Computer-assisted interpretation of mass spectra

Under Defense Advanced Research Projects Agency sponsorship, APL is developing a miniature time-of-flight (TOF) mass spectrometer for early warning against exposure to chemical/biological agents. Intended for operation by a wide range of military and civilian personnel, the instrument must be able to detect and identify pathological agents within minutes. Key to this mission is the spectrometer operator's interpretation of the data. Typically, interpretation of mass spectra has been the realm of professional chemists and biochemists. Other operators must rely on computer classification of the TOF mass spectrometer's output. We describe algorithms that can be used to interpret mass spectra and that have been successful on a limited data set. These algorithms handle precisely known, and partially unknown, signatures. For precisely known signatures, a vector space problem can be formulated to estimate the optimum approximation of the measured spectrum with a combination of stored library signatures of threat agents. For partially unknown signatures, a Bayesian probabilistic approach has been taken to relate the potentially variable signature of a bacterial threat to likelihoods of chemical composition of bacterial lipids.