Alchemical free energy simulations have been very popular in drug design problems due to their promise of high accuracy.  Nevertheless, these calculations can be notoriously difficult to converge to the level of precision required for quantitative prediction required for ranking in lead optimization.  Very recently this landscape has changed with the development of GPU-accelerated free energy simulations that are orders of magnitude more efficient.  Here we report recently developed robust and efficient free energy estimation (?-SAMS) and analysis (2D-vFEP(?)) methods for GPU-accelerated alchemical free energy simulations of protein-liagnd binding.  The ?-SAMS method runs the Gibbs-sampler along a fine grid in ?-space to produce a statistically optimal distribution and solves the phase space overlap problem that is usually seen in FEP/BAR/MBAR type simulations.  The 2D-vFEP-(?) analysis, operating on dU/d? data generated by ?-SAMS, uses a maximum likelihood approach to create an optimally smooth analytic 2D map of the (?, dU/d?) distribution that enables highly accurate TI.  Preliminary benchmark results show a 3-5 fold speed up relative to more traditional replica exchange enhanced sampling methods and MBAR analysis.  This technology enables the calculation of alchemical free energy transformations using conventional MM force fields to sufficiently high precision to allow high-level quantum mechanical corrections to be made using new free energy "book-ending" approaches to correct endpoint states.  Together, these methods create the foundation for a new flexible and efficient free energy pipeline for protein-ligand binding affinity predictions using both traditional and new quantum mechanical force fields for lead optimization.<br/>

Towards accurate and precise binding affinity predictions: the

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.