Applications of Physical Property Prediction Software to the Screening of Combinatorial Libraries

(LogP, pK_a, LogD and associated properties)

Antony Williams, Eduard Kolovanov and Maria Foster

1. Abstract
2. Introduction
3. ACD Prediction Algorithms
4. LogD Suite: Algorithms and Databases
5. User Database Examples
6. LogD Example

7. LogP Example
8. pK_a Example
9. Batch Predictions
10. ISIS Integration
11. Conclusion
12. References

I. Abstract

Desktop software utilizing physical property prediction algorithms has been developed to allow the prediction of LogP, pK_a and LogD values directly from molecular structures.

Using batch mode prediction capability to generate these values it is now possible to populate combinatorial libraries with predicted values and utilize these parameters as constraints for the screening of these libraries according to physical property characteristics.

One recent development is the ability to perform LogP predictions following generation of a User Database containing molecular structures and measured LogP values unique to a particular laboratory and therefore to a specific set of molecules.

II. Introduction

Predictive software utilizing algorithms developed using molecular structure and physical property correlations is an extremely useful tool for medicinal chemists. A number of commercial products for physical property prediction exist and these include boiling point, vapor pressure, LogP (Kow), pK_a, LogD and toxicity predictors.

The experimental determination of such properties can be time consuming and tedious as well as, in some cases, being subject to large experimental variation and errors. To address the need to provide reduced time-scale for determination of physical property values, as well as provide significant cost benefits, prediction software is often utilized. Our software uses algorithms based on an internal data set of experimental values.

The accuracy of the prediction can also be improved using your own data. This important feature of many ACD programs is called the USER Database.

Dealing with large databases it is possible to run our property prediction algorithms in Batch mode. With the databases populated with predicted values for different physical properties virtual libraries or structural libraries can be screened and filtered using these important properties as criteria. Other properties predicted using our algorithms include the Bioconcentration Factor (BCF), Aqueous Solubility at 25C, the Adsorption Coefficient (Koc) and the Boiling Point.