Report 77-20 Computer Assisted Structure Elucidation Using Stanford KSL Automatically Acquired 13C NMR Rules. Computer-Assisted Structure Elucidation Using Automatically Acquired '3C NMR Rules Carbon-13 nuclear magnetic resonance (CMR) has developed into an important tool for the structural chemist. A CMR spectrum exhibits a wide range of shifts which have been shown to have a strong correlation with structure(1 2). A natural abundance CMR spectrum which is fully proton decoupled consists of a number of sharp peaks which correspond to the resonance frequencies in an applied magnetic field of the various types of carbon atoms present. A C-13 shift is the amount an observed peak is shifted from that of a reference peak, usually tetramethylsilane (TMS). Molecular structure elucidation using CMR consists of establishing a set of rules which summarize the CMR behavior for a set of compounds and then using the rules to identify unknown compounds. In the traditional approach to structure elucidation using CMR the chemist forms a set of empirical rules by sorting through a large amount of data looking for correlations between structural arrangements in the molecuies and the observed C-13 shift. The total shift is then given as a function of these structural parameters. The functional fort, is usually chosen to be a linear combination of independent parameters. The optimized value of the coefficient of each structural parameter is obtained by a curve fitting procedure.

This characterization of the domain transformations leads naturally to the use of abstract objects and transformations in planning experiments. The abstractions correspond to the conceptual entities and transformations of the geneticist and will be an important part of the knowledge base.

The Experimental Steps in Detaii... 5 IV. Review of Knowledge Used in This Experiment 10 IV51 Proposing the Experiment..... 10 IV.2 Establishing Some Subgoals.

The DENDRAL research project was started in 1965 by Professors J. Lederberg and E.A. Feigenbaum and now includes Professor C. Djerassi in the Chemistry Department and about 20 other persons in Computer Science, Chemistry and Genetics. There are several aspects to the whole project, including research in chemistry and genetics, development of new chemical instrumentation and supporting computer programs, as well as artificial intelligence research. We have had two main computer science goals in this work: to study scientific inference and to aid working scientists. The two programs described below illustrate these concerns. Heuristic DENDRAL The Heuristic DENDRAL Program is designed to aid organic chemists determine the molecular structure of unknown compounds. Parts of the program have been highly tuned to work with experimental data from an analytical instrument known as a mass spectrometer. Mass spectrometry is a new and still developing analytical technique. It is not ordinarily the only analytic technique used by chemists, but is one of a broad array of analytic techniques including NMR, IR, UV, and "wet chemistry" analysis. It is particularly useful when the quantity of the sample to be identified is very small, for mass spectrometry requires only micrograms of sample.

A range of different encoding techniques have been developed, along with a number of approaches to applying knowledge. Most of the effort to daze however, has concentrated on representing and manipulating knowledge about a specific domain of application, like game-playing ([14D, natural language understanding ([153, [19]), speech understanding ([8], [II)), chemistry ([7]), etc. This paper explores a number of issues involving representation and use of what we term meta-level knowledge, or knowledge about knowledge'. It begins by defining the term, then exploring a few of its varieties and considering the range of capabilities it makes possible. Four specific examples of meta-level knowledge are described, and a demonstration given of their application to a number of problems, including interactive tranfer of expertise and the "intelligent" use of knowledge. Finally, we consider the long term implications of the concept and its likely impact on the design of large programs.

Crystallographers have been fascinated by computing The problem of deriving the coordinates for a trial devices for many years and have done much pioneering protein structure, given an electron density map, the work in the design and utilization of such devices for amino-acid sequence and the stereochemical principles crystallographic research. Machines such as the 1948 and constraints known to apply, is one which currently analogue Fourier summation device X-RAC (Pepinsky.

C. Richard Johnson, Jr. provided very helpful comments on adaptive control systems. We received many valuable suggestions from members of the Heuristic Programming Project at Stanford. 2 Supported by the Research and Development Branch of the Department of National Defence of Canada.

A candidate elimination algorithm has been shown whicn will find all rule versions consistent with all training instances. Backtracking is not required for noise-free training instances, and the final result is independent of the order of presentation of instances. Version spaces provide at once a compact summary of past training instances and a representation of all plausible rule versions. Pecause they provide an explicit representation for the space of plausible rules, version spaces allow a program to represent "how much it doesn't know" about the correct form of the rule. This suggests the utility of the version space approach to problems such as intelligent selection of training instances and merging sets of independently generated rules.

Computer facilities were provided by t"e SUMEX-AIM facility at Stanford University under National Institutes of health grant RR-00785. The author is supported by the Research and Development Branch of the Department of National Defence of Canada.

Report 77-11 Structure Elucidation Based on Computer S Stanford -- KSL Analysis of High and Low Resolution Mass Spectral Data. A tremendous effort has been directed toward development of advanced instrumentation for mass spectrometrlc analysis. Advancements include everincreasing sensitivities and resolving powers, new ionization techniques, metastable ion probes of ion decomposition and structure and computer systems for rapid acquisition and reduction of data. We sometimes lose sight of the fact that these developments are designed to provide information about chemical and biochemic,a1 structures at greater TeTTE--aTrin greater detail than previously available. The ultimate goal in most research in mass spectrometry is to provide powerful tools for molecular structure elucidation, either directly, by exploitation of existing techniques, or indirectly by development of new techniques. Concurrently, several computer-based techniques designed to assist chemists in the analysis and interpretation of mass spectral data have been developed. Reprinted with permission from Smith, Dennis H. and Carhart, Raymond E. in "High Performance Mass Spectrometry: Chemical Applications," Michael Gross, Ed., in ACS SYMPOSIUM SERIES, No. 70; American Chemical Society: Washington, D.C., 1978, pp.325-347. Library search procedures (2) and their extensions (I) or attern recoviitioz programs (4) may provide clues to t e identity of the structure or be used to determine the structure uniquely. A computer program for analysis of spectra based on classspecific fragmentation rules, is available (5). There are several reasons or this lag: There is no formal theory.