APPLICATIONS OF ARTIFICIAL INTELLIGENCE FOR CHEMICAL INFERENCE

The purpose of this paper is to discuss an interactive computer program which can be used to study questions of plausible ion structures. The program can be constrained in a variety of ways by the user, based on information inferred from (i) the structure of the intact molecule which gave rise to the ion, (ii) metastable defocusing experiments to identify progenitors, and (iii) existing knowledge of the fragmentations and rearrangements of related molecules, including deuterium labeled analogs. Implementation of these constraints is discussed primarily with the example of triethylamine, an apparently simple compound whose behavior is examined in detail with the aid of the program because of the complexity of its fragmentation. KNOWLEDGE of the structures of gaseous ions is fundamental to our understanding of the fragmentations and rearrangements of molecules subsequent to ionization in a mass spectrometer. The processes by which molecules decompose frequently involve highly energetic pathways, placing correlations with known structures of ions near or in the ground state (e.g. solution chemistry) on a largely speculative basis. Yet these correlations have contributed a great deal to our rationalization, if not deeper understanding, of many fragmentation processes. Of more value to a deeper understanding, however, is the knowledge of the actual ion structures. Recent advances in instrumental techniques, e.g. ion cyclotron resonance (i.c.r.)2 metastable ion analysis,j collision-induced metastable ions4 and field ionization analysis over the course of ion decomposition,5 have provided a variety of tools for indirect studies of ion structures. From these studies, structural features of ions can be inferred with much more certainty than in the past. Studies using i.c.r. or metastable ions normally require that two criteria be met. For completeness, one should have available all plausible structures for an ion of given composition (empirical formula). This ensures that no reasonable alternatives will be overlooked. For each reasonable alternative, one must have a means of producing the ion in an unambiguous way, so that its behavior under the experimental conditions can be standardized for comparison with unknown ion structures. This requires a search for compounds which on ionization or fragmentation can yield only one reasonable structure for the given decomposition. Using the structure generator,6 with constraints (‘CONGEW) where appropriate, we can now explore questions of possible isomeric ion structures. The availability of the program makes it possible to meet the first criterion mentioned in the previous paragraph. The remainder of this report is devoted to a discussion of how the program can be used under constraints inferred from a variety of sources, including the structure of the intact molecule, studies of metastable ions and mass spectra of labeled analogs. The example discussed in most detail is that of ion structures from the