Reproduce “CellCLIP -- Learning Perturbation Effects in Cell Painting via Text-Guided Contrastive Learning” 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

Mingyu Lu, Ethan Weinberger, Chanwoo Kim, Su-In Lee

CellCLIP, from researchers at the University of Washington, introduces a text-guided contrastive learning framework for Cell Painting data, creating a unified latent space for diverse cellular perturbations. This approach significantly outperforms existing baselines in cross-modal retrieval and enhances biologically meaningful intra-modal tasks, including zero-shot gene-gene relationship recovery.

Key insights

  • Natural language text prompts, incorporating specific metadata like SMILES strings, provide a highly effective and generalizable way to represent diverse biological perturbations.
  • Efficiently combining pre-trained natural image encoders with a novel, lightweight Transformer architecture (CrossChannelFormer) for cross-channel reasoning significantly improves performance and computational efficiency in Cell Painting analysis.
  • Aggregating individual cell profiles into a single, attention-pooled representation per perturbation effectively addresses the many-to-one data structure, speeding up training by 6.7 times and improving retrieval.

Summary by alphaXiv.

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