(238b) Quantitative Structure-Property Relationship Model for Prediction of IGC50 Toxicitiy Potentials

The effect of chemicals on animals can be multifold and requires laborious experimentation to test chemicals for their toxicity. Quantitative structure-property relationship (QSPR) models offer a substitute for experimentation and thereby save resources and avoid undue animal experimentation. However, this task is too complex to be addressed by a simple QSPR model. Therefore, as a first step, the toxicity of the chemicals on small microbial organisms is used as an indication of the effect on animals. The toxicity data used in this study involved the microbe T. pyriformis as the test organism. The 50% growth impairment concentration, which is denoted as IGC₅₀, was used as an indication of the toxicity potential of the chemicals.

This study was completed under a QSPR challenge conducted in 2009 by Case studies on the Development and Application of in-Silico Techniques for Environmental hazards and Risk assessment (CADASTER). The objective was to develop successfully a QSPR model on known IGC₅₀ values and apply it to an external test set of unpublished IGC₅₀ values. In this study, we present our new quantitative structure-property relationship (QSPR) model for predicting IGC₅₀ values. Several descriptors were generated for the 1213 molecules in the database. A wrapper approach, which involves differential evolution combined with neural networks, was implemented to find the optimum set of 20 network descriptors and the network architecture, simultaneously. Our ensemble network resulted in an RMSE value of 0.817 for the external dataset considered.