Development of an expert system for amino acid sequence identification

An expert system for amino acid sequence identification has been developed. The algorithm uses heuristic rules developed by human experts in protein sequencing. The system is applied to the chromatographic data of phenylthiohydantoin-amino acids acquired from an automated sequencer. The peak intensities in the current cycle are compared with those in the previous cycle, while the calibration and succeeding cycles are used as ancillary identification criteria when necessary. The retention time for each chromatographic peak in each cycle is corrected by the corresponding peak in the calibration cycle at the same run. The main improvement of our system compared with the onboard software used by the Applied Biosystems 477A Protein/Peptide Sequencer is that each peak in each cycle is assigned an identification name according to the corrected retention time to be used for the comparison with different cycles. The system was developed from analyses of ribonuclease A and evaluated by runs of four other protein samples that were not used in rule development. This paper demonstrates that rules developed by human experts can be automatically applied to sequence assignment. The expert system performed more accurately than the onboard software of the protein sequencer, in that the misidentification rates for the expert system were around 7%, whereas those for the onboard software were between 13 and 21%.