Repository navigation
Releases
Releases · steineggerlab/folddisco
Folddisco Release 3-3e193e3
Folddisco Release 3 improves default ranking, adds geometry- and substitution-tolerant search and novelty screening, and loads large indices much faster. On M-CSA catalytic sites, Sens@1FP rises from 0.45 to 0.5 (report).
New features
--sensitive(=--expand-radius 2): also searches neighbouring distance and angle bins, recovering motifs whose geometry deviates from the query.--aa-subst blosum62|group|size, or:*per residue: allows similar amino acids, ranked below exact matches.--confident: reports only matches covering most of the query within 1 Å RMSD.--novelty-mode: reports one KNOWN, PARTIAL or NOVEL verdict per query, with the best hit as evidence.index --expand-radius,--aa-subst: Optional index-time expansion for small databases.- Multi-character and numeric chain IDs (
AA_250,10_250,--chain-sep), suggested by @adobles96 (#44), with thanks to @Immortals-33 (#48) and @kamurani (#33).
Changes
- Results are now sorted by
match_scoreper match andstructure_scoreper structure instead ofidf. Use--sort-by idf:desc,rmsd:ascto restore the previous order.
Performance
- Index lookups are memory-mapped and cached on first use, so the AFDB50 lookup loads in about 18 ms instead of 3 s.
Bug fixes
- Substituted residues (
:H,:ND) are no longer dropped during matching. - Malformed
-qstrings now produce an error message instead of a panic.
Full Changelog: 2-9375a2d...3-3e193e3
Folddisco Release 2-9375a2d
Folddisco Release 2 at a glance: new sparse index format, E‑value statistics for hits, structure-similarity metrics with better sorting
and filtering, new `analyze` command, column‑based output formatting, and faster hashing/retrieval. Folddisco has its own Marv now and was published:
Breaking changes
- New unified index format. The old
SimpleHashMap/alloc indexing modes were replaced by a singleFolddiscoIndex(built on the newSparseIndex). Not compatible with older indices (cb099b5, 157497d, ac1e91e, 9dd06f1). - New hashes are limited to 30 bits, changing the on‑disk hash layout (21300eb).
- Default sort strategy is now IDF+RMSD (1a33f06, 4579cff).
New features and enhancements
- E‑value estimation for matches based on the IDF score, with per‑match IDF and multiple fitting functions (#32, d395c8a, c452eca, fdfeadb, 5ab3881, 3043123, de68e15) by @jwyoon05.
- Structure‑similarity metrics exposed to the CLI, enabling result filtering and metric‑aware sorting (16696a4, 408ffa9, a99e45a, 013f5c8).
- Column‑based TSV/CSV output formatting and updated match‑result header (88dc390, 1e115a9, f92169f).
mmaptype now handles index files larger than available memory (765aeeb).- Version handling via environment variable for build environments without git (716f25f).
- Two new binning schemes, including shift binning for higher sensitivity (217bb7d, d3700a8, c708dea).
- Improved foldcomp path resolution (95abc73, 8d7a776).
Performance
- Faster hashing with
FxHasher(23133c0). - Accelerated residue mapping (5f3796b).
- Residue retrieval checks distances first, then fetches features, and drops an unnecessary distance‑map copy (c6f57e5, 60a70e0).
- Query residues are no longer sorted unnecessarily (d589728).
Bug fixes
- Fixed CIF parsing for structures with numeric values (e.g. ligand code
919) and relative paths; no longer truncates multi‑char chain IDs (#41, 1bab90c) by @adobles96. - Fixed duplicated entry issue in the index (82cd989).
- Fixed a benchmark bug (#29, 3392bf2).
Documentation & developer notes
- Add
analyzecommand for index distribution analysis: reports total/empty/density and amino‑acid pair counts, with a--top-noption (10d42cb, c28c502, 58aa0f7, 7d6c911, 10eb601). - Add docker containers and conda/docker install instructions to README (9c13b9d, c6ee2ef).
- Updated preprint version, download links, and pre‑built indices; added Cargo and PDB dataset links (f928970, 335a295, 9375a2d).
- Removed obsolete doctests and the foldcomp feature test; test cleanup (2ebad32, 0f1d715, 5fedfb4).
Folddisco Release 1-7514114
First release of Folddisco
Folddisco is tool for searching discontinuous motifs in protein structures.
It is designed to handle large-scale protein databases with efficiently, enabling the detection of structural motifs across thousands of proteomes or millions of structures.
Publications
Webserver
Search protein structures motifs against the AlphaFoldDB and PDB in seconds using the Folddisco webserver (code): search.foldseek.com/folddisco 🚀