Predicting of Drug-Disease Associations via Sparse Auto-Encoder-Based Rotation Forest

Computational drug repositioning, designed to identify new indications for existing drugs, significantly reducing the cost and time involved in drug development. Confirming the association between drugs and disease is a critical process in drug development. At present, the most advanced method is to apply the recommendation system or matrix decomposition method to predict the similarity between drugs and diseases. In this paper, the association between drugs and diseases is integrated and a novel computational method based on sparse auto-encoder combined with rotation forest (SAEROF) is proposed to predict new drug indications. First, we constructed a drug-disease similarity based on drug-disease association and integrated it into sparse auto-encoder to obtain the final drug-disease similarity. Then, we adopt a rotation forest algorithm to predicted scores for unknown drug–disease pairs. Cross validation and independent test results show that this model is better than the existing model and has reliable prediction performance. In addition, the case study of two diseases further proves the practical value of this method, and the results obtained can be found in CTD database.