深度协同过滤算法实现药物-靶标关系预测

Deep collaborative filtering algorithm for drug-target interaction prediction

为提高药物-靶标相互作用(drug-target interaction,DTI)预测效果,提出一种深度协同过滤算法实现DTI预测。在约束非负矩阵分解中融入多输入深度自编码器,通过添加药物、靶标双重正则化,约束矩阵分解中潜在影响因子的学习,缓解新药物、新靶标的冷启动问题。针对DTI矩阵的稀疏问题,设计多输入深度自编码器,实现同时提取DTI矩阵和药物、靶标辅助信息的潜在特征。对4类数据集设计两组实验,实验结果表明,深度协同过滤算法优于其它7种算法。

To improve the effects of drug-target interaction(DTI)prediction,a deep collaborative filtering algorithm was proposed to achieve DTI prediction.A multi-input deep auto-encoder was combined with constrained non-negative matrix factorization.By adding the dual regularization of drugs and targets,the learning of potential influence factors of drugs and targets in the matrix factorization was constrained,and the cold start problem of new drugs and new targets was alleviated.To overcome the sparse problem of DTI matrix,the multi-input deep auto-encoder was designed to extract the potential features of DTI matrix,drug and target auxiliary information.Two sets of experiments were designed for the four types of data sets.The results show that the deep collaborative filtering algorithm outperforms the other seven algorithms.