April 2 - 4, 2001. San Diego, California, USA, 221st American Chemical Society (ACS) National Meeting

Impacting PhysChem Property Prediction and Chromatography Simulation with User Training

Robert S. DeWitte, Michael McBrien, and Eduard Kolovanov

Abstract

A fundamental limitation of physical property prediction methods is the functional scope of the training set: in other words, has the algorithm ever seen anything like my compound? If the answer is no, then nearly all methods are likely to produce unsatisfactory results in some cases. Advanced Chemistry Development provides a solution to this limitation in a feature called "user training". This approach seamlessly employs measurements performed on proprietary compounds to enhance the predictive accuracy of LogP, LogD and pKa calculations. By coupling the relationships among these properties, user training on any one property positively impacts the predictions of the others. With ACD software, scientists can gain insight into entire families of highly novel structures by doing experiments on a very few examples. Physicochemical interactions form the basis for chromatographic separation. Since these parameters can be predicted from chemical structures, it follows that structure-based chromatograpy simulation is possible. User Training offers the ability to predict chromatographic retention times for novel species based on physicochemical data for related compounds, offering the method development chromatographer a unique evaluative/development tool.