Related Experiment Video
Updated: Jul 19, 2025
06:50
Author Spotlight: A Computational Approach to Decipher Amino Acid Preferences in Multispecific Protein-Protein Interactions
Published on: January 26, 2024
1.9K
Comparative Performance of High-Throughput Methods for Protein pKa Predictions
Wanlei Wei1, Hervé Hogues1, Traian Sulea1
1Human Health Therapeutics Research Centre, National Research Council Canada, 6100 Royalmount Avenue, Montreal, Quebec H4P 2R2, Canada.
Journal of Chemical Information and Modeling
|August 7, 2023
Summary
Accurate prediction of protein residue acidity (pKa) is crucial for engineering therapeutics. This study benchmarks seven computational methods, finding DeepKa, PKAI+, PROPKA3, and H++ useful, with consensus models improving accuracy for pH-dependent protein design.
Area of Science:
- Computational Biology
- Protein Engineering
- Biochemistry
Background:
- Protein engineering demands accurate pKa prediction for designing therapeutics targeting various biological pH environments.
- Existing high-throughput pKa prediction methods lack comprehensive benchmarking, hindering reliable computational protein design.
- Accurate pKa prediction is vital for optimizing protein function, especially in pH-sensitive applications like antibody-antigen binding.
Purpose of the Study:
- To systematically evaluate and benchmark seven accessible computational pKa prediction methods.
- To identify the most accurate and efficient algorithms for predicting protein residue pKa shifts.
- To provide guidance for selecting appropriate computational tools for protein engineering workflows.
Main Methods:
- Systematic testing of seven pKa prediction methods (PROPKA3, DeepKa, PKAI, PKAI+, DelPhiPKa, MCCE2, H++) against a non-redundant dataset of 408 experimental pKa shifts from the PKAD database.
- Statistical bootstrapping was employed to assess the confidence and utility of each method.
- Development of consensus models by averaging the predictions of the best-performing empirical methods.
Main Results:
- While no single method consistently outperformed null hypotheses, DeepKa, PKAI+, PROPKA3, and H++ demonstrated utility.
- DeepKa showed consistent performance across different amino acid types, exhibiting lower errors and higher correlations.
- Consensus models combining top empirical predictors achieved a root-mean-square error of 0.76 pKa units and R² of 0.45, improving accuracy and transferability.
Conclusions:
- DeepKa, PKAI+, PROPKA3, and H++ are valuable tools for pKa prediction in protein engineering.
- Consensus approaches significantly enhance the accuracy and reliability of computational pKa predictions.
- This benchmark provides a foundation for method development and guides the selection of computationally inexpensive pKa predictors for applications like pH-dependent antibody optimization.
