SGANRDA: semi-supervised generative adversarial networks for predicting circRNA-disease associations

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNAs (circRNAs) are increasingly recognized for their roles in complex human diseases.
• Traditional experimental methods for identifying circRNA-disease associations are costly and time-consuming.
• Existing computational models have limitations in effectively predicting these associations.

Purpose of the Study:

• To develop an advanced computational model for predicting circRNA-disease associations.
• To overcome limitations of current models by integrating diverse biological data.
• To improve the efficiency and accuracy of identifying potential circRNA-disease links.

Main Methods:

• Proposed a semi-supervised generative adversarial network (GAN) model named SGANRDA.
• Fused natural language processing features of circRNA sequences with disease semantics, and Gaussian interaction profile kernel features.
• Employed a two-stage training process: pre-training the GAN with all circRNA-disease pairs, followed by fine-tuning with labeled samples.
• Utilized an extreme learning machine classifier for final prediction.

Main Results:

• Achieved high performance with AUC scores of 0.9411 (leave-one-out cross-validation) and 0.9223 (5-fold cross-validation).
• SGANRDA outperformed existing models on benchmark datasets.
• Case studies showed that 25 out of 30 top predicted circRNA-disease pairs were validated by recent literature.

Conclusions:

• SGANRDA is an effective computational tool for predicting circRNA-disease associations.
• The model's innovative approach, incorporating circRNA sequences and utilizing all available data, enhances prediction accuracy.
• SGANRDA provides reliable candidates for further biological investigation, aiding in disease diagnosis and prognosis.