Repository navigation
indel search: label each group's hits from its own index - #37
Merged
Merged
Conversation
indel_search splits a batch by query length (short queries on a 2-mer table, the rest on their own k) but joined every group's hits to the last group's index for metadata. Protein numbers are each index's own FASTA ordinals, so when the per-k .pepidx files come from different versions of the FASTA (they're reused by name, eg. after the FASTA is edited in place) a short query's hit got named after whichever protein holds that number in the other file. Each group's frame is now built from its own index and the frames are stacked with vertical_relaxed. With consistent index files nothing changes. An all-miss indel search also returned Matched Sequence and Indel Positions as null-typed columns; they're cast to String so every search has the same schema. Recall is unchanged: 100% on all 11 local indel benchmark sets with outputs identical to master, and 9,000 random mixed batches (a third with stale tables) match the brute-force oracle. New tests cover per-index labeling with stale tables, a batch equal to its short and long queries searched separately, and the all-miss types. test_metadata_rows' stale-index test now checks the error hint directly, since indel search no longer reads another group's index.
With one edit, a placement is valid exactly where the two strings agree before it and after it, so the valid positions are one run between the common prefix and the common suffix. _indel_placements now reads that run off the two lengths instead of building agreement runs for every shift and walking them. Two-edit pairs keep the general path. Placements are identical and in the same order: test_indel_positions_equivalence checks them against the original brute-force implementation exhaustively over small alphabets, and 1,000,000 random pairs agree as well. Annotation time halves, and output building for 50k 1-indel queries drops from 0.91 s to 0.65 s.
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
2 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.
Indel search splits a batch by query length: short queries go on a 2-mer table and the rest use their own k. Every group's hits were labeled from the last group's index, so if the per-k .pepidx files came from different versions of the FASTA, a short query's hit could be named after the wrong protein. Each group now uses its own index.
Also:
Recall is unchanged: 100% on all 11 indel benchmark sets with output identical to master (every column, row order and type), and 9,000 random mixed batches match the brute-force oracle. Full test suite passes on top of #35 and #36.