Reproduce “Attribution-based diagnostics for single-cell foundation models” 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
Farnoush Rezaei Jafari, Tristan Lazard, Kenza Bouzid, Shruthi Bannur, Jeremiah Wander, Ashley Mae Conard, Stephanie Hyland, Lorin Crawford and 1 more
The authors use LRP to trace predictions from pretrained single-cell foundation models (scFMs) back to input genes, then compare signed relevance rankings with biological references. Across tested cell-typing tasks, these rankings recover known markers better than expression and random rankings. In a synthetic perturbation system, gene-LRP modestly recovers direct edges beyond degree-preserving controls (p = 0.001), but not response signs or total effects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.