Prediction of potential miRNA-disease associations based on stacked autoencoder

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Molecular Medicine

Background:

• MicroRNAs (miRNAs) are crucial in human complex disease development.
• Discovering miRNA-disease associations aids disease diagnosis and treatment.
• Computational methods offer efficient and cost-effective identification of these associations.

Purpose of the Study:

• To propose a novel computational model, SAEMDA, for predicting potential miRNA-disease associations.
• To leverage unsupervised and supervised learning for enhanced prediction accuracy.
• To validate the model's performance on complex disease datasets.

Main Methods:

• Developed SAEMDA, a Stacked Autoencoder-based model.
• Pre-trained Stacked Autoencoder (SAE) in an unsupervised manner using all samples.
• Fine-tuned SAE using labeled and selected negative samples with a softmax classifier.

Main Results:

• SAEMDA achieved high AUC values (0.9210 global, 0.8343 local LOOCV).
• Achieved an average AUC of 0.9102 ± 0.0029 in 5-fold cross-validation.
• Case studies showed high validation rates: 82% for breast, 100% for lung, and 90% for esophageal neoplasms.

Conclusions:

• SAEMDA demonstrates superior predictive performance compared to existing models.
• The model effectively utilizes unlabeled data for robust predictions.
• SAEMDA is a reliable tool for predicting potential miRNA-disease associations, supporting clinical applications.