Building chemometric models based on spectral data: A study of model performance and its applications

ANYL 77

Michelle D'Souza, michelle_d'souza@bio-rad.com, Bio-Rad Laboratories, Informatics Division, Two Penn Center Plaza, Suite 800, 1500 John F. Kennedy Blvd., Philadelphia, PA 19102, Gregory M. Banik, gregory_banik@bio-rad.com, Informatics Division, Sadtler Software & Databases, Bio-Rad Laboratories, Two Penn Center Plaza, Suite 800, 1500 John F. Kennedy Blvd., Philadelphia, PA 19102, and Scott Ramos, Infometrix, Inc, Suite 250, 10634 East Riverside Dr, Bothell, WA 98011.
In this study, we generated a series of hierarchical chemometric models that can be used to predict properties and/or generate classifications for unknown samples. Models were built using a large collection of Sadtler IR, 1H, and 13C NMR spectra. Currently, the following model types are supported:

- Categorical: bin a sample into a single class (asymmetric case), two classes (Boolean) or multiple classes. - Regression: predict values for a single property or multiple properties. - Consensus: combine Boolean models of different spectral technologies or different algorithms to generate a more accurate conclusion. Available scenarios are “percent agreement”, “majority rules”, “best case” and “worst case”.

This talk will discuss examples of these models' performance and applicability to analytical chemistry.

 

General Posters
7:00 PM-9:00 PM, Sunday, August 19, 2007 BCEC -- Exhibit Hall - B2, Poster

Division of Analytical Chemistry

The 234th ACS National Meeting, Boston, MA, August 19-23, 2007