Predicting Octanol-Water Partition Coefficients: Are Quantum Mechanical Implicit Solvent Models Better than Empirical...

Area of Science:

• Computational Chemistry
• Physical Chemistry
• Medicinal Chemistry

Background:

• Accurate prediction of octanol-water partition coefficients (log P_ow) is crucial for drug discovery and environmental fate assessment.
• Implicit and explicit solvent models are commonly used, but their performance varies.
• Empirical fragment-based methods offer an alternative approach for log P_ow prediction.

Purpose of the Study:

• To evaluate the performance of contemporary quantum mechanical implicit solvent models and empirical fragment-based methods for predicting log P_ow.
• To compare these methods against each other and against explicit solvent simulations.
• To identify the most accurate methods for log P_ow prediction.

Main Methods:

• Tested implicit solvent models (SMD, SM8, SM12, ADF-COSMO-RS) and empirical fragment-based methods (ALOGP, miLOGP).
• Utilized two test sets: 34 organic molecules and 55 fluorinated alkanols/carbohydrates.
• Compared results with explicit solvent simulations (GAFF, GAFF-DC) and the SAMPL6 log P_ow challenge.

Main Results:

• Implicit solvent models showed good performance with mean absolute errors (MAEs) around 0.6 log units.
• Empirical fragment-based methods (ALOGP, miLOGP) achieved significantly lower MAEs (0.2 to 0.4 log units).
• ALOGP demonstrated strong performance on the SAMPL6 challenge with an MAE of 0.32 log units.

Conclusions:

• Empirical fragment-based methods are superior to implicit and explicit solvent models for log P_ow prediction.
• ALOGP and miLOGP offer high accuracy and are recommended for log P_ow calculations.
• Systematic errors in implicit models lead to cancellation in transfer free energy calculations but do not match empirical method accuracy.