Reproduce “PerturbCellRL: Aligning Distributions and Grounding Biology via Post-Training Perturbation Generators” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Dongxia Wu, Mingyu Li, Yuhui Zhang, Anurendra Kumar, Emma Lundberg, Serena Yeung-Levy, Emily B. Fox

PerturbCellRL post-trains a cell-response generator with rewards that score generated cells for population-level distribution match and biological plausibility. Starting from pretrained scDFM, it combines an energy witness with expression-support and pathway-discrimination checks. On ComboSciPlex, mean MMD drops from 0.0462 to 0.0366, a 20.8% reduction. The evaluation spans four settings, but rewards require real target cells and manually designed biological features.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.