The idea for this paper arose during talking over coffee at the BioExcel Alchemical Free Energy workshop in May 2019. We’d previously shown that alchemical free energy methods could successfully predict which mutations in S. aureus DHFRÂ confer resistance to trimethoprim (and crucially, which do not). That is all well and good, but to do this at scale, we’d need to be able to run such calculations quickly, hence this paper.
Part of the answer is making use of high performance computing, but part is also accepting that the primary goal of the calculations is not quantitative accuracy and precision, but instead resolving which side of a free threshold the change in antibiotic binding free energy induced by the mutation lies. That in turn enables the use of large numbers of very short lambda simulations which can be run in parallel, reducing the time-to-solution even further.
This, or similar methods, could be used in drug development (to assess how many codon mutations could allow a protein to escape the action of an inhibitor) or in diagnostics.
The paper is part of a special issue on Computational Medicine.
