CN105701365A - Cancer-related genes finding method by using miRNA expression data - Google Patents

Abstract

The invention discloses a method for discovering cancer-related genes by using miRNA expression data. Based on the pan-cancer project PanCancer under the Cancer Genome Atlas TCGA, statistical analysis and machine learning algorithms are used to analyze and process gene expression data to identify genes related to complex diseases. including: sorting out sample data; performing statistical analysis on miRNA data; sorting miRNA according to the average rate of change; selecting target genes; extracting corresponding disease samples and normal samples; genes are sorted. The invention can discover a plurality of risk genes related to complex diseases such as cancer, which is of great significance to biological target therapy, biological drug development, pathogenic mechanism interpretation and risk prediction of complex diseases.

Description

A method for discovering cancer-related genes using miRNA expression data

technical field

The invention belongs to the technical field of data processing, and in particular relates to a method for discovering cancer-related genes by using miRNA expression data.

Background technique

Bioinformatics is an emerging discipline combining life science and computer science, which studies the collection, processing, storage, dissemination, analysis and interpretation of biological information, and reveals complex biological information through the comprehensive use of biology, computer science and information technology. Biological mysteries hidden in data. Genes are the carriers of genetic information, and the exploration of genes will help to deepen the understanding of diseases. There are currently more than 20,000 genes known to humans, and the mRNA gene expression data obtained from the corresponding sequencing data has reached more than 20,000 dimensions, and the genes related to each disease are different. Some disease-related genes have been discovered, but most Most of the related genes need further study. It can be seen that direct analysis of mRNA gene expression data needs to deal with high-dimensional data, and the computational complexity is extremely high.

miRNA is a kind of endogenous small RNA with a length of about 20-24 nucleotides. There are more than 1000 miRNAs known to human beings, which have various important regulatory functions in cells. miRNA can regulate many genes in the human body, that is, each miRNA can have multiple target genes, and multiple miRNAs can also regulate the same gene. There are three ways in which miRNAs regulate genes. The first mode of action is to split the molecular structure of the target gene. In this case, the two are completely complementary in structure, and the function of miRNA is very similar to that of siRNA. Most miRNAs in plants have this mode of action. The second mode of action is to hinder the translation of the target gene. In this case, the two are incompletely complementary in structure. This incomplete complementarity leads to the blockage of the translation of the target gene, which then affects the stability of gene expression. Non-plant organisms This is the most common mode of action found in C. elegans. For example, lin-4 of Caenorhabditis elegans affects the growth and development of C. elegans in this way, but this mode of action is rare in plants. The third way of action is the combination of the first two ways. Some miRNA parts are complementary to the target gene. At this time, it is shown to cut the target gene, while the remaining part is not completely combined with the target gene. At this time, it is shown to hinder the target gene. translate. In view of the small dimension of miRNA expression data, by processing miRNA expression data to obtain disease-risk miRNAs, and then using miRNA target gene mRNA data for analysis, the purpose of predicting disease-related genes can be achieved while reducing data dimensions.

The existing technology directly deals with complex disease mRNA gene expression data with high dimensionality and heavy calculation.

Contents of the invention

The purpose of the present invention is to provide a method for discovering cancer-related genes using miRNA expression data, aiming to solve the problems of high dimensionality and large amount of calculation in the prior art to directly process mRNA gene expression data of complex diseases.

The present invention is achieved in this way, a method for discovering cancer-related genes using miRNA expression data, the method for discovering cancer-related genes using miRNA expression data is based on the pan-cancer project PanCancer under the Cancer Genome Atlas TCGA, using statistical analysis and machine learning Algorithms to analyze and process gene expression data to identify genes associated with complex diseases, including:

Sample data sorting, obtaining miRNA expression data and mRNA expression data of a certain disease, both of which include disease samples and corresponding normal samples;

Perform statistical analysis on the miRNA data to obtain the average expression values of normal samples and disease samples, and the influence of zero values should be excluded in this process;

Sort the miRNAs according to the average change rate, the higher the change rate, the higher the ranking, and select the top 10 miRNAs as related miRNAs;

The five target gene prediction softwares miRanda, miRDB, miRWalk, RNA22, and Targetscan were used as tools for predicting mRNA to obtain the target genes of the corresponding miRNAs, and the selected target genes followed the following conditions: The maximum number of prediction software that predicts the same target gene at the same time, N _k represents the number of genes predicted by K target gene software at the same time, as a pre-selected gene must be at least R (0≤R≤K) target gene prediction software predicted at the same time;

According to the selected mRNA, extract corresponding disease samples and normal samples from the initial mRNA expression data;

Use the Relief algorithm to sort the genes in the above extracted mRNA samples, arrange them in descending order of importance, and take the top 45 genes as the predicted disease-related genes.

