"Big Data" Fast Chemoinformatics Model to Predict Generalized Born Radius and Solvent Accessibility as a Function of...

Area of Science:

• Computational chemistry
• Molecular modeling
• Biophysics

Background:

• The Generalized Born (GB) solvent model offers a good balance between accuracy and computational cost for solvation energy calculations.
• Estimating Effective Born Radii (EBRs) in GB models typically involves computationally expensive volume integration.
• Existing methods often simplify EBR calculations, potentially limiting accuracy for systems with significant geometric changes, such as in molecular dynamics or evolutionary algorithms.

Purpose of the Study:

• To develop a computationally efficient method for estimating Effective Born Radii (EBRs) and atomic solvent-accessible areas (SA) within the Generalized Born (GB) solvent model.
• To integrate this estimation method into the Sampler for Multiple Protein-Ligand Entities (S4MPLE) evolutionary algorithm, enabling rapid updates of EBRs during simulations with large geometric changes.
• To ensure the developed method maintains accuracy comparable to traditional, more computationally intensive approaches.

Main Methods:

• Developed a quantitative structure-property relationship (QSPR) to predict EBRs and SA as linear functions of atomic topological and geometric descriptors.
• Compiled a large and diverse training dataset (>5 million entries) including various molecular systems (fragments, drugs, proteins, complexes) and generated multiple conformers for each.
• Employed a bootstrapping multilinear regression with descriptor selection for model training and validated the approach on an independent external dataset (>2000 systems).

Main Results:

• Successfully developed linear models to accurately predict EBRs and SA using atomic descriptors.
• The QSPR models demonstrated high performance on both training and external validation sets.
• Solvation energies calculated using the estimated EBRs and SA showed good agreement with energies obtained through standard, computationally intensive methods, within typical force-field error margins.

Conclusions:

• The developed QSPR-based method provides a fast and accurate way to estimate EBRs and SA for the Generalized Born solvent model.
• This approach significantly reduces the computational cost associated with EBR updates in simulations involving substantial geometric variations.
• The method's broad applicability across diverse molecular systems makes it a valuable tool for computational chemistry and drug discovery efforts.