Peptide outlier guard and decoupled deff gate - #149
Merged
Merged
Conversation
ammarcsj
commented
Oct 1, 2026
ammarcsj
commented
Member
- Peptide outlier filtering may now only lower a protein's significance, never raise it.
- residual_decorrelation_tolerance default update
- New residual_deff_gate_tolerance param, the deff gate no longer uses the pruning tolerance
Peptide outlier candidates are flagged on their two-sided p-value, which carries no direction, while the protein-level aggregation in sum_and_re_scale_zvalues sums *signed* z-values. Discarding a peptide whose z-value opposes the protein's direction therefore raises |sum(z)| and lowers n, making the protein more significant - the opposite of what the filter is for, and contrary to what its docstring claimed. apply_peptide_outlier_filtering now delegates each protein to _filter_and_aggregate_protein, which re-aggregates and, if |z| grew, un-flags every peptide and restores the unfiltered aggregate. Removing peptides can now only ever lower a protein's significance. Verified by replaying both Spectronaut species-mixture benchmarks from their exported iontrees (the replay reproduces the pipeline's stored per-peptide is_outlier_peptide flags and gene p-values exactly): - LargeFC: 1005 of 8595 proteins had their significance inflated, worst case 46x (p 0.945 -> 0.021). All were unregulated background. The sets of significant proteins are identical before and after this change at 5%, 1% and 0.1% FDR, and the empirical FDR is unchanged at 3.30%. - SmallFC: 189 of 10634 affected, worst case 37222x. 7 of 1337 significant proteins are lost (0.5% at 5% FDR), all spiked-in true positives that reached significance through the removal of a single peptide. Empirical FDR unchanged at 1.50%. - No protein gains significance at any threshold in either dataset, and the mechanism never rendered an unregulated protein significant. A sign-aware variant (drop only candidates concordant with the protein) was tested and rejected: it is worse than no fix at all, because the discordant candidate that is then retained keeps inflating the aggregate once its concordant partner has been removed. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
residual_decorrelation_tolerance default 0.10 -> 0.08. New residual_deff_gate_tolerance (default 0.05): the deff gate no longer uses the pruning tolerance. None restores the old coupled behaviour. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
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
1 participant
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.