Many Needles in a Haystack: Active Hit Discovery for Perturbation Experiments

High-throughput gene perturbation experiments can test several genetic interventions in parallel, yet experimental budgets remain limited. A central goal is hit discovery: identifying as many perturbations as possible whose phenotypic effect exceeds a predefined threshold. Pure exploration strategies are statistically inefficient, wasting budget on low-value regions. Bayesian optimization methods offer a principled alternative but target a single global optimum, over-exploiting dominant modes while neglecting other high-value regions. We formalize hit discovery as a sequential experimental design problem and propose Probability-of-Hit, an acquisition function that directly targets threshold exceedance by ranking candidates according to their posterior probability of being a hit. We prove asymptotic optimality of this approach and demonstrate strong empirical performance on both synthetic benchmarks and real biological immunology datasets, including upto 6.4\% improvement over baselines on the Schmidt IL-2 dataset.

Identifying which genetic interventions produce meaningful biological effects such as killing cancer cells or blocking a disease pathway requires testing thousands of candidates in the lab. But running each experiment is costly, so scientists can only afford to test a small fraction of all possibilities. Existing computational strategies to choose which genes to test fall into two traps: some cast a wide net, spending budget on regions that yield no hits, while others greedily exploit best-known sites and overlook equally valuable deposits elsewhere. We introduce Probability-of-Hit (PoH), an algorithm that forms a key component of lab-in-the-loop framework. Operating across multiple rounds, each candidate gene is scored by how likely it is to exceed a predefined threshold of biological effect. Rather than hunting for one peak, our approach actively seeks out multiple promising candidates simultaneously. Based on observed effects from lab experiments on the chosen set, the algorithm updating its priorities for the next round. Our results show that PoH recovers substantially more high-effect genes per experiment than standard approaches (up to 6.4% more hits on a real immunology dataset). By making gene-perturbation screens more efficient, this work can lower the cost of functional genomics studies and accelerate the process of drug target discovery.