MoAgent: A Hypothesis-Driven Multi-Agent Framework for Drug Mechanism of Action Discovery

Jun Hyeong Kim, Seokhyun Moon, Seonghwan Kim, Junhyeok Jeon, Taein Kim, Jisu Seo, Songmi Kim, Woo Youn Kim

Keywords: Multi-agent system, Virtual Cell, Mechanism of Action, Hypothesis testing

TL;DR: We introduce MoAgent, a multi-agent framework that reframes mechanism of action (MoA) inference as a hypothesis-driven scientific discovery process.

Abstract:

Determining the mechanism of action (MoA) of novel chemical compounds is a critical yet challenging task in drug discovery.
We introduce MoAgent, a multi-agent framework that reframes MoA inference as a hypothesis-driven scientific discovery process.
MoAgent integrates multi-modal data from chemical structure, gene expression, and biological pathways, deploying a committee of specialized agents to collaboratively generate and validate mechanistic hypotheses.
The framework operates through an iterative cycle of evidence triangulation and hypothesis validation, where a bioinformatician agent assesses causal plausibility using a knowledge graph and a medicinal chemist agent verifies direct target engagement.
Our experiments demonstrate that the integrated, hypothesis-driven strategy significantly enhances the accuracy and reliability of MoA inference, and maintains robust performance even in zero-shot scenarios.
By emulating scientific reasoning, MoAgent offers a more effective paradigm for accelerating drug discovery.

Submission Number: 29