Skip to content

Use the schema's variant types consistently across datasets - #538

Open
jjacobson95 wants to merge 1 commit into
mainfrom
fix-mutation-classes
Open

jjacobson95 wants to merge 1 commit into
mainfrom
fix-mutation-classes

Conversation

@jjacobson95

@jjacobson95 jjacobson95 commented Nov 12, 2026 •

Copy link
Copy Markdown
Collaborator

Datasets named some variant types differently. For example, CPTAC used Start_Codon_SNP where others use Translation_Start_Site, and BeatAML used Frameshift_Variant where others split frameshifts into Frame_Shift_Ins / Frame_Shift_Del. This maps every dataset to the variant types listed in our schema, one name per variant type, and fixes the problems listed in #532:

Dataset Change
cell lines (CCLE, CTRPv2, PRISM, FIMM, gCSI, NCI60, GDSCv1, GDSCv2) each source value mapped once, which removes the duplicate start_lost/stop_lost rows; Start_Codon_* / Stop_Codon_* → Translation_Start_Site / Nonstop_Mutation; undetermined coding indels → in-frame or frameshift by HGVS length
BeatAML frameshifts split into Frame_Shift_Ins/Del by HGVS; protein_altering_variant → in-frame by length; start_lost → Translation_Start_Site
Pancreatic stop-gained → Nonsense_Mutation; 5′ UTR → 5'UTR; frameshifts split by HGVS
Sarcoma frameshifts split by HGVS; intergenic_variant → IGR
CPTAC, Colorectal Start_Codon_* / Stop_Codon_* → Translation_Start_Site / Nonstop_Mutation
Novartis classes read from the protein change instead of text checks; mutations no longer stored as byte strings, written as p.…; removes a ? filter that did nothing

Two exceptions remain: Novartis frameshifts are Frameshift_Variant, since the source only gives them at the protein level, and 17 CPTAC variants the source couldn't classify stay Undetermined.

Tested by running each changed build script end to end: the outputs match the corrected files prepared for this update, and no mutations are added or lost, apart from removing the duplicated cell line rows.

Closes #532

@jjacobson95 jjacobson95 added this to the 2.4 new build milestone Nov 12, 2026
@jjacobson95 jjacobson95 added bug Something isn't working data update labels Nov 12, 2026
@jjacobson95 jjacobson95 self-assigned this Nov 12, 2026
@jjacobson95
jjacobson95 requested a review from sgosline November 12, 2026 00:36
@sgosline

Copy link
Copy Markdown
Member

There are still some problems with the mutation mapping.

IGR is misleading and I think we should remove:

Screenshot 2026-11-12 at 12 08 02 PM

Undetermined is only in cptac, can we just remove:
Screenshot 2026-11-12 at 12 10 16 PM

Two other variants are only in cptac: De_novo_start_in/out of frame

Can we delete these four from the vocabulary?

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working data update

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Mutation variant classifications aren't consistent MAF names across datasets

2 participants