Abstract
Exploring latent microRNA (miRNA)-disease associations (MDAs) is vital for early screening and treatment. Compared with traditional experiments, computational methods enhance efficiency and lower costs in predicting MDAs. We trained the Accurate Matrix Completion for predicting potential MiRNA-Disease Associations (AMCMDA) model in this work, utilizing truncated nuclear norm minimization to improve the prediction accuracy. In AMCMDA, we begin by constructing a heterogeneous network incorporating both similarity and association information between miRNAs and diseases. Second, an optimization framework is designed to complete the effective approximation of the truncated nuclear norm to complement the missing values of the objective matrix. Finally, we solve this optimization problem via Alternating Direction Method of Multipliers and obtain the final prediction scores. After comparing the AMCMDA model with other models across three validation frameworks and three different datasets, we find that the AMCMDA model demonstrates robust and accurate performance. The model's excellent performance is also demonstrated by two categories of case studies on three diseases.