Neural Networks for Quantitative Structure-Property Relations (QSPRs)
Research Team: Yoram Cohen, Denise Yaffe, Francesc Giralt, Alex Arenas and Gabriela Espinosa
In order to assess the existing and potential environmental impact of chemical
contaminants it is necessary to predict their likely distribution in the environment. The
distribution of chemicals in the environment is governed by their physicochemical
and transport properties. However, given the large number of present and future chemicals
which may be of concern, it is infeasible to measure the required physicochemical
properties of all those chemicals. Therefore, property prediction methods are necessary.
Unfortunately, existing prediction methods are either cumbersome to use or do not apply
over a sufficiently wide range of chemical functionalities. Therefore, in this
program the use
of neural networks for designing a set of prediction tools is being investigated. The goal
is to develop a neural network prediction system which will allow one to estimate basic
physicochemical properties such as boiling points, vapor pressure, Henry's law constants, octanol-water partition coefficients, aqueous solubility,
aqueous infinite activity coefficients and others. The tools generated
by this research will be directly applicable for use in models of contaminant transport
and exposure assessment models.
The initial phase of this project focused on the demonstrating the Neural Networks approach for the prediction of boiling points of organic compounds using both a back-propagation network and a newly developed Fuzzy ARTMAP model.
Last update:
11/09/2004