Further, when analyzing miRNA data, it is necessary to exclude the influence of zero values. When calculating the mean value of miRNA, first calculate the number m of non-zero values in each sample, and then calculate the sum of the expression values of miRNA samples Sum, then calculate The mean value of the sample is Sum/m, the mean value of the expression value of the normal sample is n, and the mean value of the expression value of the disease sample is c, then the corresponding expression value change rate is |n-c|/n, and the sample expression value change is determined according to the mean change rate of miRNA The top 10 miRNAs are related miRNAs.

Further, as a preselected gene, at least R (0≤R≤K) target gene prediction software must be simultaneously predicted. When N _k >10, R=K; when N _k <10 and N _k-1 >10, R =K-1; similarly, if N _k-1 <10 and N _k-2 >10, then R=K-2, and so on.

Furthermore, the selected feature selection method is the Relief algorithm, and the feature weight calculation formula is as follows:

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; -</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; Same</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; +</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; miss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; k</span>}}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; ;</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; ;</span>$

Among them, s _i (i=1,...,p) represents the i-th sample, p is the number of samples; Same _j represents the j-th similar sample of s _i , Miss _j represents the j-th heterogeneous sample of s _i , k represents the number of neighbors; w _f (f=1,...,q) represents the weight of feature f, that is, the importance of the fth pre-selected gene, q is the number of pre-selected genes; r represents the sampling frequency;

The function diff is defined as follows:

$\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; |</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; f</span>}}}{{\mathrm{<span\; class="notranslate">\; Max</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Min</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; ;</span>$

Among them, s _if represents the value of feature f on the i-th sample, s _jf represents the value of feature f on the j-th sample, Max _f represents the maximum value of feature f in the sample, and Min _f represents the feature f The minimum value in the sample, the number of sampling r=10, the number of neighbors k=20, the number of iterations of the Relief algorithm is 30 times, and the weight W={w ₁ ,w ₂ ,…,w _q } of each feature is calculated, and The mRNAs are ranked according to weight W.

Another object of the present invention is to provide a system for the method of using miRNA expression data to discover cancer-related genes, the system comprising:

The sample data sorting module is used to obtain miRNA expression data and mRNA expression data of a certain disease, both of which include disease samples and corresponding normal samples;

The statistical analysis module is used to perform statistical analysis on miRNA data, and obtain the average expression values of normal samples and disease samples respectively, and the influence of zero value should be excluded in this process;

The screening and ranking module is used to sort the miRNAs according to the average rate of change, the higher the rate of change, the higher the ranking, and filter the top 10 miRNAs as related miRNAs;

The selected target gene module is used to apply miRanda, miRDB, miRWalk, RNA22, and Targetscan five target gene prediction software as a tool for predicting mRNA to obtain the target gene of the corresponding miRNA. The selected target gene follows the following conditions: For the five used Target gene prediction software, assuming that K represents the maximum number of prediction software that predicts the same target gene at the same time, and N _k represents the number of genes predicted by K target gene software at the same time;

The extraction module is used to extract corresponding disease samples and normal samples from the initial mRNA expression data according to the selected mRNA;

The sorting module is used to sort the genes in the above-mentioned extracted mRNA samples using the Relief algorithm, arrange them in descending order of importance, and take the first 45 genes as the predicted disease-related genes.

Further, the statistical analysis module further includes:

The non-zero value calculation unit is used to calculate the miRNA mean value, and first calculate the number m of non-zero values in each sample;

The sample mean value calculation unit is used to obtain the sum Sum of miRNA sample expression values, and calculate the sample mean value as Sum/m;

Expression value change rate calculation unit, the average expression value of normal samples is n, and the average expression value of disease samples is c, then the corresponding expression value change rate is |n-c|/n;

The ranking unit is used to determine the top 10 miRNAs in the change rate of sample expression value as related miRNAs according to the mean change rate of miRNAs.

Another object of the present invention is to provide a bio-targeted treatment system using the method for discovering cancer-related genes using miRNA expression data.

Another object of the present invention is to provide a biopharmaceutical development process using the method for discovering cancer-related genes using miRNA expression data.

