Post

PREPRINT (not peer-reviewed). For one drug-resistance dataset, a mutation's position was more informative than tested biophysical scores. Jia He mapped variants from four fibroblast growth factor receptor (FGFR) subtypes onto common residue positions, then tested predictions on each subtype held out from training. Across 7,522 unique variants, the position-based lookup had AUROC scores of 0.84–0.89 for FGFR1–3 and 0.70 for FGFR4; 0.5 represents chance-level ranking. The signal weakened substantially when the alignment was shifted by one residue. This is a way to prioritise sites for study, not a validated clinical predictor of individual variants: discrimination among substitutions at the same position was weak. Find the paper on ChemRxiv by searching its full title: Drug-resistance phenotypes in FGFR deep mutational scanning are carried by a single-residue positional statistic that transfers across FGFR1–3 and weakens in FGFR4.

10:43:22 PM · Nov 22, 2026
Sign in to reply and join the conversation.