ScreenShot: A Foundation Model for Few-Shot Combination Drug Screening
ScreenShot predicts combination-therapy response from a few observed patient measurements, without molecular profiling or fine-tuning.
The model is a hierarchical transformer pretrained on 40 drug screening datasets covering 3,700 drugs and 6,000 biological samples. Given a few-shot context from a new patient, it uses in-context learning on functional measurements to predict combination drug response. On four held-out datasets, the authors say it beat baselines on accuracy and on finding selectively effective treatments. They also used its representations for active learning, matching uniform screening hit detection with one-third of the budget. ArXiv · AI/CL/LG's note
The model is a hierarchical transformer pretrained on 40 drug screening datasets covering 3,700 drugs and 6,000 biological samples. Given a few-shot context from a new patient, it uses in-context learning on functional measurements to predict combination drug response. On four held-out datasets, the authors say it beat baselines on accuracy and on finding selectively effective treatments. They also used its representations for active learning, matching uniform screening hit detection with one-third of the budget. ArXiv · AI/CL/LG's note
score 5