Predicting protein-nucleic acid interactions via protein language models with biophysical and evolutionary priors

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Protein-nucleic acid interactions are vital for cellular functions.
• Predicting these interactions is crucial for understanding biological processes and developing therapeutics.
• Existing methods often require complex structural data or have limitations in accuracy.

Purpose of the Study:

• To develop a novel computational framework, PNABPred, for predicting protein-nucleic acid interactions.
• To integrate diverse data types including biophysical and evolutionary information into a protein language model.
• To enhance the accuracy and efficiency of identifying nucleic acid-binding proteins.

Main Methods:

• Developed PNABPred, a multi-modal framework integrating semantic, evolutionary, and biophysical representations.
• Utilized a protein language model architecture.
• Input requires only protein sequences, enabling prediction even in intrinsically disordered regions.

Main Results:

• PNABPred significantly outperforms state-of-the-art sequence-based methods in RNA-binding protein classification and DNA-binding site prediction.
• Achieved high performance metrics: MCC of 0.889 and AUROC of 0.990 for RNA-binding protein classification.
• Demonstrated superior performance over CLAPE-DB in DNA-binding site prediction on the Test_129 dataset.

Conclusions:

• PNABPred offers a powerful and accurate method for predicting protein-nucleic acid interactions using sequence data alone.
• The framework's ability to identify binding sites in disordered regions is a significant advancement.
• PNABPred supports scalable applications in basic research, biotechnology, and therapeutic development.