Tag
5 posts
AlphaFold's low-confidence regions aren't always errors — many are intrinsically disordered, where the absence of stable structure IS the function. Why pipelines that penalize disorder systematically misread half the proteome.
AI drug discovery's hardest problem isn't a model — it's a coordination problem between engineers who don't speak biology and biologists who don't ship code. Three high-leverage plays for bridging the two cultures.
Filter NVIDIA AI Enterprise's model catalog to Life Sciences and 14 models appear — spanning genomics, structure prediction, binder design, molecule generation, and docking. Here's where each one sits in the drug discovery pipeline, and what's conspicuously still missing.
OpenFold3 predicts the 3D structure of any protein complex. NVIDIA AI Enterprise makes it callable as an API. I wired them together into a voice-guided molecule viewer — so anyone, not just structural biologists, can explore a folded protein.
Graphics cards turned game worlds into clarity. The same engine could turn drug discovery into a fluid, real-time visual workflow. The Scientific Visual OS thesis.