Back

BioScientist Agent: Designing LLM-Biomedical Agents with KG-Augmented RL Reasoning Modules for Drug Repurposing and Mechanistic of Action Elucidation

Zhang, F.; Zhao, Y.; Zhang, W.; Lai, L.

2025-08-12 bioinformatics
10.1101/2025.08.08.669291 bioRxiv
Show abstract

Drug discovery is protracted, resource-intensive, and afflicted by attrition rates exceeding 90 %, which leaves most diseases, particularly rare or neglected indications, without effective therapies. Drug repurposing offers a cost effective alternative, yet systematic identification of novel drug indication pairs and mechanistic rationales remains hindered by the scale and heterogeneity of biomedical knowledge. We present BioScientist Agent, an end to end framework that unifies a billion-fact biomedical knowledge graph with (i) a variational graph auto-encoder for representation learning and link prediction driven drug repurposing, (ii) a reinforcement learning module that traverses the graph to recover biologically plausible mechanistic paths, and (iii) A large language model (LLM) multi-agent layer that orchestrates these components, enabling inference of target pathways for a drug disease pair, and automatic generation of coherent causal reports. In all downstream tasks, the BioScientist Agent surpasses existing state of the art baseline models across various metrics and provides mechanistic explanations consistent with the literature. Its open and modular design accelerates hypothesis generation and reduces experimental overhead in early stage discovery.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.