Temporal Biomedical Knowledge Graph Predicts Clinical Trial Advancement.

A new Temporal Heterogeneous Biomedical Knowledge Graph (THBKG) has been developed to address the high failure rate of therapeutic programs in clinical trials, particularly in Phase II. The core innovation of THBKG is its ability to reconstruct the evidence supporting a target-disease linkage at any given past date, crucial for evaluating decisions made at the time a program entered the clinic. This graph comprises over 110,000 entities and 11.1 million edges across nineteen relation types, with each edge timestamped to reflect when its evidence changed. The THBKG enables a decision-aligned benchmark that predicts whether a target-disease pair entering Phase II will advance to Phase III, based solely on evidence available before that decision. Graph propagation over THBKG significantly outperforms direct-evidence models, achieving a 4.3-4.5 relative success rate for top-ranked pairs. This gain is particularly pronounced for the 72.8% of pairs that lack direct target-disease evidence at their decision point, where the model leverages intervening biological connections. The framework also includes a path-based explainer to decompose predictions, offering transparency into the underlying evidence landscape.