Drug-Disease Association Prediction Through Multiple Integrated Similarities and Deep Learning

The new drugs' development is a time-consuming, laborious, and costly process. Consequently, the need for computational methods to predict drug-disease associations has become increasingly evident. This paper presents a new approach for predicting drug-disease associations by using multiple integrated similarities and deep learning techniques. Specifically, we combine Disease semantic similarity with disease Gaussian interaction profile kernel similarity and Drug chemical structure similarity with drug Gaussian interaction profile kernel, to construct integrated similarity measures. Then we used a Weighted K-nearest known neighbors (WKNKN) algorithm to solve the sparsity data problem. Subsequently, we utilized a deep learning technique to predict drug-disease associations. Our method demonstrates remarkable performance, achieving the highest AUC and AUPR values of 0.9538 and 0.6607, respectively. When compared to other relevant methods using the same dataset, our approach outperforms them in both AUC and AUPR metrics. Thus, our method proves to be a valuable tool for predicting associations between drugs and diseases.