TESSERA: LLM-Guided MCTS for Drug-Disease Explanations

Researchers propose TESSERA, a neuro-symbolic framework combining large language models (LLMs) with Monte Carlo Tree Search (MCTS) over knowledge graphs to generate mechanistic explanations for drug-disease pairs. The approach addresses combinatorial challenges in extracting multi-step paths by using LLMs for local discriminative judgment rather than autonomous generation, while the knowledge graph enforces structural constraints. MCTS coordinates long-horizon search with credit assignment via backpropagation. LLMs serve as a prior policy for exploration and a comparative state evaluator for reward signals. The work is described in arXiv paper 2605.09542.