Electrophoresis can resolve thousands of proteins, but analysis of such data is extremely difficult. Automated methods for detection and recognition of proteins on two-dimensional (2D) electrophoretograms are needed. Image processing techniques and expert systems have been tried, but these are computationally intensive, require heuristic rule bases, and cannot identify proteins from analysis of their constituent polypeptide chains. Artificial neural networks are easy to train and implement, and should correctly categorize an almost unlimited number of proteins by their distinctive polypeptide patterns. Development of a protein identification system based on neural net recognition of polypeptide distribution in 2D electrophoretograms is proposed. Phase I research will develop a demonstration system to examine electrophoretograms of an unknown protein and identify it. The effort is an extension of successful neural net research by ORINCON in such diverse fields as biomedical diagnostics, underwater acoustics, and fault detection. Phase II will incorporate data fusion techniques and an expanded set of known proteins to enable analysis of entire 2D electrophoresis biological samples. This would lead to development of a commercial package for automated protein analysis of 2D electrophoretograms. It easily could be modified to categorize other constituents (e.g., amino and nucleic acids) simply by retraining the net.
