A novel method for quantitatively predicting non-covalent interactions from protein and nucleic acid sequence.

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Non-covalent interactions, including hydrogen bonds and van der Waals contacts, are fundamental to biochemical interactions between proteins and nucleic acids.
• Accurate prediction of these interactions is essential for understanding molecular recognition and complex formation.

Purpose of the Study:

• To develop a novel computational method for quantitatively predicting the number of hydrogen bonds and van der Waals contacts in hypothetical protein-nucleic acid complexes.
• To identify key sequence-based features that govern the formation of these dominant non-covalent interactions.

Main Methods:

• Utilized Support Vector Machine Regression (SVR) models.
• Developed a hybrid feature incorporating sequence-length fraction, conjoint triad for protein sequences, and gapped dinucleotide composition.
• Evaluated model performance based on prediction accuracy.

Main Results:

• The SVR-based models demonstrated excellent predictive performance.
• Amino acid polarity was identified as a critical factor influencing hydrogen bond and van der Waals contact formation.
• A web server, H-VDW, was developed for public access to the prediction models.

Conclusions:

• The developed method provides an effective way to predict key non-covalent interactions in protein-nucleic acid complexes from sequence data.
• Sequence-derived features and amino acid properties are valuable for computational prediction of molecular interactions.
• The H-VDW web server facilitates further research in structural bioinformatics and drug design.