piRNA-disease association prediction based on multi-channel graph variational autoencoder.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Piwi-interacting RNAs (piRNAs) are non-coding small RNAs abundant in mammalian testes.
• piRNAs are linked to various human diseases, but experimental validation of these associations is resource-intensive.
• Developing efficient computational methods for piRNA-disease association prediction is crucial.

Purpose of the Study:

• To propose a novel computational method for predicting piRNA-disease associations.
• To leverage a multi-channel graph variational autoencoder (MC-GVAE) for enhanced prediction accuracy.
• To integrate multiple similarity networks for a comprehensive analysis of piRNA-disease relationships.

Main Methods:

• Utilized a multi-channel graph variational autoencoder (MC-GVAE) framework.
• Integrated four similarity networks: piRNA sequences, disease semantics, piRNA GIP kernel, and disease GIP kernel.
• Employed a three-layer neural network classifier for final association prediction.

Main Results:

• The MC-GVAE method achieved state-of-the-art performance on a benchmark dataset.
• Achieved an average Area Under the Curve (AUC) of 0.9310 and Area Under the Precision-Recall Curve (AUPR) of 0.9247.
• Demonstrated superior effectiveness in predicting piRNA-disease associations compared to existing methods.

Conclusions:

• The proposed MC-GVAE method is highly effective and accurate for predicting piRNA-disease associations.
• This computational approach offers a valuable tool for understanding the role of piRNAs in human diseases.
• The study highlights the potential of integrating diverse biological data for complex association predictions.