Another object of the present invention is to provide a pathogenic mechanism elucidation system using the method for discovering cancer-related genes using miRNA expression data.

Another object of the present invention is to provide a disease-causing risk prediction system using the method for discovering cancer-related genes using miRNA expression data.

The method for discovering cancer-related genes using miRNA expression data provided by the present invention is based on the pan-cancer project PanCancer under the Cancer Genome Atlas TCGA (The Cancer Genome Atlas), and uses statistical analysis and machine learning algorithms to analyze and process gene expression data to identify complex diseases. related genes. The invention can discover multiple risk genes related to complex diseases such as cancer, which is of great significance to the biological target therapy of complex diseases, the development of biological drugs, the interpretation of pathogenic mechanism and the risk prediction, etc., and can be designed according to the obtained risk genes Gene-targeted therapy; select highly sensitive drugs or develop new drugs based on gene markers; based on the discovered related genes, it can analyze the development process of complex diseases to determine their formation mechanism; it can also conduct susceptibility genes for predicted risk genes testing to reduce the risk of disease. The present invention takes into account that the large amount of mRNA expression data makes it difficult to process sample data, so miRNAs with small data amounts and regulatory effects on mRNA are used as analysis points. The prior art deals with more than 20,000 dimensional mRNA expression data, while this method The analysis is more than 1,000-dimensional miRNA expression data, and the dimension is reduced by 20 times, so the calculation complexity is reduced, the calculation time is shortened, and unfavorable factors such as excessive calculation time caused by a large amount of data are avoided. The present invention uses miRNA to rapidly locate pathogenic mRNA, and is not limited to a certain complex disease or a certain cancer, but can use this method to analyze related genes for all complex diseases. The present invention determines the target gene by analyzing the miRNA expression data of a certain disease, and screens the risk gene from the mRNA expression data of the target gene, and does not need any information related to the disease. Therefore, as long as the miRNA and mRNA expression data of a certain disease are given, this method can be applied for analysis and has wide applicability.

Description of drawings

Fig. 1 is a flowchart of a method for discovering cancer-related genes using miRNA expression data provided by an embodiment of the present invention.

detailed description

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the examples. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

The present invention utilizes the characteristics of small amount of miRNA expression data and the targeting relationship between miRNA and mRNA to obtain cancer-related genes by analyzing miRNA expression data, thereby solving the problem of excessive data volume when directly using mRNA expression data analysis in the prior art .

The application principle of the present invention will be described in detail below in conjunction with the accompanying drawings.

As shown in Figure 1, the method for discovering cancer-related genes using miRNA expression data according to the embodiment of the present invention includes the following steps:

S101: Screen miRNAs related to cancer based on the expression difference of miRNAs in normal samples and disease samples;

S102: Utilizing the mapping relationship between miRNA and mRNA, targeting the mRNA on which the miRNA acts;

S103: Discover cancer-related genes by analyzing targeted mRNA expression data.

The data used in the present invention, including normal and disease samples in miRNA and mRNA, are all from the pan-cancer research project of TGCA.

The specific implementation steps of the present invention are as follows.

Step 1, data processing

Divide the sample data into the following four groups: miRNA expression data of normal samples, miRNA expression data of disease samples, mRNA expression data of normal samples, and mRNA expression data of disease samples. Here, it is necessary to ensure that the miRNA and mRNA names of the samples are consistent.

Step 2: Screen out miRNAs with a high rate of change in expression values

1. For the miRNA data, calculate the mean value of the corresponding expression value of each miRNA in the normal and disease sample data. Because the miRNA expression value in the sample is now 0, the number of zero values needs to be counted during processing. If the total number of non-zero values in a certain miRNA sample is m, and the sum of the sample values is Sum, then the The mean of miRNA samples is Sum/m. This method was used to calculate the mean value of normal and disease samples for each miRNA.

2. Calculate the change rate of each miRNA according to the sample mean value. If the normal sample mean value of a certain miRNA is n, and the disease sample mean value is c, then the change rate is |n-c|/n.

3. Sort all miRNAs according to their rate of change, and select the top 10 miRNAs with a large rate of change.

Step 3, get the target gene of the selected miRNA

Using five target gene prediction software, miRanda, miRDB, miRWalk, RNA22 and Targetscan, the target genes of selected miRNAs were obtained. For the five target gene prediction software used, it is assumed that K represents the maximum number of prediction software that simultaneously predicted the same target gene, and N _k represents the number of genes predicted by K target gene software at the same time. This method requires at least R (0≤R≤K) target gene prediction software to simultaneously predict the preselected gene. When N _k >10, R=K; when N _k <10 and N _k-1 >10, R=K-1; similarly, if N _k-1 <10 and N _k-2 >10, then R=K-2, and so on.

