Bridge-Aware Reinforced Compositional Exploration and Decomposition for Biomedical Multi-Hop Answering

Biomedical question answering requires multi-hop reasoning over implicit relationships—e.g., identifying gene-associated diseases to retrieve target drugs. However, closed-book frontier language models achieve only ~14.57% precision on two-hop queries under single-answer prompting and answers drawn from parametric knowledge lack verifiable evidence, limiting their clinical utility. In this paper, we propose Bridge-Aware Reinforced Compositional Exploration and Decomposition (BRACED). A planner trained with Group-Relative Policy Optimization emits sub-queries, and a deterministic type-aware two-hop traversal over PrimeKG converts them into an answer set, so every returned entity is traceable to a graph path. On the BioHopR benchmark, BRACED reaches micro-F1 0.4241 and macro-F1 0.6381, improving over supervised fine-tuning of the same planner by 30.2%. With retrieval and the cap held fixed, five backbones of 1.7B–8B finish within 0.0034 of one another and BRACED reaches its result with one language-model call per question, from a 7B open-weight model that runs on the premises. Two typed hops already reach 95.1% of gold answers on this PrimeKG-derived benchmark: performance is bounded by candidate selection under the cap, not by decomposition quality.