 500pg Escherichia coli DNA of ELSA, SWT and NEB full methylome sequencing indicators (deep titration)

Table 2-2 The method of this application and the method of this application + UMI correction (data simulation)

Table 2-3 The application method and the application method + UMI correction (test)

Table 2-4 EM-seq and the method of this application

Note 1: Sequencing data were downloaded from the NCBI Sequence Read Archive (SRC) with accession PRJNA591788 and PRJNA534206. Note 2: In order to objectively compare these methods, all data were downsampled to ~100M read length

Table 2-5 The quality control indicators of this application method using 2 to 30ng human cfDNA input

Table 2-6 uses 2 to 30ng human WBC input method quality control index of this application

Table 4-1 ddPCR detection of hotspot mutations using titrated SW48 and H2228 cancer cells (biological replicates)

Table 4-2 HS-UMI detection of hotspot mutations using titrated SW48 and H2228 cancer cells (biological replicates)

Table 4-3 HS-UMI detection of non-hotspot mutations using titrated SW48 and H2228 cancer cells (biological replicates)

Table 4-4 QC indicators of HS-UMI using cancer cell lines (NCI-H2228, SW48) spike-in experiments

Table 5-1 "Descriptor" Organization Classification (n=157)

Table 5-2 "Specifier" plasma classification (n=2,473)

Table 6 Multivariate analysis of ctDNA detection by the method of this application

Table 7 Group 2 "Quadruple Comparison" Summary

Table 7-1 Summary of ctDNA detection by HS-UMI and ddPCR in "Group 2 - Quadruple Comparison"

Table 7-2 Summary of ctDNA mutations detected by ddPCR in "Group 2 - Quadruple Comparison"

Table 7-3 ddPCR assays used in "Group 2 - Quadruple Comparisons"

The foregoing detailed description has been offered by way of explanation and example, not to limit the scope of the appended claims. Variations on the presently recited embodiments of this application will be apparent to those of ordinary skill in the art and remain within the scope of the appended claims and their equivalents.

Claims (46)

A method for detecting the methylation modification of a target nucleic acid, the method comprising the following steps: step (a-1) is based on the correlation coefficient of the CpG site in the target nucleic acid, the methylation level of the CpG site, and The position information of the CpG site determines the co-methylation block; and/or step (a-2) is based on the correlation coefficient of the CpG site in the target nucleic acid, the amount of information of the candidate co-methylation block and the division balance degree of the candidate co-methylation block, determining the co-methylation block, and step (b) determining the methylation degree of the co-methylation block in the sample to be tested based on the methylation degree of the co-methylation block The presence and/or amount of the target nucleic acid. The method according to claim 1, which detects the presence and/or content of the target nucleic acid with methylation modification in the sample to be tested. The method of any one of claims 1-2, the correlation coefficient of the CpG sites comprising a Pearson correlation coefficient between two or more of the CpG sites in the target nucleic acid. The method of any one of claims 1-3, said methylation level of said CpG site comprising an average methylation level between two or more of said CpG sites in said target nucleic acid The difference in methylation allele frequency (mAF). The method of any one of claims 1-4, said methylation level of said CpG site comprising the mAF between two or more said CpG sites in said target nucleic acid The ratio of the difference to the sum of mAFs. The method of any one of claims 1-5, the location information of the CpG sites comprising differences in genomic locations between two or more of the CpG sites in the target nucleic acid. The method according to any one of claims 1-6, wherein the position information of the CpG sites comprises the genome position distance between two or more of the CpG sites in the target nucleic acid and the The ratio of the lengths of the target nucleic acids. The method according to any one of claims 1-7, said step (a-1) comprising: determining the corrected correlation coefficient between every two said CpG sites of said target nucleic acid, site i The corrected correlation coefficient d _ij with site j is calculated by the following formula:

