SAMPL6 logP challenge: machine learning and quantum mechanical approaches.

Area of Science:

• Computational Chemistry
• Cheminformatics

Background:

• Predicting the octanol-water partition coefficient (logP) is crucial for drug discovery and environmental fate assessment.
• Accurate logP prediction aids in optimizing molecular properties for desired applications.

Purpose of the Study:

• To evaluate two distinct computational strategies for predicting logP coefficients.
• To assess the performance of these methods in the context of the SAMPL6 logP blind prediction challenge.

Main Methods:

• Utilized density functional theory (DFT) for electronic structure optimization and calculation of molecular descriptors (e.g., van der Waals areas, HOMO/LUMO energies, dipole moments).
• Employed machine learning models, including multilinear regression and partial least squares, trained on a dataset of 97 molecules.
• Investigated electronic structure vertical solvation approaches, incorporating DFT and domain-based local pair natural orbital methods with solvation models.

Main Results:

• Calculated various molecular descriptors from optimized electronic structures.
• Developed and trained regression and machine learning models using these descriptors.
• Applied electronic structure vertical solvation methods for comparative analysis.

Conclusions:

• The study explored advanced computational techniques for logP prediction.
• The findings contribute to the ongoing development of predictive models in cheminformatics and computational chemistry.