The objective of this research was to improve systems for computer examination of the mass spectra of unknown pollutants. For this we have developed a new probability based matching (PBM) system for the retrieval of mass spectra from a large data base, and have substantially improved the interpretation of unknown mass spectra using the self-training interpretive and retrieval system (STIRS). PBM was designed as a prefilter to STIRS; if an unknown mass spectrum can be identified with a sufficiently high confidence by PBM, interpretation of the spectrum using STIRS is not necessary. The PBM system provides more efficient retrieval than presently accepted systems; it incorporates a 'reverse search' algorithm, and through the use of weighted mass and abundance data provides a statistically valid prediction of the confidence of the matches found. STIRS has been improved to give a confidence-level prediction of the presence of approximately equal to 200 particular substructural features in the unknown molecule. Extensive studies have been made to improve the data selection for most data classes used by STIRS, resulting in a much higher level of overall system performance. Operation efficiencies of both PBM and STIRS have been improved dramatically so that both require less than 1 minute on a laboratory PDP-11/45 computer. STIRS has been made available for outside use by long-distance phone connections to this PDP-11/45, and recently both PBM and STIRS have been made operational on the Cornell IBM-370/168 so that these are available internationally over the TYMNET computer network system.

"October 1976." Includes bibliographical references (pages 56-58). Prepared by Cornell University, Ithaca, New York, under grant no. Contract Number: R-801106.