DCSGMDA: A dual-channel convolutional model based on stacked deep learning collaborative gradient decomposition for...

Area of Science:

• Biomedical Informatics
• Genomics
• Computational Biology

Background:

• MicroRNAs (miRNAs) are crucial biomarkers in human diseases, with abnormal expression linked to disease onset and progression.
• Understanding miRNA-disease relationships is vital for elucidating pathogenesis and advancing therapeutic strategies.
• Numerous miRNA-disease associations remain undiscovered, hindering comprehensive research.

Purpose of the Study:

• To develop a novel computational model for predicting potential miRNA-disease associations.
• To leverage deep learning techniques for enhanced feature extraction and correlation prediction.
• To identify novel miRNA-disease relationships for further biological investigation.

Main Methods:

• Constructed miRNA and disease similarity networks, and an association network.
• Employed stacked deep learning and gradient decomposition for feature mining.
• Utilized dual-channel convolutional neural networks and a multilayer perceptron for predicting associations.
• Validated the model using 5-fold and 10-fold cross-validation on HMDD datasets.

Main Results:

• The proposed dual-channel convolutional model (DCSGMDA) demonstrated strong performance in predicting miRNA-disease associations.
• Cross-validation experiments confirmed the model's efficacy across different datasets.
• Ablation, parametric, and comparative studies further validated the model's robustness and superiority.

Conclusions:

• DCSGMDA effectively predicts miRNA-disease associations, offering a valuable tool for biomedical research.
• The model aids in uncovering novel relationships, contributing to a deeper understanding of disease mechanisms.
• This approach has the potential to accelerate drug discovery and personalized medicine.