where _ρij represents the Pearson correlation coefficient, E(y _i ) represents the average methylated allele frequency (mAF) of all samples at site i, and E(y _j ) represents the average methylated allele frequency (mAF) of all samples at position j. λ allele frequency (mAF), pos _i represents the genomic position of locus i, pos _j represents the genomic position of locus j, L represents the length of the target nucleic acid region, and λ ₁ and λ ₂ are independently selected from 0 or greater number. The method according to claim 8, the value range of said _λ1 is 0 to 1. The method according to any one of claims 8-9, the value range of said λ ₂ is 0 to 1. The method according to any one of claims 1-10, the correlation coefficient based on the CpG site in the target nucleic acid in the step (a-2) comprises the correction of the CpG site in the target nucleic acid Post-correlation coefficient, the post-correction correlation coefficient of the CpG site in the target nucleic acid comprises the correction based on the methylation level of the CpG site and/or the position information of the CpG site in the target nucleic acid Pearson's correlation coefficient between two or more of said CpG sites. The method of claim 11 , the corrected correlation coefficient of the CpG site in the target nucleic acid comprises the corrected correlation coefficient in the method of any one of claims 8-10. The method according to any one of claims 1-12, wherein the information amount of the candidate co-methylation block comprises the quantity information of the CpG sites of the candidate co-methylation block. The method according to any one of claims 1-13, wherein the degree of partition balance of the candidate co-methylation blocks comprises the difference in the number of the CpG sites of the different candidate co-methylation blocks . The method according to any one of claims 1-14, wherein the degree of partition balance of the candidate co-methylation blocks comprises variations in the number of CpG sites of the different candidate co-methylation blocks coefficient. The method according to any one of claims 1-15, said step (a-2) comprising: maximizing the block index of said candidate co-methylation block, determining said co-methylation area block, a candidate co-methylated block in the target nucleic acid

The block index of

Calculated by the following formula:

in,

B _i represents the number of the CpG sites of the ith candidate co-methylation block, and α ₁ and α ₂ are independently selected from 0 or greater numbers. The method according to claim 16, determining the block breakpoints of the co-methylation blocks through an iterative method of unique breakpoints. The method according to any one of claims 16-17, wherein the value of _α1 ranges from 0 to 10. The method according to any one of claims 16-18, the value range of _α2 is 0 to 10. The method according to any one of claims 1-19, said step (b) comprising: based on the length of the continuous CpG of each sequencing read (read) of said co-methylation block, said read The number of CpGs on the above and the total number of reads of the co-methylated block determine the presence and/or content of the target nucleic acid. The method according to any one of claims 1-20, said step (b) comprising: determining the methylation block score of said co-methylation block, said methylation block score MBS passing Calculated with the following formula:

Wherein, the n is the total number of reads covering all CpG sites of the co-methylation block, L _i is the number of CpG sites contained on the i-th read, and l _ij is the continuous methylation CpG on the i-th read The length of the site, m is the sequencing depth on the i-th read. The method according to any one of claims 1-21, applying UMI correction to the summary of the sequencing data of the sample to be tested. The method according to any one of claims 1-22, the method further comprises: extracting the feature value of the MBS of the co-methylation block of the tumor sample and the healthy sample through a machine learning model, based on the The MBS of said co-methylated blocks in a sample determines the presence and/or amount of said target nucleic acid. An analysis device for detecting the methylation modification of a target nucleic acid, the device comprising: a block division module (a-1), based on the correlation coefficient of the CpG site in the target nucleic acid, the methyl group of the CpG site The methylation level and the position information of the CpG site determine the co-methylation block; and/or the block division module (a-2), based on the corrected correlation coefficient of the CpG site in the target nucleic acid, the candidate The amount of information of the co-methylation block and the division balance degree of the candidate co-methylation block, determining the co-methylation block, and the determination module (b) based on the co-methylation block The degree of methylation determines the presence and/or amount of the target nucleic acid in the sample to be tested. A method for detecting ctDNA using methylation sequencing, comprising: - Selection of differentially methylated CpG sites; - separating said CpG sites into a plurality of co-methylation blocks based on the similarity of the methylation status of said CpG sites; - Sequence the samples to obtain methylation sequencing reads; - Measure the average methylation level of each co-methylated block in the sample for further DNA methylation analysis. The method according to claim 25, wherein the differentially methylated CpG sites are selected from the TCGA database generated by the Infinium Human Methylation 450K array. The method according to any one of claims 25-26, wherein said average methylation level is the average methylated allele frequency. The method according to any one of claims 25-27, wherein the number of co-methylation blocks is between 1/30 and 1/5 of the number of CpG sites. The method according to any one of claims 25-28, wherein said co-methylated blocks are further restricted by comparing a specific tumor sample with a normal tissue sample, eg a primary lung tumor with a normal lung tissue sample. The method according to claim 29, wherein co-methylation blocks of insufficient depth (<100) on most samples (>80%) are excluded from downstream analysis. A method according to any one of claims 25-30, wherein co-methylation blocks are separated based on a modified correlation matrix called "block index". The method of claim 31, wherein (1) Express the iAF correlation matrix of each CpG site in region r as

And the correlation coefficient between site i and site j is calculated as

_ρij denotes the Poisson correlation coefficient, E(y _i ) denotes the mAF of all samples at position i, pos _i denotes the genomic position, L denotes the length of the original region, and _λ1 and _λ2 are parameters estimated by using prior information ; (2) For the new split in the original area r

