A feature pair-based neural network embedded decision tree for synergistic drug combination prediction

Jiayu Zou, Lianlian Wu, Kunhong Liu, Yong Xu, Song He, Xiaochen Bo

Abstract: In the realm of combination therapy, it's a vital step to evaluate the synergistic effects of anti-tumor drug pairs. However, most existing machine learning methods only focus on the attributes of individual drugs, overlooking the implicit relationships of drug pairs, which are essential for understanding their synergistic effects. To address this issue, this paper constructs a novel Neural network Embedded Decision Tree model (NEDT) under a novel paradigm of synergistic drug combination prediction, synonymous feature pairing for drug pairs. It matches synonymous features of drug pairs to construct molecular-level correlations, capturing the implicit relationships of drugs. Our work distinguishes itself from previous neural decision trees by introducing a bi-objective optimization strategy into the fine-tuning process. Experimental results validate that NEDT performs well in predicting synergistic drug combinations. Systematically interpretability analyses demonstrate that NEDT can yield valuable insights into drug synergy, confirming its potential in the biomedical field.