Advancing morphological profiling with self-supervised learning through comparative insights into representation learning of cellular perturbations.

DINO and DINOv2 SSL foundation models pretrained on JUMP-CP images outperform CellProfiler in compound-replicate retrieval on JUMP-Target2, achieving P ≤ 0.001 for 302 annotated compounds.

• Why it matters: Classical morphological profiling relies on segmentation and handcrafted features, limiting scalability and transfer; SSL aims to learn label-free image representations for phenotypic profiling.
• What they did: The study evaluated contrastive (DINO), masked generative (MAE, iTPN), and hybrid (DINOv2) models pretrained on 119,876 compounds (~1.77 million images) from JUMP-CP across 7 laboratories, then benchmarked retrieval and mechanism-of-action tasks.
• The result: On EU-OPENSCREEN Bioactives (2,464 compounds), DINO/DINOv2 improved site-averaged mean average precision for mechanism-of-action retrieval by 31–37% versus CellProfiler and consistently beat masked-based models across benchmarks.

AI summary of the abstract · may contain errors · not medical advice · read the paper.