A protein-water interaction model for one-site water: incorporating the effect of water polarity to capture hydrophilicity and hydrophobicity at atomic resolution

Coarse-grained water models have been developed to reduce the size of simulating systems and their computational cost. One-site water models have clear benefits over cluster models, including finer granularity, but they may face challenges when applied to biomolecular systems, such as protein complexes, because water molecules are electrically neutral. In this work, we propose a mathematical model that describes interactions between protein atoms and coarse-grained one-site water molecules, by incorporating the effect of water polarity to describe interactions with charged protein atoms. This model is developed based on observations made in this work and in our previous work. First, at the atomic level, protein-water interactions are mostly controlled by electrostatic interactions between protein atoms and waters’ hydrogens and by the collective dynamics of all waters’ hydrogens. Second, interactions between protein atoms and water molecules can be composed of three components: (i) interactions by statistical preferences of the water orientations around protein atoms, (ii) perturbations of those interactions by neighboring protein atoms, and (iii) interactions with bulk water. Our results show that the proposed model accurately captures the hydrophilic and hydrophobic properties of protein atoms, including force profiles that indicate potential energy wells and repulsive walls, respectively, and reproduces the hydrophilic, hydrophobic, and amphipathic properties of amino acids at the atomic level. This work explains how individual protein atoms can exhibit hydrophilic or hydrophobic behavior. The proposed model has the effect of screening protein-water electrostatic interactions and the potential to significantly reduce the computational cost of the molecular dynamics simulations. In this work, we additionally propose the least-squares closest-distance (LSCD) fitting method, which has slight advantages over least-squares (LS) fitting by avoiding the overweighting of large-magnitude values, maintaining accuracy for values close to zero, and reducing the need for extremely small thresholds when fitting functions have complex forms.