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

2026-05-11

AI Summary

This study addresses the challenge of efficiently identifying multiple gene perturbations whose phenotypic effects exceed a predefined threshold—referred to as “hits”—under limited experimental budgets. Framing the problem as a sequential experimental design task, the work introduces Probability-of-Hit, the first acquisition function specifically tailored for multi-hit discovery. This function ranks candidate perturbations by their posterior probability of being hits, directly optimizing for threshold exceedance. Built within a Bayesian optimization framework, the approach combines Gaussian process modeling with the proposed acquisition function to actively explore high-value regions of the search space and is proven to be asymptotically optimal. Empirical evaluations on both synthetic benchmarks and real immunological datasets, including the Schmidt IL-2 dataset, demonstrate up to a 6.4% improvement in hit discovery efficiency over established baselines.

Abstract

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 up to 6.4% improvement over baselines on the Schmidt IL-2 dataset.

Problem

• hit discovery
• gene perturbation
• sequential experimental design
• threshold exceedance
• high-throughput screening

Innovation

• Probability-of-Hit
• hit discovery
• Bayesian optimization
• sequential experimental design
• gene perturbation