Cherimoya genomics model lifts count score to 0.78, nearly halves VEP error
jmschreiber91 · x · 2026-10-31
A Cherimoya team member shares that a mix of architecture and data optimizations lifted count performance from 0.72 to 0.78 while nearly halving VEP error — key changes included reducing max jitter from 500 to 50 and switching profile convolution to k=1. The thread also notes that observational holdout gains often fail to correlate with variant effect prediction, possibly linked to homology leakage, and that fitting cell-line CNAs can boost scores but ruin external validity.
More from Research
- Lacuna's Standing On traces any repo's research roots — nanoGPT stands on 140 papers — hugo_larochelle · 2026-10-31
- Three Mathematicians Join Lex Fridman to Discuss AI Solving Navier-Stokes, Quasi-Riemann — khademinori · 2026-10-31
- ANSER-Bench: 2,235 analytical search tasks trip up SOTA Deep Research in ~50% of cases — CShorten30 · 2026-10-31
- Stanford: AI reads routine ECGs to flag sudden cardiac death risk in 86% with no known warning signs — oziadias · 2026-10-31
- Zip-NeRF Gets Orders of Magnitude Faster to Train, but Rendering Still Crawls at 3-4 FPS — jonstephens85 · 2026-10-31
- Researchers reconstruct Meta Muse's hidden charter from 884 chain-of-thought samples — sethlazar · 2026-10-31