Machine learning for drug repositioning: Recent advances and challenges

Because translating the growing body of knowledge about human disease into treatments has been slower than expected, the application of machine learning techniques to drug repositioning has become attractive. An effective and comprehensive understanding of the current state of drug repositioning can help researchers to investigate more efficient and accurate algorithms. In this study, we first present the theoretical rationale for drug repositioning analysis. Then, we conduct a comprehensive review on machine learning algorithms for drug discovery, which include (1) traditional machine learning-based models using linear and logistic regression, support vector machines, random forest, and decision tree, (2) network transmission-based models using drug–disease similarity and network similarity-based reasoning, (3) matrix completion and matrix factorization-based methods using matrix completion, logistic matrix factorization, collaborative matrix factorization, and regularized matrix factorization, and (4) deep learning-based methods using deep neural networks, convolutional neural networks, recurrent neural networks, and graph convolutional networks. This is followed by a review of commonly used data sources for drug repositioning, as well as an introduction to particular data sources that can be employed by researchers. To conclude, we discuss the future developments and challenges of drug repositioning methods.