Tag
This paper analyzes the Virtual Cell Challenge benchmark for held-out CRISPRi perturbation prediction, finding that simple magnitude-based scalar features outperform deep MLP encoders, and that magnitude-only predictors transfer better across cell types.
PerturbCellRL introduces a reinforcement learning framework that post-trains a pretrained single-cell transcriptomic generator using cell-level verifiers as rewards, improving biological consistency of perturbation predictions beyond distributional matching.