LoL-align: probability scoring, output, clustering and CI - #595
Open
lasseReifenrath wants to merge 1 commit into
Open
LoL-align: probability scoring, output, clustering and CI#595lasseReifenrath wants to merge 1 commit into
lasseReifenrath wants to merge 1 commit into
Conversation
Collaborator
|
Hello! Did you mean to upload the .gitignore file as well? |
Contributor
Author
|
No, this was by mistake |
Collaborator
|
@lasseReifenrath could remove the .gitignore please? Also |
…--blocklen fixes Add the LoL-align alignment mode (--alignment-type 3) end to end and make it robust for block lengths other than the default 16. Alignment mode: - Homology-probability classifier (LoLAlignProbability) with a 39-feature MLP; store the probability in result.score (*10000) and the normalized LoL-score in result.eval. - tmalign.cpp: self-hit handling for the LoL-align path. - convertalis: register --alignment-type (forwarded by easy-search) and rescale score back to [0,1] on output for the bits/prob/raw columns, the default and empty format-output tab, BLAST+len (-m 2) and HTML (-m 3). SAM (-m 1) stays on the bit-score path. - StructureCluster: dispatch --alignment-type 3 in the cascaded clustering workflow (main alignment + reassignment steps). - Regression: bump submodule to the updated LoL-align search targets (SFAM 0.448889, FOLD 0.27193). - mac-arm64 CI: install rust via rustup.rs (Homebrew rustup no longer ships rustup-init) and export cargo to PATH for the block_aligner build. - README: add the LoL-align publication and document the new parameters. Non-default --blocklen crash fixes (heap corruption / segfault), all in the FwBw/LoLAlign block path: 1. Anchor buffers were sized with the unpadded max sequence length, but the blocklen-padded max_targetLen is passed to initQuery as maxTLen and bounds the anchor-zeroing loop. When blocklen padded past the real max length (e.g. 32/64), the loop wrote past anchorQuery/anchorTarget and corrupted the heap. Size the LoLAlign buffers with the padded max_targetLen. 2. The FwBw SIMD kernel pads each block to a multiple of VECSIZE_FLOAT, so a block length that is not a multiple of VECSIZE_FLOAT overran the per-block buffers (blocklen 17, 63). Round the requested block length up to VECSIZE_FLOAT. 3. When the forward-backward posterior overflows to all-zeros (maxP == 0, only at pathologically large block lengths), recover the alignment the way the refinement phase does: raise the temperature (which shrinks the score exponents back into float range) and re-run the FwBw until a usable posterior is found, giving up past temperature 30 (initScoreMatrix is re-called each retry because it divides the score matrix by the temperature). An empty-anchor bail-out plus bounds guards on the neighborhood loop and alignStartAnchors act as a fallback so the sentinel-0 seed can never index out of bounds. Also register --blocklen on the lolalign command so it can actually be set. Verified valgrind-clean for block lengths 1-512; the healthy range (16/32/64) is numerically unchanged (blocklen-16 output byte-identical), and block lengths that previously produced all-zero posteriors (>= ~80 on ~150-residue proteins) now recover full results.
lasseReifenrath
force-pushed
the
master
branch
from
August 10, 2026 14:40
3660a71 to
1f05441
Compare
Contributor
Author
|
I removed the gitignore, and I will check out the db type Idea. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add the LoL-align alignment mode (--alignment-type 3) end to end: