Stanford's Kundaje: SOTA perturb models' power comes from prior knowledge, not virtual cells
anshulkundaje · x · 2026-10-06
In posts 7-8 of his thread critiquing single-cell foundation model (scFM) benchmarks, Anshul Kundaje argues that most of the predictive power of SOTA perturbation models comes from prior-knowledge information — they are not causal transcription regulation models, virtual cells, or world models. He also notes the models that actually perform somewhat well (PRESAGE, Rheister, Arc's new PIE model, GenbioAI's nearest-neighbor models) use no single-cell pretraining on observational atlases at all, undercutting the scFM pretraining narrative.
More from Research
- User Sim Index is broken: trivial bot scores 95% across behavioral dims — ericzelikman · 2026-10-06
- Used OpenAI Dots as a Free Agent Swarm to Break a 47-Year-Old Math Record — jaxchang · 2026-10-06
- Stanford prof says current experiment designs can't train causal virtual cell models — anshulkundaje · 2026-10-06
- TIDES dataset on multi-party and multi-agent collaboration to debut at COLM 2026 — josephseering · 2026-10-06
- LakeQuest QA benchmark, testing RAG on messy enterprise tables and docs, hits COLM — hllo_wrld · 2026-10-06
- Frontier Data Summit lineup: Chollet, Dawn Song headline batch of new agent benchmarks — StanfordAILab · 2026-10-06