CircSI-SSL: circRNA-binding site identification based on self-supervised learning

Area of Science:

• Biochemistry
• Bioinformatics
• Molecular Biology

Background:

• Circular RNAs (circRNAs) are gaining attention for their roles in protein binding.
• Current protein-binding site identification algorithms for circRNAs are supervised and require large amounts of labeled data.
• Acquiring labeled data for circRNA studies involves extensive and difficult biological experiments.

Purpose of the Study:

• To develop a novel self-supervised learning algorithm for circRNA protein-binding site identification.
• To overcome the limitations of supervised methods that require substantial labeled training samples.
• To provide a more efficient and scalable approach for predicting circRNA-protein interactions.

Main Methods:

• Proposed CircSI-SSL, a self-supervised learning algorithm for circRNA-binding site prediction.
• Combined multiple feature coding schemes and utilized RNA_Transformer for cross-view sequence prediction.
• Employed self-supervised learning to capture mutual information from multi-view data, followed by fine-tuning with minimal labeled samples.

Main Results:

• CircSI-SSL demonstrated excellent performance on six circRNA datasets, outperforming previous algorithms.
• Achieved high accuracy even with a 1:9 training-to-test data ratio.
• Showcased good scalability through successful transplantation experiments on six lncRNA datasets without modifications.

Conclusions:

• CircSI-SSL offers a powerful alternative to traditional supervised algorithms for circRNA-binding site prediction.
• The self-supervised approach significantly reduces the dependency on labeled data, making circRNA research more accessible.
• The algorithm's scalability suggests broad applicability in various RNA-related prediction tasks.