To address the challenge of discovering target pathways in biological knowledge bases, this paper proposes ExPAth, a novel subgraph inference framework that explicitly integrates experimental data. ExPAth classifies diverse graphs (biological networks) within biological databases and considers links (representing pathways) that contribute to the classification as target pathways. This framework seamlessly integrates biologically based models to encode experimental molecular data, and it presents machine learning-oriented biological evaluations and novel metrics. Experimental results, including evaluations of 301 biological networks, demonstrate that pathways inferred by ExPAth are biologically meaningful, achieving up to 4.5x higher Fidelity+ (necessity) and 14x lower Fidelity- (sufficiency) than existing methods, while preserving up to 4x longer signaling chains.
