October 18 - 20, 2005, EUM 2005, Obernai, France

Improving ACD/PhysChem Predictions: Accuracy Extender vs User Training Databases

David C. Adams, Ed Kolovanov

Abstract

A powerful feature in ACD/PhysChem Prediction software is the ability to train the algorithms to improve prediction accuracy. Training can be done in two different ways. The first option is to build a User Database containing representative compounds and experimental data from the chemical classes of interest and to use this database for algorithm training. The second training option, ACD/PhyChem Accuracy Extender, allows users to define their own new fragments and aromatic or aliphatic interactions, and assign corresponding values to these increments to dramatically increase the accuracy of LogP and Solubility prediction for specific classes. Similarly, ACD/pK_a Accuracy Extender allows you to design custom Hammett/Taft equations for particular novel chemical classes. This talk will discuss the differences between these two approaches.

Download the presentation in MS PowerPoint (897 Kb ZIP file) or Adobe Acrobat format (579 Kb PDF file).

Relevant Products: LogP, LogD, pK_a, Solubility, LogD Sol suite, LogD suite