A Simplified Input Strategy for Predicting Multi-Type Associations in miRNA-LncRNA-Disease Network via Stacked Deep Matrix Factorization

Ning Ai, Haoliang Yuan, Zhonghua Lu, Yong Liang, Qi Hong Lai, Loi Lei Lai, Hongmin Cai, Dong Ouyang

Abstract: Understanding miRNA-disease (MDA), lncRNA-disease (LDA), and lncRNA-miRNA (LMI) associations is biologically crucial. Most existing methods predict these separately, relying on similarity networks and external data, which limits generalization. Few models unify all three predictions. Here, we propose SimpleMAP ($\textit{S}$ implified input strategy for $\textit{M}\,ultiple \,\textit{A}\,ssociations \,\textit{P}$ rediction), a novel framework that simultaneously predicts MDAs, LDAs, and LMIs using only known associations, without similarity or auxiliary biological data. This minimal-input design reduces feature contamination and improves applicability. SimpleMAP constructs a three-layer heterogeneous network to model interactions among miRNAs, lncRNAs, and diseases. It employs stacked deep matrix factorization (SDMF) to extract latent features from sparse association data and integrates multiple fusion strategies to enhance representation learning. As one of the first models to jointly predict three bio-entity associations with such simplicity, SimpleMAP achieves state-of-the-art performance across multiple benchmarks. It also demonstrates robustness on two independent datasets involving miRNA-circRNAdisease associations. Case studies with biological evidence confirm its ability to identify potential biomarkers for complex diseases. SimpleMAP establishes a new paradigm in association prediction.enabling multi-task, high-accuracy inference with minimal input complexity. Its scalability and biological interpretability make it a powerful tool for computational biology and disease mechanism studies. The code is available at: https://github.com/AINING96/SimpleMAP.git.

External IDs:dblp:journals/tcbb/AiYLLLLCO25