Predicting lncRNA-protein interactions through deep learning framework employing multiple features and random forest...

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• RNA-protein interactions (RPI) are fundamental to numerous biological processes.
• Experimental identification of RPI is time-consuming and costly.
• Current computational methods for RPI prediction lack sufficient robustness and generalizability.

Purpose of the Study:

• To develop an advanced computational model, LPI-MFF, for accurate RPI prediction.
• To improve the robustness and generalizability of RPI prediction models.
• To address limitations in existing machine learning and deep learning-based RPI prediction approaches.

Main Methods:

• LPI-MFF integrates multi-source information including protein-protein interactions, sequence features, secondary structure, and physicochemical properties.
• A random forest algorithm was utilized for effective feature screening.
• A convolutional neural network (CNN) served as the final classification model.

Main Results:

• LPI-MFF achieved high accuracy rates of 97.60% on RPI1807 and 97.67% on NPInter datasets via fivefold cross-validation.
• The model demonstrated strong performance on an independent test set (RPI1168) with 84.9% accuracy.
• An accuracy of 90.91% was recorded on the Mus musculus dataset.

Conclusions:

• LPI-MFF exhibits superior robustness and generalization capabilities compared to prevalent RPI prediction methods.
• The multi-source information fusion strategy effectively enhances RPI prediction performance.
• This model offers a promising computational tool for advancing RPI research.