基于核方法的协同药物组合预测方法

Kernel-based prediction of a synergistic drug combination

肿瘤因其异质性和复杂的代谢途径,会对单一药物产生耐药性.具有协同抑制效应的药物组合策略,是解决上述问题的有效途径之一.然而,筛选有效药物组合往往需要通过一系列药理学、分子生物学实验,耗时且费用高昂.生物信息学方法通过对已知协同用药的实验数据进行建模分析,可以实现有效药物组合的高通量筛选.本文提出一种基于相似性特征的预测药物组合和细胞系(drug-drug-cell line,DDC)关系的新方法,用于筛选出特异于细胞系的协同或拮抗的药物组合.具体地,首先使用S-kernel和高斯核分别计算药物组合和细胞系基于相似性的特征向量,然后拼接两向量得到药物组合-细胞系的特征向量,以此作为机器学习模型的输入特征.基于药物协同实验的DDC关系作为机器学习的输出.三种机器学习模型,包括深度神经网络(deep neural network,DNN)、随机森林(random forest,RF)和支持向量机(support vector machine,SVM),交叉验证结果表明,新方法稳定可行,且深度神经网络和随机森林分类准确率高达89%~91%.重要的是,基于新方法的预测模型能够预测包含未知药物分子或细胞系的全新DDC组合.本文提出的特征计算方法能够使机器学习模型准确预测药物组合同细胞系之间的关系,为药物组合协同预测提供了一种新方法.

Complex diseases,such as cancer,often exhibit resistance to single drug therapy because of their heterogeneity and complex metabolic pathways.Combination therapy is an efficient strategy to overcome drug resistance.As experimental screenings consume considerable resources and have low efficacy,the computational method is a good alternative.Thus,this article proposes a new method for computing features of a drug-drug-cell line(DDC)combination based on similarity,where the S-kernel and Gaussiankernel methods are used to calculate the drug-drug combination similarity and cell line similarity,respectively.The final feature vector for machine learning input was obtained by concatenating these two vectors.The output for machine learning was based on the experimental results of the synergistic drug combination.Cross validation was performed on three machine learning algorithms,including the random forest,support vector machine,and deep neural network models.The results suggested that the novel method had a robust performance with an area under the curve value of 0.89–0.91.Importantly,the model predicted the novel DDC combinations with new drugs or new cell lines based on unique input features.In conclusion,this novel method improved predictive performance and provided a new strategy for predicting synergistic drug combinations.