Morton E. Munk

A neural network approach to infrared spectrum interpretation

The simple linear neural network model was investigated as a method for automated interpretation of infrared spectra. The model was trained using a database of infrared spectra of organic compounds of known structure. The model was able to learn, without any prior input of spectrum-structure correlations, to recognize and identify 76 functional groupings with accuracies ranging from fair to excellent. The effect of network input parameters and of training set composition were studied, and several sources of spurious correlations were identified and corrected.

Neural network models for infrared spectrum interpretation

A neural network model having a layer of hidden units is described which can identify functional groups in organic compounds, based on their infrared spectra. This network shows substantially better performance than the simple linear model reported earlier. The effect of the training set size and composition, the number of hidden units used, and the training time were studied.

Structure generation by reduction: a new strategy for computer-assisted structure elucidation

A problem common to computer programs for structure elucidation is the efficient and prospective use of the input information to constrain the structure generation process. The input may consist of potentially overlapping substructure requirements and alternative substructure interpretations of spectral data. Other useful information may be structural features that must not be present in the output structures. All of these may interact in a complex manner that is impossible to determine by use of a bond-by-bond structure assembly algorithm. A new method is described called structure reduction. In contrast to structure assembly, this method begins with a set of all bonds and removes inconsistent bonds as structure generation progresses. This results in a more efficient use of the input information and the ability to use potentially overlapping required substructures. Several examples illustrate the application of our computer program COCOA, which uses this method to solve real-world structure elucidation problems.

SPECTRA ESTIMATION FOR COMPUTER-AIDED STRUCTURE DETERMINATION

A recently developed program for estimating 1H-NMR chemical shifts has been interfaced to a structure generator. It provided predicted chemical shifts for 89% of the protons occurring in ca. 110 00...

An Approach to the Assignment of Canonical Connection Tables and Topological Symmetry Perception

Many nonnumerical computer applications in chemistry require algorithms for topological symmetry perception (constitutional symmetry, the automorphism partitioning problem), the assignment of canonical connection tables (the coding problem), or the detection of graph isomorphism. Important principles in designing such algorithms (vertex-classification, depth-first construction of sequence number permutations, and the use of automorphisms to restrict the construction process) are presented. In addition, a detailed description is presented for the algorithm used in program CASE for the assignment of canonical connection tables and topological symmetry perception.

Graph automorphism perception algorithms in computer-enhanced structure elucidation.

The concept of graph symmetry is explained in terms of the vertex automorphism group, which is a subgroup of the complete vertex permutation group. The automorphism group can be deduced from the automorphism partition of graph vertices. An algorithm is described which constructs the automorphism group of a graph from the automorphism vertex partitioning. The algorithm is useful especially for graphs which contain more than one vertex-partition set. Several well-known topological symmetry perception algorithms that yield automorphism partitions are compared. The comparison is favorable to the Shelley-Munk algorithm, developed in the framework of the SESAMI system for computer-enhanced structure elucidation.

An approach to automated partial structure expansion

Abstract An algorithm (ASSEMBLE) to construct all structures consistent with the structural implications of the chemical and spectroscopic properties of an unknown molecule is described. The design of ASSEMBLE takes cognizance of the need to supply some nonoverlapping substructure information in addition to the molecular formula, and the use of structural constraints that cannot be directly expressed as non-overlapping fragments. ASSEMBLE employs several heuristics (rules) intended to avoid the assembly of identical (isomorphic) graphs. To provide a non-redundant list of structures, duplicate structures are recognized and removed by a naming algorithm. ASSEMBLE also perceives different π-resonance forms as identical structures even when they are topologically non-equivalent.

Computer-assisted interpretation of infrared spectra

Abstract Pattern recognition and artificial intelligence programming techniques for the interpretation of infrared spectra are compared in an effort to determine the best technique for assisting the solution of actual structural elucidation problems. For several reasons, artificial intelligence is the method of choice. When information pertaining to a large number of classes is required, an excessively large training set would be needed with pattern recognition procedures. If the program makes a mistake, it must be alterable so that a similar error will not occur again. Artificial intelligence programs are amenable to this correction procedure.

A novel formalism to characterize the degree of unsaturation of organic molecules.

The existing formalism to calculate the degree of unsaturation from the molecular formula of organic molecules cannot be applied to charged and/or disconnected species. Moreover, the calculated value depends on the assumed formal valence of each of the elements. In this work, we introduce a new formalism that eliminates these problems. The suggested property, degree of unsaturation, can be calculated from the molecular formula as well as from any structural representation of a molecule corresponding to that molecular formula.

The characterization of structure by computer

Abstract A comprehensive approach to a computer-based system of structure characterization must include capabilities in both spectrum interpretation and structure generation, and a link between the two that is transparent to the user. Program CASE is undergoing development with these requirements in mind. Structure generation is based on the new concept of structure reduction, which appears to offer advantages in efficiency of execution over procedures based on structure assembly. An experimental version of the program possesses some problem-solving ability.