BertNDA: a Model Based on Graph-Bert and Multi-scale Information Fusion for ncRNA-disease Association Prediction

Abstract

Non-coding RNAs (ncRNAs) are a class of RNA molecules that lack the ability to encode proteins in human cells, yet play crucial roles in various biological process. Understanding these relationships and how different ncRNAs interact with each other to affect diseases can vastly contribute to their diagnosis, prevention, and treatment. However, predicting tertiary interactions between ncRNA-disease associations by utilizing structural information across multiple scales remains a challenging task. It should be noted that research on predicting tertiary interaction between trinary ncRNA-disease associations is scarce, highlighting the need for further studies in this area. In this work, we propose a predictive framework, called BertNDA, which aims to predict association between miRNA, lncRNA and disease. The framework employs Laplace transform of graph structure and WL (Weisfeiler-Lehman) absolute role coding to extract global information. Local information is identified by the connectionless subgraph which aggregates neighbor feature. Moreover, an EMLP (Element-wise MLP) structure is designed to fuse the multi-scale feature representation of nodes. Furtherly, feature representation is encoded by using a Transformer-encoder structure, the prediction-layer outputs the final correlation between miRNA, lncRNA and diseases. The 5-fold cross-validation result furtherly demonstrates that BertNDA outperforms the state-of-the-art method in predicting assignment. Furthermore, an online prediction platform that embeds our prediction model is designed for users to experience. Overall, our model provides an efficient, accurate, and comprehensive tool for predicting ncRNA-disease associations. The code of our method is available in: https://github.com/zhiweining/BertNDA-main.