We have developed a new WWW-based search engine for the CCRC-Net for a data set of 1H-NMR spectra of glucuronoxylomannans (GXMs) from Cryptococcus neoformans compiled by Drs. Cherniak and Morris. This search engine is capable of quantitatively analyzing GXM 1H-NMR spectra, giving an accurate estimate of the mixture ratio of the six possible GXM chemotypes in the sample. The importance of C. neoformans as an opportunistic pathogen in immuno-compromised patients has increased dramatically in recent years, particularly in patients with AIDS. Greater than 90% of the cell envelope of the yeast is carbohydrate; the major cell envelope component is the capsular carbohydrate GXM. GXM serves as the basis for the current serotyping scheme for the yeast. The unique difficulty of this project was that all the spectra that were presented to the neural network during training were spectra of mixtures. The results demonstrate the ability of neural networks to analyze the components of a mixture spectrum without having seen pure spectra of its individual components during training. This recognition capability is particularly useful in the case of mixtures, as it is often difficult to separate their components by chemical methods. An article is being prepared for publication describing this quantitative analytical technique. A grant proposal is also being prepared by Drs. Robert Cherniak and Faramarz Valafar as coinvestigators for submission to the National Institutes of Health.
Showing the most recent 10 out of 245 publications
