Prediction of DNA-binding residues from sequence features.

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Protein-DNA interactions are fundamental to essential cellular processes like transcription and DNA metabolism.
• Understanding the molecular basis of protein-DNA recognition is critical, yet remains challenging despite available structural data.
• The increasing volume of genomic data necessitates computational methods for identifying DNA-binding proteins and their residues.

Purpose of the Study:

• To develop and evaluate a predictive method for identifying DNA-binding residues directly from protein sequence data.
• To investigate the utility of various sequence-derived features in predicting DNA-binding residues.
• To establish a highly accurate computational classifier for DNA-binding residue prediction.

Main Methods:

• Training neural networks using five sequence-derived features: molecular mass, hydrophobicity index, side chain pKa, solvent accessible surface area, and conservation score.
• Evaluating the predictive performance using sensitivity, specificity, and Receiver Operating Characteristic (ROC) curves.
• Testing the classifier on different classes of DNA-binding proteins.

Main Results:

• The side chain pKa value emerged as the most significant feature for predicting DNA-binding residues.
• Combining multiple sequence-derived features substantially improved predictive accuracy.
• The final classifier, utilizing all five features, achieved 72.71% sensitivity and 67.73% specificity.
• This performance represents a significant advancement over existing sequence-based prediction methods.

Conclusions:

• Amino acid side chain ionization state is a critical determinant of DNA-binding.
• A multi-feature neural network approach provides a powerful and accurate tool for predicting DNA-binding residues from protein sequences.
• This method offers a valuable resource for genomic analysis and understanding protein function.