Transfer Learning for Predicting ncRNA-Protein Interactions.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Noncoding RNAs (ncRNAs) regulate gene expression and cellular functions through protein interactions.
• Accurate prediction of ncRNA-protein interactions (ncRPI) is vital for biological understanding and therapeutic development.
• Experimental identification of ncRPI is costly and time-consuming, necessitating computational approaches.

Purpose of the Study:

• To develop a novel framework, Transfer-RPI, for enhancing ncRPI prediction accuracy.
• To leverage transfer learning and deep feature learning to overcome data limitations in ncRPI prediction.
• To improve the generalization and performance of ncRPI prediction models, especially on smaller datasets.

Main Methods:

• Utilized a transfer learning-based framework, Transfer-RPI.
• Employed RiNALMo and ESM models for comprehensive feature extraction from RNA and protein sequences, respectively.
• Integrated rich feature sets and fine-tuned deep learning architectures for complex interaction pattern recognition.

Main Results:

• Transfer-RPI demonstrated superior performance compared to existing methods across multiple datasets.
• Achieved high accuracies, including 80.1% (RPI369) to 95.4% (NPInter v2.0), under 5-fold cross-validation.
• Deep learning with advanced feature representations and transfer learning significantly boosted prediction accuracy.

Conclusions:

• Transfer learning effectively addresses data limitations in ncRPI prediction.
• Transfer-RPI provides a powerful computational tool for uncovering ncRPI.
• The findings pave the way for deeper molecular biology insights and novel therapeutic innovations.