A comparative analysis of machine learning classifiers for predicting protein-binding nucleotides in RNA sequences.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• RNA-protein interactions are fundamental to cellular processes but their molecular mechanisms remain poorly understood.
• Experimental determination of RNA-protein complex structures is challenging, limiting mechanistic insights.
• High-throughput sequencing yields vast RNA sequence data, outpacing structural data availability.

Purpose of the Study:

• To develop computational methods for identifying protein-binding sites on RNA sequences without requiring structural information.
• To investigate the efficacy of various machine learning classifiers and sequence-derived features for this prediction task.

Main Methods:

• Utilized machine learning classifiers, including random forest models.
• Explored features derived from nucleotide triplets and quartets for sequence analysis.
• Evaluated model performance using metrics such as accuracy, sensitivity, specificity, MCC, and AUC.

Main Results:

• Random forest models incorporating nucleotide-triplet and nucleotide-quartet features demonstrated superior performance.
• Achieved high prediction accuracy (84.8%), sensitivity (83.2%), specificity (86.1%), MCC (0.70), and AUC (0.93).

Conclusions:

• The developed computational approach effectively predicts RNA-protein binding sites from sequence data.
• A user-friendly webserver, Nucpred, has been created and made publicly accessible for broader research application.