Proteomic Pattern Analysis Using Neural Networks

Protein profiling of biologic samples by techniques such as surface-enhanced laser desorption/ ionization (SELDI) or matrix assisted laser desorption/ ionization (MALDI) yields massive amounts of data that require use of automated techniques to detect expression patterns. This paper suggests a neural network based classification and clustering technique for the analysis of proteomic data on serum samples collected from human subjects exposed to diesel exhaust fumes (DEF). Data were collected on samples from 93 subjects exposed to DEF. Proteomic patterns were analyzed using Neuralware Predict (R) software obtained from Neuralware Inc. The cascade correlation algorithm was used as the classification algorithm and self-organizing maps (SOM) was used as the clustering algorithm. The protein peaks were identified using the Ciphergen Software. The most discriminating peaks were identified by applying a student t-test and using the p-value as the criterion for discrimination. The classification and clustering algorithms were applied to the two data sets. The use of a neural network program for analysis of proteomic patterns from serum samples obtained from human subjects exposed to DEF or not exposed to DEF showed excellent discrimination. Such an approach has potential to play an important role in determining deleterious effects of occupational exposures and discovery of biomarkers.