Blinded predictions of standard binding free energies: lessons learned from the SAMPL6 challenge

• DOI: 10.1007/s10822-018-0154-6
• Publication Date: 2018-08-29T21:55:54Z
• Authors: Michail Papadourakis; Stefano Bosisio; Julien Michel
• Keywords: Bridged-Ring Compounds; Macrocyclic Compounds; Osmolar Concentration; Carboxylic Acids; Imidazoles; Molecular Conformation; Proteins; Cycloparaffins; Models, Theoretical; Molecular Dynamics Simulation; Ligands; 01 natural sciences; 0104 chemical sciences; 0103 physical sciences; Solvents; Thermodynamics; Protein Binding

In the context of SAMPL6 challenges, series blinded predictions standard binding free energies were made with SOMD software for a dataset 27 host-guest systems featuring two octa-acids hosts (OA and TEMOA) cucurbituril ring (CB8) host. Three different models used, ModelA computes energy based on double annihilation technique; ModelB additionally takes into account long-range dispersion state corrections; ModelC introduces an empirical correction term derived from regression analysis SAMPL5 previously SOMD. The performance each model was evaluated setups; buffer explicitly matches ionic strength assays, whereas no-buffer merely neutralizes net charge counter-ions. ModelC/no-buffer shows lowest mean-unsigned error overall (MUE 1.29 < 1.39 1.50 kcal mol-1, 95% CI), while explicit modelling improves significantly results CB8 host only. Correlation experimental data ranges excellent TEMOA (R2 0.91 0.94 0.96), to poor 0.04 0.12 0.23). Further investigations indicate pronounced dependence modelled strength, variable reproducibility between simulation packages.