Step 4: Use the Relief algorithm to screen related genes

According to the selected target gene mRNA, normal samples and disease samples were screened out from the mRNA expression data, and the data of the two types of samples were integrated together, and the mRNA was sorted in descending order of importance using the Relief algorithm. The number of times r=10, the number of neighbors k=20, and the number of iterations of the algorithm is 30 times. According to the results of Relief sorting, the top 45 genes were selected as the predicted related genes.

The application effects of the present invention will be described in detail below in conjunction with experiments.

In experiment 1, the breast cancer expression data (BRCA) in the TCGAPanCancer project was selected as the experimental object, and there were 1045 miRNAs and 20530 mRNAs in the data. Breast cancer expression data were processed according to the above experimental steps:

1. Import the miRNA sample data, first screen all the normal sample data to ensure that there are no zeros in the normal sample, if there is a miRNA in the normal sample data that is all zero, delete the miRNA data in the normal sample at the same time, Data corresponding to miRNAs in disease samples were also deleted.

2. For the screened miRNA data, average the normal and disease samples respectively, and calculate the rate of change.

3. Rank the miRNAs according to their rate of change, and select the top 10 miRNAs with the highest rate of change after sorting. In this experiment, the following 10 MicroRNAs were finally selected: hsa-mir-133b, hsa-mir-133a, hsa-mir-208b, hsa-mir-206, hsa-mir-551b, hsa-mir-145, hsa-mir- 378, hsa-mir-451, hsa-mir-144, hsa-mir-1.

4. For the selected miRNAs, use the aforementioned five target gene prediction software to predict target genes, and obtain 725 target gene mRNAs.

5. Select the data corresponding to 725 target gene mRNAs from the mRNA data, and then use the Relief algorithm to sort the importance of the mRNAs, set the number of sampling r=10, the number of neighbors k=20, and the number of algorithm iterations is 30 times, The top 45 important mRNAs were selected as related genes. The 45 mRNAs are as follows: RXFP2, GYPA, OTX2, PRDM9, CYP11B1, MMD2, CHRNA4, NEUROD1, PABPC1L2B, RIT2, CNTN5, NEUROD4, SLC4A1, PRDM7, FBXO40, GABRG2, GPR6, ZIC3, SPINLW1, DMRT1, CYP3A4, DPCR1, LHX9 , ISL2, LIPI, SOST, HHLA2, S100A7, RIPPLY1, TRHDE, BMP3, KCNMB2, PAX5, PAX3, ANGPT4, DSCAM, EREG, OR7D2, DRD1, GFRA3, LEP, GPR26, LIX1, ZIC1, GDAP1L1.

The function of the obtained genes in breast cancer and their functional relationship with known important genes of breast cancer will be analyzed below to illustrate the correlation between these genes and breast cancer, thereby verifying the effectiveness of the method proposed in this study.

NEUROD1 is a basic bHLH transcription factor of the NeuroD family. It can combine with other bHLH transcription factors to produce heterodimers and activate the transcription of a special DNA sequence called E-box. It also contributes to various cell differentiation pathways regulation. HeidiFiegl found that NEUROD1 was abnormally methylated in breast tissue tumors and lung tumor samples, and the higher the degree of tumor progression, the higher the level of methylation in samples. The protein encoded by SLC4A1 is a member of the AE protein family, which plays an important role in red blood cells, as a transport protein medium to help the corresponding substances pass through the cell membrane. A Gorbatenko's study showed that SLC4A1 was down-regulated in all breast cancer subtypes, suggesting that SLC4A1 may have some influence on breast cancer lesions. CYP3A4 encodes the cytochrome P450 enzyme involved in the metabolism of half of today's drugs, such as acetaminophen, codeine, cyclosporine A, and diazepam and erythromycin, as well as some steroids and carcinogens metabolism. CKeshava found that the mutation of CYP3A4 may lead to the imbalance of hormone metabolism in breast cancer, and may also activate foreign substances to cause cancer. It is an important gene related to breast cancer. The protein encoded by HHLA2 exists on the surface of monocytes. This protein can bind to receptors on lymphocytes, thereby regulating cell-mediated immunity and inhibiting the proliferation of monocytes. MJanakiram found that the copy number of HHLA2 increased by 29% in breast cancer by analyzing TCGA-related expression data, resulting in overexpression of HHLA2 in breast cancer, which shows that HHLA2 has a certain impact on breast cancer lesions. The protein encoded by S100A7 is a member of the S100 protein family. S100 proteins are widely present in the cytoplasm and nucleus, and participate in many cellular processes, such as the regulation of cell cycle and differentiation. Emberley detailed the research status of S100A7 in breast cancer and the specific mode of action and expression of S100A7 in breast cancer. Similarly, Haddadd also explained the relationship between S100A7 and breast cancer. PAX3 is a member of the PAX transcription factor family, which contains a paired box domain and a paired homeodomain, and these genes play a very important role in fetal development. In his article, WJ Tan specifically described the clinical expression of PAX3 in breast cancer, and analyzed the impact of PAX3 on breast cancer. LEP encodes a protein secreted by white blood cells. LEP plays an important role in regulating body weight. It can inhibit food intake and regulate energy expenditure. Cleveland described the relationship between LEP gene variation and the incidence of breast cancer. This shows that the abnormal expression of LEP may not only lead to weight imbalance, but also may cause breast cancer lesions.

