A deep learning approach to predict inter-omics interactions in multi-layer networks

Abstract Despite enormous achievements in production of high throughput datasets, constructing comprehensive maps of interactions remains a major challenge. The lack of sufficient experimental evidence on interactions is more significant for heterogeneous molecular types. Hence, developing strategies to predict inter-omics connections is essential to construct holistic maps of disease. Here, Data Integration with Deep Learning (DIDL), a novel nonlinear deep learning method is proposed to predict inter-omics interactions. It consists of an encoder that automatically extracts features for biomolecules according to existing interactions, and a decoder that predicts novel interactions. The applicability of DIDL is assessed with different networks namely drug-target protein, transcription factor-DNA element, and miRNA-mRNA. Also, the validity of novel predictions is assessed by literature surveys. Furthermore, DIDL outperformed state-of-the-art methods. Area under the curve, and area under the precision-recall curve for all three networks were more than 0.85 and 0.83, respectively. DIDL has several advantages like automatic feature extraction from raw data, end-to-end training, and robustness to sparsity. In addition, tensor decomposition structure, predictions solely based on existing interactions and biochemical data independence makes DIDL applicable for a variety of biological networks. DIDL paves the way to understand the underlying mechanisms of complex disorders through constructing integrative networks.