DINO and DINOv2 outperform CellProfiler in cell imaging analysis
Researchers published a large-scale evaluation in Nature Machine Learning comparing self-supervised learning (SSL) foundation models for morphological profiling. The study tested contrastive (DINO), masked-based generative (MAE, iTPN), and hybrid (DINOv2) models pretrained on the JUMP–CP dataset, which contains 119,876 compounds and approximately 1.77 million images from 7 laboratories.
Key points
- DINO and DINOv2 outperformed CellProfiler in compound retrieval on JUMP–Target2 with P ≤ 0.001.
- Models achieved 31–37% mean average precision gains on EU-OPENSCREEN Bioactives dataset.
- Evaluation used 119,876 compounds and 1.77 million images from the JUMP–CP dataset.
The results show that top-performing SSL models, specifically DINO and DINOv2, outperformed the classical CellProfiler tool in compound-replicate retrieval on the held-out JUMP–Target2 set with statistical significance (P ≤ 0.001). On the EU-OPENSCREEN Bioactives dataset, these models achieved 31–37% gains in mean average precision for mechanism-of-action retrieval compared to CellProfiler. The study highlights that contrastive and hybrid approaches generalize better across diverse cell types and perturbations than masked-based models.
This work provides a systematic benchmark for selecting SSL approaches in high-content imaging. By bypassing the need for cell segmentation and handcrafted feature engineering, these pretrained foundation models offer a scalable path to accelerate drug discovery. The authors note that while postprocessing steps like Harmony improved batch effect correction, aligning data across different laboratories remains a challenge.
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- TinyLeague updates benchmarks: drops AA Index, adds KV efficiency metrics · 1 src
- Artificial Analysis releases Intelligence Index v5 with private datasets · 1 src
- DeepSeek’s R1 model with guideline knowledge base improves junior doctors’ arthroplasty decisions · 1 src
- Google DeepMind releases Genie 4 world model with native audio and Gemini 3.8 Flash guidance · 12 src
- Researchers propose synthetic document finetuning to detect alignment-faking LLMs · 3 src
Comments
via GitHub Discussions