From the above analysis, it can be seen that the predicted genes will have an impact on breast cancer lesions, but the specific pathogenic principles of these genes still require in-depth analysis by relevant technical personnel.

The pathway analysis tools in the David database and the STRING-DB database are used to analyze the predicted genes as a whole. These two analysis methods can explain from the side that the predicted gene causes the occurrence of the disease through its effect on the important genes of cancer, and verifies the correlation between the predicted gene and the cancer gene. The important genes of breast cancer selected in this experiment are PIK3CA, TP53, PTEN, AKT1 and SF3B1.

Using the David database, it was found that there is a risk pathway between the predicted gene and the important gene, and there is also a related risk pathway in the predicted gene. There are pathways between EREG and PIK3CA and AKT1 in important genes, and there are risk pathways between LEP and PIK3CA and AKT1. It is worth noting that there are also related pathways between LEP and screening genes DRD1, GABRG2 and RXFP2, as shown in Table 1. Pathway analysis also found that there is a biological metabolic connection between LEP, EREG and breast cancer important genes PIK3CA and AKT1, and there is a pathway connection between DRD1, GABRG2, RXF2 and LEP, which may lead to the occurrence of the disease source.

Table 1 Pathways involved in breast cancer-related genes

For the important genes of breast cancer and the 45 genes screened out, check the interaction relationship between them on STRING-DB, and the analysis results are shown in Table 2. Some genes have no connection with other genes, such as OR7D2, HHLA2, DPCR1, etc., which does not mean that they have no effect on breast cancer. These genes may act on breast cancer alone (such as HHLA2, which has been analyzed before. lesions), and may also interact with other important genes in breast cancer. There are many interactions among the remaining genes, and these genes constitute a relationship network, and the predicted genes may positively or negatively affect the important breast cancer genes in the network through certain biological functions, thus resulting in breast cancer lesions.

Table 2 Association between breast cancer-related genes and important genes

In the second experiment, the kidney cancer expression data (KIRC) in the TCGAPanCancer project was selected as the experimental object, and there were 1045 miRNAs and 20530 mRNAs in the data.

Using the same method as in Experiment 1, the rate of change of samples was sorted, and the top 10 samples were selected as target miRNAs after sorting. Selected miRNAs are as follows: hsa-mir-200c, hsa-mir-514b, hsa-mir-506, hsa-mir-508, hsa-mir-514-2, hhsa-mir-141, hsa-mir-514- 3, hsa-mir-514-1, hsa-mir-184, hsa-mir-934. Using the aforementioned five target gene prediction software to predict the target genes of the selected miRNAs, 504 mRNAs were obtained. These mRNA sample data were selected from the initial mRNA expression data, and then the weight was calculated using the Relief algorithm, and the top 45 mRNAs were selected as target mRNAs. The 45 selected mRNAs are as follows: ODAM, KLHL1, TAC1, NPY2R, HYAL4, FOXE1, TTR, SLC6A14, GLRA3, FUT9, GRIA2, KCNA1, CXorf41, TFAP2B, SFTPB, CRISP1, PDE6H, AGXT2L1, LHFPL4, SLC30A8, STXBP5L, TMEM196, IL1F5, ASTN1, CRISP3, HTR2C, LIN28B, TRIM42, KIAA1486, COL9A1, GCM1, TNNI1, SCG3, ANXA10, BTC, SORCS1, KCND2, LRRN1, MSTN, ERBB4, PRG4, NAPB, ARHGAP12, C12orf53, RAD52.

