AutoScreen: AI-driven multi-agent platform for automated, evidence-based target discovery and prioritization in functional genomics

Background

Functional genomics is a rapidly evolving field that leverages high-throughput technologies—such as CRISPR screens, RNA sequencing, and genome-wide association studies (GWAS)—to decipher gene function and identify potential therapeutic targets. As these technologies have matured, they now generate vast and complex datasets, offering unprecedented opportunities for biological discovery and drug development. However, the process of translating raw experimental data into actionable insights, such as prioritizing which genes to target for further study or intervention, remains a significant bottleneck. This challenge is compounded by the need to integrate diverse sources of biological evidence, including scientific literature, structured databases, and experimental results, in a manner that is both comprehensive and context-aware. The growing scale and complexity of functional genomics data have outpaced traditional manual curation and expert-driven review, creating a pressing need for automated, transparent, and evidence-based approaches to target discovery.

Current approaches to target prioritization in functional genomics are limited by several key shortcomings. Manual review processes are labor-intensive, subjective, and often inconsistent, leading to variability in target selection and missed opportunities for discovery. Existing computational methods, including statistical ranking and knowledge-based algorithms, frequently struggle to synthesize evidence from heterogeneous sources or to adapt their recommendations to the specific context of a given experiment. Many AI-driven solutions lack transparency, making it difficult for researchers to understand or trust the rationale behind gene rankings. Furthermore, these methods often fail to incorporate the latest advances in large language models and reinforcement learning, which can enhance evidence integration and align predictions with experimental validation. As a result, researchers face significant barriers in efficiently and reliably identifying high-confidence gene targets, ultimately slowing the pace of functional genomics research and its translation into therapeutic innovation.

Technology

AutoScreen is an AI-powered platform designed to automate and enhance the discovery of gene targets in functional genomics, particularly for high-throughput screening applications such as CRISPR, RNA-seq, and GWAS. The system features a modular, multi-agent architecture that operates in two primary modes: pre-screen library design and post-screen hit re-ranking. Its workflow involves a Deep Research Agent that gathers information from literature, web, and LLMs; an Information Restructure Agent that converts this unstructured data into structured search terms; and a Knowledge Base Search Agent that queries 26 specialized databases in parallel. The Information Synthesis Agent then integrates these diverse evidence streams using LLM-driven feature selection, assigning importance scores and rationales, while reinforcement learning optimizes this process based on experimental validation. Finally, a Target Review Agent applies practical filters to produce transparent, evidence-based gene prioritizations, all accessible through a user-friendly Resource Hub.

What differentiates AutoScreen is its comprehensive, context-aware integration of multi-source evidence and its transparent, explainable decision-making process. Unlike traditional or purely statistical methods, AutoScreen dynamically synthesizes literature, database, and experimental data, leveraging reinforcement learning to continually refine its feature selection in alignment with real-world biological outcomes. Its iterative extraction and review process ensures high recall and precision, while the modular agent framework allows for extensibility and adaptation to new experimental contexts. Benchmarking on nearly 400 CRISPR screens demonstrates AutoScreen’s superior hit recovery and ranking precision, often requiring smaller libraries for the same number of true positives, thus reducing costs and accelerating research. The platform’s Resource Hub further democratizes access, enabling researchers to rapidly generate, review, and act on evidence-backed gene target reports, setting a new standard for scalable, reproducible, and transparent target discovery in genomics.

Benefits

• Automates and accelerates gene target discovery in functional genomics, reducing manual effort and time.
• Integrates multi-source evidence (literature, databases, web) for comprehensive and precise target prioritization.
• Employs reinforcement learning to optimize feature selection based on experimental validation, enhancing prediction accuracy.
• Provides transparent, explainable, and evidence-based target rankings with detailed rationale and data attribution.
• Supports both pre-screen library design and post-screen hit re-ranking for flexible experimental workflows.
• Outperforms existing methods in hit recovery and ranking precision, improving experimental efficiency and reducing costs.
• Includes a Resource Hub offering extensive, searchable target reports to facilitate community access and hypothesis generation.
• Modular multi-agent architecture allows extensibility and adaptation to diverse experimental contexts and data types.

Applications

• Automated CRISPR screen hit prioritization
• AI-driven gene library design
• Evidence-based drug target discovery
• Re-analysis of public genomics datasets
• Interactive target discovery resource hub

Additional info

This AI system automates target discovery in functional genomics, operating in pre-screen library design and post-screen hit re-ranking modes. Its modular, multi-agent architecture gathers, structures, and synthesizes evidence from diverse sources and 26 databases using LLMs and reinforcement learning. It provides transparent, evidence-based target prioritization, outperforming current methods.