Self-supervised learning on graphs predicts non-coding RNA and disease associations.

Area of Science:

• Oncology
• Bioinformatics
• Genomics

Background:

• Non-coding RNAs (ncRNAs) are critical regulators in cancer development and progression.
• Accurate identification of ncRNA-disease associations (RDAs) is vital for developing targeted cancer therapies.
• Existing graph convolutional network methods for RDA prediction face challenges with data noise and poor generalization.

Purpose of the Study:

• To develop a robust and generalizable framework for predicting ncRNA-disease associations (RDAs).
• To enhance the accuracy of identifying potential ncRNA-disease relationships for improved therapeutic interventions.

Main Methods:

• Proposed SSLGRDA, a novel scheme integrating graph self-supervised learning and machine learning.
• Constructed heterogeneous and homogeneous graphs incorporating known RDAs and node similarities.
• Employed multiple contrastive or generative strategies for graph self-supervised learning to extract robust ncRNA and disease embeddings.
• Utilized machine learning for predicting latent RDA probabilities.

Main Results:

• SSLGRDA demonstrated strong generalization capabilities across nine ncRNA-disease datasets.
• The proposed method outperformed several state-of-the-art RDA prediction techniques.
• Case studies confirmed SSLGRDA's effectiveness in discovering novel ncRNA-disease associations.

Conclusions:

• SSLGRDA offers a powerful and generalizable approach for predicting ncRNA-disease associations.
• The method holds significant potential for advancing precision oncology and therapeutic strategy design.
• SSLGRDA effectively enhances prediction ability and model generalization through robust node embeddings.