The function of the obtained genes in renal cancer will be analyzed below to illustrate the correlation between these genes and renal cancer, so as to verify the effectiveness of the method proposed in this study.

KLHL1 is a protein-coding gene, a member of the muscle tissue protein family, expressed in tissue cells of the kidney, and also expressed in many brain tissues. This suggests that mutations in KLHL1 may lead to functional problems in certain cells in the kidney, which could affect the development of cancer in the kidney. The protein encoded by NPY2R is the receptor for Y2 in neuropeptide (NPY) Y, and NPY receptors are involved in a variety of biological actions, including food intake, stimulation of anxiety, circadian pain modulation, and transmission and control of pituitary hormone release. There are 293 kinds of cells in the human kidney that are regulated by genes including NPY2R. It can be seen that NPY2R plays an important role in kidney function, and whether the normal expression of NPY2R gene will play a role in kidney disease. KCNA1 belongs to the potassium channel 6-TM family whose genes include active Ca2(+-), which contributes to the formation of tetramers, and also participates in the formation of subfamily proteins, such as KV1.1, KV1.2 , KCNQ2 and KCNQ3 etc. Mutations in KCNA1 affect the function of the kidney. Analysis of overexpressed human kidney cells revealed that KCNA1 was mutated in a non-functional region and also negatively affected the functional region of KV1.1. FOXE1 is a member of the transcription factor family, and this gene may have related effects on the mutation of thyroid disease, and then play a certain role in kidney disease. It has been reported that genes including FOXE1 can affect the kidney by affecting the function of the thyroid gland, leading to kidney deformities and lesions. The enzyme encoded by SLC6A14 is a member of the lytic carrier family 6. The lytic carrier family is mainly used to help transport sodium and chlorine elements in human neurons. The encoded protein is also involved in the transport of neutral and cationic amino acids, and also acts as a β-amino acid Propionic acid carrier. It has been found that SLC6A14 is overexpressed in renal lesion tissues, which indicates that the variation of SLC6A14 may have an impact on renal function. The Rex oligosaccharide fucosyltransferase encoded by FUT9 is a member of the glycosyltransferase family. It mainly exists in the Golgi apparatus and plays an important role in the development of organ embryos. FUT9 is also responsible for regulating CD15 in mature granules. expression in cells. A 1.8-fold decrease in the expression of FUT9 in the kidney will lead to a 1.8-fold increase in the expression of CD24A in the kidney, which will directly and seriously affect the normal function of the kidney. Although FUT9 does not directly cause renal lesions, it indirectly affects renal function by affecting the expression of CD24A, and its role in the kidney cannot be underestimated. STXBP5L is an important paralogous gene that encodes a protein that binds to syntaxin. As a protein, STXBP5 interacts with synaptic neurons, exists in large quantities in the kidney, and also plays a significant role in affecting kidney function, indicating that the variation of STXBP5L has a great impact on kidney function.

The KEGG pathway tool of the David database and the STRING-DB database are used to conduct an overall analysis of the discovered kidney cancer-related genes. The important KIRC genes selected here are TP53, CDH1, VEGFA, MUC1 and EGFR. Through the analysis, it was found that there were associated pathways between the predicted genes and important genes, and there were also associated pathways between the predicted genes. There is a risk pathway between BTC, ERBB and EGFR, an important gene of kidney cancer. There are also pathways associated with EGFR, such as VEGFA and HTR2C. The pathway diagram of EGFR also has pathways of VEGFA, GRIA2 and TP53. In addition, the predicted kidney cancer-related genes HTR2C, There are also risk pathways among GRIA2, GLRA3, and NPY2R, as shown in Table 3.

Table 3 Pathways involved in renal cancer-related genes

The STRING-DB database was used to check the interaction between the predicted kidney cancer-related genes and important genes, and the results are shown in Table 4. It can be seen from Table 4 that there are many connections between the predicted genes and important genes, indicating that the predicted genes may act on the important genes of kidney cancer in some ways, thereby affecting the normal expression of important genes and leading to kidney cancer lesions. . Four genes, TTR, KCNA1, FOXE1 and ODAM, should be paid special attention to. They are related to many important genes of renal cancer and may act on multiple important genes at the same time.

