GeneDisco: A Benchmark for Experimental Design in Drug Discovery
What this work does
Evaluates batch acquisition policies and counterfactual outcome estimators over curated genetic-intervention screens.
Connection to discovery
A reproducible finite-pool baseline for sample-efficient experiment selection using biological measurements.
Evidence
The body specifies a finite collection of measured outcomes, treatment embeddings and alternative acquisition functions, with predictive error as a benchmark objective.
Scope and limitations
The benchmark reveals already recorded outcomes. Improving predictive MSE or selecting hits is narrower than recovering a mechanistic causal explanation or executing new wet-lab experiments.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.