A confidence predictor for logD using conformal regression and a support-vector machine.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Lipophilicity, quantified by the water-octanol distribution coefficient (logD), is crucial for predicting Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) properties.
• Accurate prediction of logD is essential for efficient drug candidate selection and optimization in pharmaceutical research.

Purpose of the Study:

• To develop and validate large-scale predictive models for logD.
• To provide reliable logD predictions with associated confidence intervals for drug discovery projects.

Main Methods:

• Utilized ACD/logD data for 1.6 million compounds from the ChEMBL database.
• Employed a support-vector machine with a linear kernel and conformal prediction methodology.
• Evaluated model performance using metrics such as predictive ability and median prediction interval width.

Main Results:

• Achieved a predictive ability of [Formula: see text].
• The best performing model provided median prediction intervals of [Formula: see text] log units at 80% confidence and [Formula: see text] log units at 90% confidence.
• Developed an online service with an OpenAPI interface and a web page for predictions.

Conclusions:

• The developed logD prediction models are accurate and provide valuable confidence intervals.
• The models and associated data are made available through various online services and downloadable formats to support drug discovery efforts.
• The large-scale prediction of logD for millions of compounds facilitates broader application in cheminformatics and drug design.