The block index of is defined as the following function,

where B _i represents the number of sites in the i-th block, α ₁ and α ₂ are the penalty coefficients for unbalanced split and over-split respectively; in this study, based on the expected block length and uniformity, Set both α ₁ and α ₂ to 1.0; (3) Repeat the partitioning process by iterative division with unique breakpoints; assuming k breakpoints, then by using

Reach the maximum to select the (k+1)th new breakpoint; Terminate the interactive algorithm when the (k+1)th breakpoint coincides with the kth breakpoint. A method according to any one of claims 25-32, wherein the method further comprises normalizing the depth difference of each methylation block using a methylation block score (MBS), the depth difference being used to differentiate Very small tumor signals, such as 0.1%, 0.2%, 0.5%, and 1%. The method of claim 33, wherein said MBS is:

For a given block, n is the total number of reads covering multiple CpG sites; L _i is the number of CpG sites covered on the i-th read; l _ij represents consecutive methylated CpG sites (>1) and m is the total count on the ith read length. The method according to any one of claims 25-34, wherein the sequencing reads to be analyzed are trimmed by any one or more of the following criteria: (i) Query the segment number of G bases longer than the fixed value m; (ii) the fraction of non-G-bases is less than a fixed number p; (iii) The next base is high quality A/T/C (Phred score >30). A method according to any one of claims 25-35, wherein duplicate removal is applied with a tolerance of +/- 3 bp at both the start points of R1 and R2, so that artifacts associated with inappropriately assigned fragment end positions are reduced to minimum. A method according to any of claims 25-36, wherein the UMI is applied in the correction. The method of claim 37, wherein the UMI assisted correction comprises: (1) Template generation: Generate a library of 2kb DNA segments and "cut" them into various lengths by calculation (mean value = 170, sd = 30); (2) Tail addition: based on empirical observations, a tail sequence (90%G) was added to the 5'-end of R2 with different lengths; (3) UMI incorporation: add a random 6-base UMI tag (corresponding to the modified linker 1) at the 5'-end of R2 to label the original template; (4) Technical errors: Three sets of data sets were generated with typical low/medium/high substitution error rates (0.005, 0.01, 0.02) for Illumina sequencers; all errors (eg, A->T; T->A) were Equal likelihood setting; considering homopolymer properties, introducing indels (+/- 1 base) in the tail region; by 10 rounds of PCR reactions with a repeat probability of 0.9 (1 means complete doubling)/round Mimics accumulation of PCR errors; simulates incomplete elongation with random positional fluctuations (<10 bases) at the 5'-end of R1; (5) UMI and non-UMI schemes: a) UMI: UMI extraction (5′-end of R2) and shift correction (5′-end of R1, fw=3) were applied separately for each end; b) Non-UMI: shift correction is performed on both ends (5'-end of R1 and 5'-end of R2, fw=3); (6) Stencil count: set the original (post-BC) fragment depth at 250-500X, and set the original fragment depth at 500-1000X. The method according to any one of claims 25-38, wherein the average methylation level of each co-methylated block of unmethylated phage lambda DNA is detected to measure genome-wide "technical noise" (read length C/C+T in 1, G/G+A in read 2). A method according to any one of claims 25-39, wherein a machine learning classifier of methylation patterns is applied to assess tumor levels, preferably in early tumor screening. A kit for detecting ctDNA with methylation sequencing, wherein the kit can be used to capture at least 50, 100, 150, 200, 300, 500, 800, 1000, 1500 or 2000 as in Table 5-2 Co-methylation blocks shown; preferably, this can be used to capture at least all of the co-methylation blocks as shown in Table 5-1. 1. Use of the method according to any one of claims 1-23 and 25-40 in assessing tumor levels. The use according to claim 42, said use comprising assessing said tumor level in early tumor screening. The use according to any one of claims 42-43, said tumors are from homogeneous tumors (homogenous tumors), heterogeneous tumors, blood cancers and/or solid tumors; preferably, said tumors are from the following groups One or more of the following cancers: brain cancer, lung cancer, skin cancer, nasopharyngeal cancer, throat cancer, liver cancer, bone cancer, lymphoma, pancreatic cancer, skin cancer, bowel cancer, rectal cancer, thyroid cancer, bladder cancer , kidney cancer, oral cancer, gastric cancer, solid tumors, ovarian cancer, esophageal cancer, gallbladder cancer, biliary tract cancer, breast cancer, cervical cancer, uterine cancer, prostate cancer, head and neck cancer, sarcoma, thoracic malignancy (except lung), melanoma, and testicular cancer. A storage medium recording a program capable of executing the method according to any one of claims 1-23 and 25-40. An apparatus comprising the storage medium of claim 45, and optionally comprising a processor coupled to the storage medium, the processor configured to The program in is executed to realize the method described in any one of claims 1-23 and 25-40.

Images