PNNL

Abstract

Protein pKa calculation algorithms are typically developed to reproduce experimental pKa values and provide us with a better understanding of the fundamental importance of electrostatics for protein structure and function.  However, the approximations and adjustable parameters employed in almost all pKa calculation methods means that there is the risk that pKa calculation  algorithms are ‘over-fitted’ to the available  datasets, and that these methods therefore do not model protein physics realistically.  We employ simulations of the protein pKa calculation algorithm development process to show that  careful optimization procedures and  non-biased experimental datasets must be applied to ensure a realistic description of the  underlying physical  terms.  We furthermore investigate  the effect of experimental noise and find a significant effect on  the pKa calculation algorithm optimization landscape. Finally, we comment on strategies for ensuring the physical realism of  protein pKa calculation algorithms and we assess the  overall state of the field with a view to predicting future directions  of development.