Phenotypic similarity of adverse drug reactions and disease phenotypes is a bridge to mechanistic discovery.

Adverse drug reactions (ADRs) remain a major barrier to safe therapeutic developments. A key challenge is our limited understanding of their underlying mechanisms. In this study, we investigated whether ADRs and diseases phenotypes (DPs) with similar clinical manifestations share mechanistic similarities. To this end, we constructed a comprehensive knowledge graph and applied a graph representation learning to quantify mechanistic similarities between phenotypically similar ADRs and DPs. Our analysis reveals substantial mechanistic overlap among ADRs and DPs within specific system organ classes, including cardiac, psychiatric, and metabolic disorders. These findings suggest that drugs interacting with proteins linked with specific DPs are more likely to cause ADRs with similar phenotypes. By integrating drug-induced and disease-related phenotypes, our approach offers new insights into ADR mechanisms and supports the prioritization of drugs with lower ADR risk. This work contributes to advancing safer and more targeted therapeutic development by bridging phenotypic similarity and molecular mechanisms.