CT-Perturb: LLMs Reach 82% of Full-Data Cell Perturbation Prediction with 10% of Measurements
david_van_dijk · x · 2026-11-03
CellType reports that CT-Perturb, a language-model approach for predicting cell responses to perturbations, achieves 82% of its full-data adaptation score on two new cell lines using just 10% of measurements (69% at 1%, 91% at 50%), far above an additive baseline (0.26). The result suggests LLMs can extrapolate perturbation responses to unseen cell contexts from sparse measurements, cutting functional genomics experiment costs.
More from Research
- Intology's Locus automates molecular design, beating expert solutions and LLM agents — ChengleiSi · 2026-11-03
- Imperial College team raises $40M for Instant Policy, a graph-diffusion imitation learning robot startup — philfung · 2026-11-03
- The '7x accuracy' weight-editing claim was mostly just turning off thinking mode — ChengleiSi · 2026-11-03
- Chetty and Opportunity Insights: digital twins can recover causal effects, validated by RCTs — soumitrashukla9 · 2026-11-03
- AI models have learned to update other models' weights directly, says Brynjolfsson — soumitrashukla9 · 2026-11-03
- Chetty's team shows digital twins recover causal effects, validated against RCTs — danielrock · 2026-11-03