Dataset Information

Accurate Prediction of Aqueous Free Solvation Energies Using 3D Atomic Feature-Based Graph Neural Network with Transfer Learning.

ABSTRACT: Graph neural network (GNN)-based deep learning (DL) models have been widely implemented to predict the experimental aqueous solvation free energy, while its prediction accuracy has reached a plateau partly due to the scarcity of available experimental data. In order to tackle this challenge, we first build a large and diverse calculated data set Frag20-Aqsol-100K of aqueous solvation free energy with reasonable computational cost and accuracy via electronic structure calculations with continuum solvent models. Then, we develop a novel 3D atomic feature-based GNN model with the principal neighborhood aggregation (PNAConv) and demonstrate that 3D atomic features obtained from molecular mechanics-optimized geometries can significantly improve the learning power of GNN models in predicting calculated solvation free energies. Finally, we employ a transfer learning strategy by pre-training our DL model on Frag20-Aqsol-100K and fine-tuning it on the small experimental data set, and the fine-tuned model A3D-PNAConv-FT achieves the state-of-the-art prediction on the FreeSolv data set with a root-mean-squared error of 0.719 kcal/mol and a mean-absolute error of 0.417 kcal/mol using random data splits. These results indicate that integrating molecular modeling and DL would be a promising strategy to develop robust prediction models in molecular science. The source code and data are accessible at: https://yzhang.hpc.nyu.edu/IMA.

SUBMITTER: Zhang D

PROVIDER: S-EPMC9038704 | biostudies-literature | 2022 Apr

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Accurate Prediction of Aqueous Free Solvation Energies Using 3D Atomic Feature-Based Graph Neural Network with Transfer Learning.

Zhang Dongdong D Xia Song S Zhang Yingkai Y

Journal of chemical information and modeling 20220414 8

Graph neural network (GNN)-based deep learning (DL) models have been widely implemented to predict the experimental aqueous solvation free energy, while its prediction accuracy has reached a plateau partly due to the scarcity of available experimental data. In order to tackle this challenge, we first build a large and diverse calculated data set Frag20-Aqsol-100K of aqueous solvation free energy with reasonable computational cost and accuracy via electronic structure calculations with continuum ...[more]

PMID: 35422122

Similar Datasets

Project description:Building on the SVPE (surface and volume polarization for electrostatics) model for electrostatic contributions to the free energy of solvation with explicit consideration of both surface and volume polarization effects, on the SMx approach to including first-solvation-shell contributions, and on the linear relationship between the electric field and short-range electrostatic contributions found by Chipman, we have developed a new method for computing absolute aqueous solvation free energies by combining the SVPE method with semiempirical terms that account for effects beyond bulk electrostatics. The new method is called SMVLE, and the elements it contains are denoted by SVPE-CDSL where SVPE denotes accounting for bulk electrostatic interactions between solute and solvent with both surface and volume contributions, CDS denotes the inclusion of solvent cavitation, changes in dispersion energy, and possible changes in local solvent structure by a semiempirical term utilizing geometry-dependent atomic surface tensions as implemented in SMx models, and L represents the local electrostatic effect derived from the outward-directed normal electric field on the cavity surface. The semiempirical CDS and L terms together represent the deviation of short-range contributions to the free energy of solvation from those accounted for by the SVPE term based on the bulk solvent dielectric constant. A solute training set containing a broad range of molecules used previously in the development of SM6 is used here for SMVLE model calibration. The aqueous solvation free energies predicted by the parameterized SMVLE model correlate exceedingly well with experimental values. The square of the correlation coefficient is 0.9949 and the slope is 1.0079. Comparison of the final SMVLE model against the earlier SMx solvation model shows that the parameterized SMVLE model not only yields good accuracy for neutrals but also significantly increases the accuracy for ions, making it the best implicit solvation model to date for aqueous solvation free energies of ions. The semiempirical terms associated with the outward-directed electric field account in a physical way for the improvement in the predictive accuracy for ions. The SMVLE method greatly decreases the need to include explicit water molecules for accurate modeling of solvation free energies of ions.

Project description:A fast and accurate method to compute the total solvation free energies of proteins as a function of pH is presented. The method makes use of a combination of approaches, some of which have already appeared in the literature; (i) the Poisson equation is solved with an optimized fast adaptive multigrid boundary element (FAMBE) method; (ii) the electrostatic free energies of the ionizable sites are calculated for their neutral and charged states by using a detailed model of atomic charges; (iii) a set of optimal atomic radii is used to define a precise dielectric surface interface; (iv) a multilevel adaptive tessellation of this dielectric surface interface is achieved by using multisized boundary elements; and (v) 1:1 salt effects are included. The equilibrium proton binding/release is calculated with the Tanford-Schellman integral if the proteins contain more than approximately 20-25 ionizable groups; for a smaller number of ionizable groups, the ionization partition function is calculated directly. The FAMBE method is tested as a function of pH (FAMBE-pH) with three proteins, namely, bovine pancreatic trypsin inhibitor (BPTI), hen egg white lysozyme (HEWL), and bovine pancreatic ribonuclease A (RNaseA). The results are (a) the FAMBE-pH method reproduces the observed pK a's of the ionizable groups of these proteins within an average absolute value of 0.4 p K units and a maximum error of 1.2 p K units and (b) comparison of the calculated total pH-dependent solvation free energy for BPTI, between the exact calculation of the ionization partition function and the Tanford-Schellman integral method, shows agreement within 1.2 kcal/mol. These results indicate that calculation of total solvation free energies with the FAMBE-pH method can provide an accurate prediction of protein conformational stability at a given fixed pH and, if coupled with molecular mechanics or molecular dynamics methods, can also be used for more realistic studies of protein folding, unfolding, and dynamics, as a function of pH.

Dataset Information

Accurate Prediction of Aqueous Free Solvation Energies Using 3D Atomic Feature-Based Graph Neural Network with Transfer Learning.

Publications

Accurate Prediction of Aqueous Free Solvation Energies Using 3D Atomic Feature-Based Graph Neural Network with Transfer Learning.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets