A computational framework oriented to molecular mechanism analysis was constructed to explore the potential chain of action of kalendolide E in improving Parkinson’s disease by regulating mitochondrial function. Firstly, based on heterogeneous feature coding and similarity constraint aggregation, mitochondrial dysfunction patterns were identified from 1248 samples, 312 candidate molecules and 8640 high-confidence interaction edges. Combined with drug response, topological proximity, semantic consistency, hierarchical coverage and direction matching, the association features between kalendolide E and mitochondrial regulatory pathways were extracted. Then, a relationship prediction model combining heterogeneous graph propagation, ternary relationship score and ranking loss is constructed to jointly rank the drug-target-pathway-phenotype chain of action. Experimental results show that the AUC of the model reaches 0.948, F1 reaches 0.921, and Hit@20 reaches 85.0%. The results showed that kalendolide E may reduce dopaminergic neuron damage mainly through AMPK-SIRT3 signaling, OPA1-mediated fusion regulation, ATP maintenance and reactive oxygen species control, which provides a computational analysis path for the mechanism of natural active ingredients in Parkinson’s disease.
