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

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• High-throughput datasets are abundant, but comprehensive molecular interaction maps are lacking.
• Predicting interactions between heterogeneous molecular types is particularly challenging due to limited experimental data.
• Developing predictive strategies for inter-omics connections is crucial for holistic disease mapping.

Purpose of the Study:

• To introduce a novel nonlinear deep learning method, Data Integration with Deep Learning (DIDL), for predicting inter-omics interactions.
• To assess the applicability and performance of DIDL across diverse biological networks.
• To evaluate the accuracy of DIDL's predictions through literature validation.

Main Methods:

• Developed DIDL, a deep learning model comprising an encoder for automatic feature extraction and a predictor for interaction prediction.
• Applied DIDL to predict interactions in drug-target protein, transcription factor-DNA element, and miRNA-mRNA networks.
• Validated the accuracy of DIDL predictions using literature surveys.

Main Results:

• DIDL demonstrated superior performance compared to state-of-the-art methods in predicting inter-omics interactions.
• Achieved Area Under the Curve (AUC) values exceeding 0.85 and Area Under the Precision-Recall Curve (AUPRC) values over 0.83 across all tested networks.
• Successfully predicted novel interactions, validated through literature review.

Conclusions:

• DIDL offers advantages including automatic feature extraction, end-to-end training, and robustness to network sparsity.
• The algorithm's independence from biochemical features makes it broadly applicable to various biological networks.
• DIDL facilitates the construction of integrative networks, advancing the understanding of complex disease mechanisms.