High Potential Negative Sampling for Drug Disease Association Prediction

As we are aware, drug development incurs substantial costs and time investments. Thus, there is a need for novel methods to efficiently support this process, such as drug repositioning. This field is also gaining significant attention from researchers, particularly with the aid of intelligent computational methods applied in machine learning. These methods open new opportunities in this domain and have yielded promising outcomes. However, a major drawback of learning lies in the necessity for constructing a well-supported dataset of positive and negative samples for prediction. Current methods largely focus on generating randomly sampled negative instances. Clearly, such randomly sampled negative instances cannot ensure the quality of negative dataset, and they undoubtedly impact the prediction performance of classifiers. The article proposes a method of selecting highly reliable negative samples for performance improvement in prediction of drug-disease association. Experimental results demonstrate that the proposed method outperforms current research, providing an appealing alternative solution for predicting drug-disease association.