Table 4 Association between kidney cancer-related genes and important genes

Working principle of the present invention:

By analyzing the small-dimensional miRNA expression data and utilizing the regulatory effect of miRNA on genes, a low-dimensional subset is targeted in the high-dimensional mRNA expression data, and then the importance of genes in each dimension is determined by using the Relief algorithm. This screens out genes associated with complex diseases. The Relief algorithm is a feature weight algorithm proposed by Kira and Rendell in 1992. It is trained through samples, and the classification weight of the sample features is obtained according to the training. The larger the weight, the greater the significance of the feature for classification. The feature weight calculation formula in the Relief algorithm is as follows:

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; -</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; Same</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; +</span>\frac{\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; k</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; miss</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; k</span>}}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; ;</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span><span\; class="notranslate">\; ...</span><span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; p</span>}$

Among them, s _i (i=1,...,p) represents the i-th sample, p is the number of samples; Same _j represents the j-th similar sample of s _i , Miss _j represents the j-th heterogeneous sample of s _i , k represents the number of neighbors; w _f (f=1,...,q) represents the weight of feature f, that is, the importance of the fth pre-selected gene, q is the number of pre-selected genes; r represents the sampling frequency. The function diff is defined as follows:

$\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; |</span>\frac{{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; the\; s</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}\mathrm{<span\; class="notranslate">\; f</span>}}}{{\mathrm{<span\; class="notranslate">\; Max</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; Min</span>}}_{\mathrm{<span\; class="notranslate">\; f</span>}}}<span\; class="notranslate">\; |</span>$

Among them, s _if represents the value of feature f on the i-th sample, s _jf represents the value of feature f on the j-th sample, Max _f represents the maximum value of feature f in the sample, and Min _f represents the feature f The minimum value in the sample. According to the set sampling times and the number of neighbors, the Relief algorithm obtains the weight W={w ₁ ,w ₂ ,…,w _q } of each feature through multiple iterations, and then sorts the features according to the weight W.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present invention should be included in the protection of the present invention. within range.

Claims (10)

1. one kind utilizes the method that miRNA expression data finds cancer related gene, it is characterized in that, the described method utilizing miRNA expression data to find cancer related gene is based on the general cancer project PanCancer under cancer gene group collection of illustrative plates TCGA, use statistical analysis and machine learning algorithm, it is analyzed gene expression data processing, identifies the gene relevant to complex disease；Including:

Sample data arranges, and obtains miRNA expression data and the mrna expression data of certain disease, and two kinds of data all comprise disease sample and corresponding normal sample；

MiRNA data being carried out statistical analysis, tries to achieve the mean expression value of normal sample and disease sample respectively, this process to get rid of the impact of null value；

By miRNA by the sequence of Change in Mean rate, the ranking that rate of change is more big is more forward, screens 10 forward miRNA of ranking as relevant miRNA；

Application five microRNA target prediction softwares of miRanda, miRDB, miRWalk, RNA22, Targetscan are as the instrument of prediction mRNA, obtain the target gene of corresponding miRNA, selected target gene follows following condition: for five microRNA target prediction softwares used, assume that K represents the maximum of the forecasting software number simultaneously predicting identical target gene, N_kRepresent simultaneously by the gene number of K target gene software prediction, at least to be predicted by R (0≤R≤K) individual microRNA target prediction software as preliminary election gene simultaneously；

According to the mRNA chosen, go out corresponding disease sample and normal sample from initial mrna expression extracting data；

Utilize Relief algorithm that the gene in said extracted mRNA sample out is ranked up, arrange from big to small by importance, take front 45 genes disease related gene as prediction。

2. utilize the method that miRNA expression data finds cancer related gene as claimed in claim 1, it is characterized in that, when miRNA data are analyzed, get rid of the impact of null value, when asking for miRNA average, first obtain the number m of nonzero value in each sample, then miRNA sample expression values summation Sum is tried to achieve, then calculating sample average is Sum/m, normal sample expression values average is n, disease sample expression values average is c, then corresponding expression values rate of change is | n-c |/n, Change in Mean rate according to miRNA, before determining sample expression values rate of change ranking, the miRNA of 10 is relevant miRNA。

3. utilize the method that miRNA expression data finds cancer related gene as claimed in claim 1, it is characterised in that at least to be predicted by R (0≤R≤K) individual microRNA target prediction software as preliminary election gene, N simultaneously_kDuring > 10, R=K；Work as N_k< 10 and N_k-1During > 10, R=K-1；In like manner, if N_k-1< 10 and N_k-2> 10, then R=K-2, by that analogy。

