RxRx3-core: Benchmarking drug-target interactions in High-Content Microscopy

2025-03-26

AI Summary

High-content screening (HCS) has long suffered from a scarcity of high-quality, publicly available datasets and the absence of standardized, representation-learning–oriented benchmarks—hindering progress in zero-shot drug–target interaction (DTI) inference. To address this, we introduce RxRx3-core, the first lightweight, high-fidelity, representation-learning–focused HCS benchmark dataset (18 GB, 222,601 images), capturing multi-condition cellular responses to 736 CRISPR knockouts and 1,674 compounds across eight concentrations. This work establishes the first unified integration of chemical perturbations, genetic editing, and high-content imaging. We provide pre-trained embeddings, a standardized evaluation protocol, and fully open-sourced code. RxRx3-core is publicly released on Hugging Face and Polaris, substantially lowering the barrier to HCS modeling and enabling reproducible, cross-institutional representation learning evaluation and biological discovery.

Technology Category

• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Machine Learning: Representation Learning
• Computer Vision: Representation Learning for Vision

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

High Content Screening (HCS) microscopy datasets have transformed the ability to profile cellular responses to genetic and chemical perturbations, enabling cell-based inference of drug-target interactions (DTI). However, the adoption of representation learning methods for HCS data has been hindered by the lack of accessible datasets and robust benchmarks. To address this gap, we present RxRx3-core, a curated and compressed subset of the RxRx3 dataset, and an associated DTI benchmarking task. At just 18GB, RxRx3-core significantly reduces the size barrier associated with large-scale HCS datasets while preserving critical data necessary for benchmarking representation learning models against a zero-shot DTI prediction task. RxRx3-core includes 222,601 microscopy images spanning 736 CRISPR knockouts and 1,674 compounds at 8 concentrations. RxRx3-core is available on HuggingFace and Polaris, along with pre-trained embeddings and benchmarking code, ensuring accessibility for the research community. By providing a compact dataset and robust benchmarks, we aim to accelerate innovation in representation learning methods for HCS data and support the discovery of novel biological insights.

Problem

• Lack of accessible datasets for HCS representation learning
• Need for robust benchmarks in drug-target interaction prediction
• High storage demands hinder large-scale HCS data utilization

Innovation

• Curated subset RxRx3-core for DTI benchmarking
• Compressed 18GB dataset with 222,601 images
• Includes pre-trained embeddings and benchmarking code