Geodesic's NovaAtom-Lite hits 79% on FoldBench antibody-antigen prediction, far above AlphaFold 3's 48%

QuanquanGu · x · 2026-11-11

AI drug discovery startup Geodesic released NovaAtom-Lite, an all-atom structure prediction model for proteins, nucleic acids and their binding molecules. On FoldBench antibody-antigen prediction (DockQ ≥ 0.23) it scores 79.07% (n=172) vs IsoDDE 75.58% and AlphaFold 3's 47.90% (n=167), using 5 seeds × 5 samples with model-ranked top-1 selection. It's live via Geodesic API and NovaDDE. Founding scientist chengyenhsieh also outlined a three-layer thesis for drug discovery: reasoning models with structural biology knowledge, vertical domain predictors, and wet-lab feedback.

Related event: Geodesic Releases NovaAtom-Lite, Beating AlphaFold 3 on Antibody-Antigen Benchmark(7 posts)→

Original post →

More from Apps

Apps channel →