Systematic modeling predicts synergistic and safe drug combinations for parasitic diseases.

Parasitic diseases impose a substantial global health burden due to the widespread transmission and diversity of protozoa and helminths, which cause numerous infections and regional outbreaks. Despite the availability of various antiparasitic drugs, their clinical utility is often constrained by high cost, toxicity, severe side effects, and the growing threat of drug resistance. Combination therapy, designed to enhance efficacy through synergistic effects while reducing toxicity, represents a promising strategy to improve treatment outcomes for parasitic diseases. In this work, we propose MetaSynMT, a novel multi-task learning framework designed to predict synergistic and safe drug combinations, with a specific focus on parasitic diseases. The model integrates a meta-path aggregation mechanism to capture both structural and high-order semantic features of drugs. Alongside the primary task of synergy prediction, we introduce a secondary task of side effect prediction, enabling the joint identification of combinations with high synergy and low toxicity. Experimental results demonstrate that MetaSynMT outperforms several state-of-the-art baselines on parasitic disease dataset and exhibits strong generalization capability across diverse real-world settings. Furthermore, based on MetaSynMT's predictions, we identified allicin and sodium stibogluconate as a promising combination therapy for echinococcosis. In vitro protoscolex culture experiments showed that the combination achieved a 100% inhibition rate at concentrations of 850 μM allicin and 36.3 μM sodium stibogluconate, significantly surpassing monotherapies. Overall, this work provides a novel computational tool and theoretical foundation for optimizing antiparasitic drug combinations and discovering potential therapeutic strategies.