Peter C. Jurs

Applications of Computers to Chemical Problems

Professor Jurs and his students are developing and using computer-assisted methods to investigate relationships that link molecular structures of organic compounds with their chemical properties or biological activities.

Computer-assisted methods can be used to investigate relationships between molecular structure and chemical properties for large sets of organic compounds. This approach is data intensive and inductive because a large training set of compounds with known property values is the starting point for each study. Such studies involve the following steps: a) graphical entry and storage of the structures to be used to develop the model along with their experimental property values, b) three-dimensional molecular modeling using either molecular mechanics or molecular orbital approaches, c) molecular structure descriptor generation, (d) analysis of the descriptors for utility, that is, feature selection, by interactive, user-guided methods or simulated annealing or genetic algorithm methods, and e) the development of quantitative predictive models using multivariate statistical methods or computational neural networks. Structural descriptors include topological, geometric, electronic, and hybrid representations of the molecular structures. Studies of computational neural networks involve feed-forward, fully-connected three-layer neural networks for quantitative prediction of properties. They also include the use of self-organizing maps and similar neural networks for classification tasks. Once a structure-property model has been developed, it can be used to predict the property for new compounds that were not part of the training set.

Examples of application areas where this methodology has been used include gas chromatography retention indices, high-performance liquid chromatography retention times, ion mobility spectrometry reduced mobility constants, Henry's law constants, normal boiling points, aqueous solubilities, supercritical carbon dioxide solubilities, surface tensions, vapor pressures, and autoignition temperatures.

Similar computer-assisted methods also can be used to investigate the area of quantitative structure-activity relationships. This area of study encompasses attempts to rationalize the connections between the molecular structures of organic compounds and their biological activities. Techniques drawn from chemical structure information handling, physical organic chemistry, pattern recognition, multivariate statistics, conformational analysis, molecular orbital theory, data analysis using computational neural networks, genetic algorithms, and other areas form this new approach to structure-activity studies. Such research has many areas of application, and work is proceeding in Professor Jurs' group on several different types of biologically active compounds.

The analysis of spectral data by computer-assisted methods is another area of current interest. The data are from fiber-optic sensors which have a number of polymeric coatings that alter the fluorescence spectra of analytes that are sensed. The raw data are a set of fluorescence time series, and the objective is to develop computational neural networks that can use these data to identify the analyte compounds. Various types of neural networks are being used to generate such classifiers. Several types of features are derived from the time series data for presentation to the neural networks. The overall objective is to develop fiber-optic sensors and analysis software that can identify and quantitate gas-phase analytes with high accuracy. This is a collaborative project, and the Jurs group is working on the analysis software part of the project.

The computer techniques involved in these studies of molecular structure and chemical properties or biological activity have been combined into an interactive computer software system called ADAPT. The software is designed to support the computations necessary for these types of studies in a seamless system that makes the studies convenient to perform. Students in the Jurs research group work both on applying the software system to new chemical properties or biological activities and in extending the system's capabilities with new algorithms and methods. The new methods focus on the development of new molecular structure descriptors, new methods for feature selection, and new methods for development of the predictive models.