Arc Institute's PIE model predicts unseen perturbations in virtual cells, 2.7x AUPRC gain, code written by AI agents
surmenok · x · 2026-10-09
Arc Institute introduced PIE (Perturbation is Everything), a virtual cell model tackling prediction of unseen perturbations in unseen cell types. It achieves 2.7x higher AUPRC (0.239 vs 0.088) when both context and perturbation are unseen, and 3.2x improvement on unseen perturbations. Instead of simulating full single-cell transcriptomes, PIE directly predicts differential gene expression using biological knowledge and experimental evidence. Notably, all code was written by AI coding agents with some experiments autonomously orchestrated. Paper on bioRxiv, code on GitHub.
More from Research
- Prompt Tuning Is Forgotten Lore — Are We Massively Underusing Finetuned Tokens? — cephaloform · 2026-10-09
- Ricardo Baeza-Yates Lecture: When Will ML Evaluation Stop Fooling Itself? — PolarBearby · 2026-10-09
- Alzheimer's Translation Challenge Launches With 150M-Cell Atlas and Wet-Lab Testing — xeophon · 2026-10-09
- Swapping harness lifts GPT-5.6 repo migration from 6.5% to 31%, paper finds — omarsar0 · 2026-10-09
- Microsoft's ThinkingBox: Kimi-K3 tops discovery at 93.89% but only 13.41% solve tasks 20/20 — tuhin_k · 2026-10-09
- Sakana AI's Continuous Memory Machine splits short-term compute from long-term storage — dair_ai · 2026-10-09