Robust, Efficient and Automated Methods for Accurate Prediction of Protein-Ligand Binding Affinities in AMBER Drug Discovery Boost

Abstract

Recent concurrent advances in methodology development, computer hardware and simulation software has transformed our ability to make practical, quantitative predictions of relative ligand binding affinities to guide rational drug design. In the past, these calculations have been hampered by the lack of affordable software with highly efficient implementations of state-of-the-art methods on specialized hardware such as graphical processing units, combined with the paucity of available workflows to streamline throughput for real-world industry applications. Herein we discuss recent methodology development, GPU-accelerated implementation, and workflow creation for alchemical free energy simulation methods in the AMBER Drug Discovery Boost (AMBER-DD Boost) package available as a patch to AMBER20. Among the methodological advances are 1) new methods for the treatment of softcore potentials that overcome long standing end-point catastrophe and softcore imbalance problems and enable single-step alchemical transformations between ligands, and 2) new adaptive enhanced sampling methods in the "alchemical" (or "λ") dimension to accelerate convergence and obtain high precision ligand binding affinity predictions, 3) robust network-wide analysis methods that include cycle closure and reference constraints and restraints, and 4) practical workflows that enable streamlined calculations on large datasets to be performed. Benchmark calculations on various systems demonstrate that these tools deliver an outstanding combination of accuracy and performance, resulting in reliable high-throughput binding affinity predictions at affordable cost.<br>

Practical Guidance for Consensus Scoring and Force Field Selection in Protein–Ligand Binding Free Energy Simulations

“…(A) Top: free energy convergence as a function of time for pair CAT-13j to CAT-4o with a 12 λ window scheme (used by Lee et al …”

“…The three different 12 λ sets are based on a linear scheme (0.000, 0.091, 0.182, 0.273, 0.364, 0.455, 0.545, 0.636, 0.727, 0.818, 0.909 and 1.000), the scheme used by Song et al (0.00922, 0.04794, 0.11505, 0.20634, 0.31608, 0.43738, 0.56262, 0.68392, 0.79366, 0.88495, 0.95206, 0.99078), 50 and the scheme used by Lee et al (0.0000, 0.0479, 0.1151, 0.2063, 0.3161, 0.4374, 0.5626, 0.6839, 0.7937, 0.8850, 0.9521, and 1.0000). 51 The convergence analysis was applied to ensure that the obtained 12 λ ΔΔG bind values were from the equilibrated simulations. For this, the trajectories used for the final ΔΔG bind were divided into 12 blocks to estimate the cumulative average ΔΔG bind in the forward direction.…”

AMBER Drug Discovery Boost Tools: Automated Workflow for Production Free-Energy Simulation Setup and Analysis (ProFESSA)

“…AFE simulations are used in a wide range of contexts, 15 but for the purposes of the current work, the focus will be placed on the calculation of absolute and relative solvation (ASFE and RSFE) and binding (ABFE and RBFE) free energies that are of primary importance to computer-aided drug discovery. 1,12,13,16,17 Pharmaceutical companies routinely use GPU-accelerated AFE calculations to design potency and selectivity to circumvent off-target effects and guide the prioritization of compounds for synthesis and testing in the lead optimization cycle. 1,12,13,16,17 Over the last several years, our lab has spearheaded the development of GPU-accelerated free-energy simulation and analysis methods in AMBER 1,9,10,18−23 and FE-ToolKit 24−26 and provided advanced β testing access to academic and industry partners through the AMBER Drug Discovery Boost (AMBER DD Boost) package 12 to facilitate method validation before its integration into the official AMBER release versions.…”

“…Alchemical free-energy (AFE) simulations have become an indispensable tool in computer-aided drug discovery. − In recent years, simultaneous advancement in computer hardware, simulation software, and free-energy methods has enabled highly efficient and increasingly accurate GPU-accelerated AFE simulations to address a broad scope of real-world drug discovery applications. ,− AFE simulations rely on physics-based atomistic models and statistical-mechanics methods ,,, and leverage the property that the free energy is a state function to enable nonphysical thermodynamic pathways to be constructed that are more amenable to practical computation. AFE simulations are used in a wide range of contexts, but for the purposes of the current work, the focus will be placed on the calculation of absolute and relative solvation (ASFE and RSFE) and binding (ABFE and RBFE) free energies that are of primary importance to computer-aided drug discovery. ,,,, …”

“…Pharmaceutical companies routinely use GPU-accelerated AFE calculations to design potency and selectivity to circumvent off-target effects and guide the prioritization of compounds for synthesis and testing in the lead optimization cycle. ,,,, Over the last several years, our lab has spearheaded the development of GPU-accelerated free-energy simulation and analysis methods in AMBER ,,,− and − and provided advanced β testing access to academic and industry partners through the AMBER Drug Discovery Boost (AMBER DD Boost) package to facilitate method validation before its integration into the official AMBER release versions. Among these have been the development of smoothstep softcore potentials and the introduction of flexible user control of interactions that enables optimization of alchemical transformation pathways, , new alchemical-enhanced sampling method (ACES) (see the description in the Supporting Information) to avoid kinetic traps and overcome local “hot-spot” problems in the λ dimension, and the release of to provide a robust set of network-wide free-energy analysis tools that includes the imposition of cycle closure and experimental constraints. − The goal of the current work is to report a new automated workflow for production free-energy simulation setup and analysis (ProFESSA) using the GPU-accelerated AMBER free-energy engine that integrates these new features, methods, and analysis tools.…”