Benchmarking pKa Prediction Methods for Residues in Proteins

Area of Science:

• Biochemistry
• Computational Biology
• Protein Chemistry

Background:

• Estimating the ionization state of protein residues is crucial for understanding protein function and interactions.
• Experimentally determining residue pKa values can be challenging, necessitating accurate computational prediction methods.

Purpose of the Study:

• To evaluate the accuracy of two computational methods, MD/GB/TI and PROPKA, for predicting protein residue pKa values.
• To compare the performance of these methods on a benchmark set of proteins with known experimental pKa values, including residues with significant pKa variations.

Main Methods:

• The molecular dynamics/generalized-Born/thermodynamic integration (MD/GB/TI) technique was applied.
• The empirical PROPKA method was utilized for pKa prediction.
• Both methods were tested on a benchmark set of 80 residues (Asp, Glu, Lys, His), with half exhibiting significant pKa shifts.

Main Results:

• Both MD/GB/TI and PROPKA yielded a root-mean-square deviation (rmsd) of 1.4 pKa units on the benchmark set.
• MD/GB/TI accuracy improved with longer simulation times and the inclusion of explicit waters near the residue.
• PROPKA showed good performance for Lys and Glu but less accuracy for Asp and His; absolute deviations were similar for both methods (5.1-5.2).

Conclusions:

• Both MD/GB/TI and PROPKA offer comparable accuracy for predicting protein residue pKa values, with rmsd of 1.4.
• Factors like residue type, solvent accessible surface area (SASA), and pKa variation influence prediction accuracy.
• Further refinement of computational methods, including explicit solvent and longer simulations, can enhance pKa prediction accuracy.