4. utilize the method that miRNA expression data finds cancer related gene as claimed in claim 1, it is characterised in that the feature selection approach chosen is Relief algorithm, and feature weight computing formula is as follows:

$w_f=w_f-\frac{\underset{j=1}{\overset{k}{\&Sigma;}}diff(f,s_i,{\mathrm{Same}}_j)}{rk}+\frac{\underset{j=1}{\overset{k}{\&Sigma;}}diff(f,s_i,{\mathrm{Miss}}_j)}{rk}f=1,\mathrm{...},q;i=1,\mathrm{...},p;$

Wherein, s_i(i=1 ..., p) represent i-th sample, p is number of samples；Same_jRepresent s_iThe similar sample of jth, Miss_jRepresent s_iJth foreign peoples's sample, k represents neighbour's number；W_f(f=1 ..., q) represent the weight of feature f, i.e. the significance level of the f preliminary election gene, q is the number of preliminary election gene；R represents frequency in sampling；

Function diff definition is as follows:

$diff(f,s_i,s_j)=\left|\frac{s_{if}-s_{jf}}{{\mathrm{Max}}_f-{\mathrm{Min}}_f}\right|;$

Wherein, s_ifRepresent feature f value on i-th sample, s_jfRepresent feature f value on jth sample, Max_fRepresent feature f maximum in the sample, Min_fThen representing feature f minima in the sample, frequency in sampling r=10, neighbour's number k=20, Relief algorithm iteration number of times is 30 times, calculates the weight W={w of each feature₁,w₂,…,w_q, and according to weight W, mRNA is sorted。

5. the system of the method utilizing miRNA expression data to find cancer related gene as claimed in claim 1, it is characterised in that described system includes:

Sample data sorting module, for obtaining miRNA expression data and the mrna expression data of certain disease, two kinds of data all comprise disease sample and corresponding normal sample；

Statistical analysis module, for miRNA data are carried out statistical analysis, tries to achieve the mean expression value of normal sample and disease sample respectively, and this process to get rid of the impact of null value；

Screening ranking module, for being sorted by Change in Mean rate by miRNA, the ranking that rate of change is more big is more forward, screens 10 forward miRNA of ranking as relevant miRNA；

Selected target gene module, for applying five the microRNA target prediction softwares of miRanda, miRDB, miRWalk, RNA22, Targetscan instrument as prediction mRNA, obtain the target gene of corresponding miRNA, selected target gene follows following condition: for five microRNA target prediction softwares used, assume that K represents the maximum of the forecasting software number simultaneously predicting identical target gene, N_kRepresent simultaneously by the gene number of K target gene software prediction, at least to be predicted by R (0≤R≤K) individual microRNA target prediction software as preliminary election gene, N simultaneously_kDuring > 10, R=K；Work as N_k< 10 and N_k-1During > 10, R=K-1；In like manner, if N_k-1< 10 and N_k-2> 10, then R=K-2, by that analogy；

Extraction module, for according to the mRNA chosen, going out corresponding disease sample and normal sample from initial mrna expression extracting data；

Order module, for utilizing Relief algorithm that the gene in said extracted mRNA sample out is ranked up, arranges from big to small by importance, takes front 45 genes disease related gene as prediction。

6. system as claimed in claim 5, it is characterised in that described statistical analysis module farther includes:

Unit asked for by nonzero value, when being used for asking for miRNA average, first obtains the number m of nonzero value in each sample；

Sample average computing unit, is used for trying to achieve miRNA sample expression values summation Sum, then calculating sample average is Sum/m；

Expression values rate of change computing unit, normal sample expression values average is n, and disease sample expression values average is c, then corresponding expression values rate of change is | n-c |/n；

Ranking unit, for the Change in Mean rate according to miRNA, it is determined that before sample expression values rate of change ranking, the miRNA of 10 is relevant miRNA。

7. the Biological target therapy system applying the method utilizing miRNA expression data discovery cancer related gene described in claim 1-4 any one。

8. the bio-pharmaceutical development technology applying the method utilizing miRNA expression data discovery cancer related gene described in claim 1-4 any one。

9. the pathogenesis system applying the method utilizing miRNA expression data discovery cancer related gene described in claim 1-4 any one。

10. the pathogenic Risk Forecast System applying the method utilizing miRNA expression data discovery cancer related gene described in claim 1-4 any one。