LANCMDA: Predicting MiRNA-Disease Associations via LightGBM with Attributed Network Construction

Since the abnormal expressions of microRNAs (miRNAs) were likely to induce human diseases and traditional biological methods for predicting miRNA-disease associations (MDAs) are costly and time-consuming, it is a great necessity to create computational methods for MDA prediction. In this article, a computational deep learning-based method called predicting miRNA-disease associations through Light Gradient Boosting Machine (LightGBM) with attributed network construction (LANCMDA) is put forward. Specifically, the integrated features obtained from multi-proximity matrices fusion are learnt by sparse autoencoder and then were trained via LightGBM classifier. What’s more, a 5-fold cross-validation is applied to evaluate the performance of this method and the area under curve (AUC) value is 0.9537. It shows that the LANCMDA is very promising. Lung neoplasms is also selected to conduct case study and 95% of the top 20 predicted miRNAs are verified by databases, which shows that LANCMDA is reliable to predict MDAs.