Using improved Restricted Boltzmann Machines for drug-disease Prediction
Jun Ma, Yongna Yuan, Mingchao Guo · 2018
Despite increased investment in pharmaceutical research and development, fewer and fewer new drugs are entering the marketplace. This has prompted studies in repurposing existing drugs for use against diseases. In this paper, firstly a common Restricted Boltzmann Machines model is used to predict the drug repositioning task in drug-disease association network, the model could receive more reliable prediction performance than PREDICT, Probs and Heats methods, the AUC value is 0.9528. Secondly the model is improved by adding a momentum when updating weights. The improved Restricted Boltzmann Machines model is used on the same drugs dataset for predicting task. In the same training time the AUC values is 0.9744, it has enhance obviously. Finally, some of the prediction results in the second model are proved to be effective by searching the relevant information. Therefore, the improved Restricted Boltzmann Machines model would be suitable for the drug-disease data and provide new powerful tool for drug repositioning in the future.