Long non-coding RNAs (LncRNAs) are a major type of noncoding RNAs. They play important roles in many biological processes, including gene expression regulation, signal transduction, metabolism pathway, etc. LncRNAs are also involved in the process of various disease developments. Considering the strong associations between lncRNAs and diseases, many lncRNAs have been designated to be biomarkers and therapeutic targets for the treatment of various cancers. In this regard, many computational models have been proposed to identify lncRNA-disease associations by fusing multiple biological information. However, most predictors merge the lncRNA similarity vectors and disease similarity vectors simply to represent the features of lncRNA-disease associations, failing to accurately describe the association features. Furthermore, noisy and irrelevant information would be introduced when combining multiple biological data sources, leading to the high rate of false positives.
In this study, a novel computational predictor called iLncRNAdis-FB is proposed based on the Convolutional Neural Network (CNN) to identify potential lncRNA-disease associations. Compared with other existing methods, the proposed predictor iLncRNAdis-FB has the following advantages: (i) From the perspective of bio-molecular interaction law, six different biological data sources are employed to construct a three-dimensional feature block for each lncRNA-disease association to more accurately describe the lncRNA-disease association features; (ii) Different biological information is fused in a supervised manner considering their contributions to the final prediction to reduce the false positive rates of the prediction results; (iii) The noisy and irrelevant information can be filtered by the feature extraction process of CNN. Experimental results show that iLncRNAdis-FB achieves better performance compared with other state-of-the-art predictors.
The benchmark lncRNA-disease association dataset
The independent lncRNA-disease association dataset