PDA-MutPred: Reliable prediction of binding affinity change upon mutation in protein-DNA complexes

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Mutations in protein-DNA complexes disrupt cellular functions by altering structural integrity and binding affinities.
• Accurately predicting changes in binding affinity (ΔΔG) due to mutations is a significant challenge in computational biology.

Purpose of the Study:

• To analyze sequence, structural, and network features influencing binding affinity changes in protein-DNA complexes.
• To develop and validate machine learning models for predicting mutation-induced ΔΔG in protein-DNA interactions.
• To provide a web server resource for predicting these affinity changes.

Main Methods:

• Compiled a dataset of 1169 mutations from 256 protein-DNA complexes using the ProNAB database.
• Analyzed sequence, structural (atom contacts, hydrogen bonding), and network-based features, including amino acid properties and unfolding energy.
• Developed and evaluated machine learning models, including classification based on DNA strand and protein functional class, using 10-fold cross-validation and blind testing.

Main Results:

• Identified atom contacts, side-chain hydrogen bonding, amino acid properties, and unfolding energy as critical predictors of binding affinity changes.
• Achieved an average correlation of 0.69 and MAE of 0.73 kcal/mol in cross-validation.
• Demonstrated strong performance in a blind test with a correlation of 0.65 and MAE of 0.60 kcal/mol, outperforming existing methods.

Conclusions:

• Key sequence, structural, and network features significantly impact binding affinity changes upon mutation in protein-DNA complexes.
• Machine learning models effectively predict ΔΔG, offering improved accuracy over existing methods.
• The developed web server serves as a valuable resource for researchers studying protein-DNA interactions and designing novel complexes.