MLMVFE: A Machine Learning Approach Based on Muli-view Features Extraction for Drug-Disease Associations Prediction

Determining the associations between drugs and diseases plays an important role in the drugs development processes. However, current drug-disease associations (DDAs) prediction methods are too homogeneous for features extraction, so a machine learning approach based on multi-view features extraction (MLMVFE) is proposed for DDAs prediction. Firstly, proteins are introduced to form a new heterogeneous network, which enriches the associations information. Then, nodes features are extracted from two perspectives: network topology and biological knowledge. Finally, the Light Gradient Boosting Machine classifier is utilized to predict DDAs. The MLMVFE achieves satisfactory results on both B-dataset and F-dataset through 10-fold cross-validation. In addition, to further demonstrate the reliability of the MLMVFE, case study is done where clozapine is used as a case. The result suggests that the MLMVFE has the potential to tap into novel DDAs.