Repository navigation
Overview of proposed classification and imputation fixes (#46–#50) #51
Description
Salut Eric,
J’espère que tu vas bien !
J’ai repéré quelques points pour lesquels je pense pouvoir donner un coup de main. J’ai préparé cinq PR, chacune avec des tests ciblés, pour que tu puisses les regarder séparément quand tu auras un moment. Il y a notamment un choix d’interprétation sur la classification des ratios, je me suis peut-être trompé sur ce point.
Je te mets le résumé en anglais ci-dessous pour que la discussion reste accessible aux autres utilisateurs.
Thierry
I revisited parts of your workflow and identified several behaviours worth addressing. I submitted five focused PRs and thought it would help to bring them together here.
Classification and summary statistics:
- Make allele-ratio classification boundaries symmetric under REF/ALT swapping #46 makes the ratio-dependent classification boundaries symmetric when REF and ALT are consistently swapped.
- Correct minor-allele carrier counting in the MAS filter #47 corrects minor-allele carrier counting, avoiding double-counting heterozygotes and distinguishing the minor allele from ALT.
- Preserve missing statistics and handle insufficient classification data #48 preserves unavailable statistics as missing, handles insufficient classification data, and separates genotype counts from allele-depth availability.
Imputation safeguards:
- Preserve missing genotypes when no complete donor calls are available #49 leaves missing genotypes unchanged when no complete donor calls are available, instead of assigning 0/0.
- Exclude self-neighbours by sample identity before selecting donors #50 excludes self-neighbours by sample identity rather than assuming the first sorted distance record is the self-comparison.
Each PR includes focused synthetic regression tests and instructions for running them. These test the specific corrections, not the biological validity of the complete workflow or imputation method.
I kept the proposals separate to make review easier. Some touch the same scripts, so their tests should also be rerun together after integration.
For #46 in particular, I'm wondering whether the symmetric boundaries match the intended screening behaviour, since this changes some high-ratio classifications...
This issue can serve as an overview, with implementation discussions remaining in the individual PRs.
Thanks for making the workflow available and maintaining it over the years!
Thierry
Activity
enormandeau commentedon Sep 18, 2026
enormandeau commentedon Sep 28, 2026
I merged 4 of the 5. The proposed changes for the imputation based on similar samples made me rethink what to do in cases where none of the N closest samples had valid genotypes. I will change the current behaviour (using "0/0") but I still want to impute these data.
Thanks for the PRs.