Data analysis and alternative modelling of MITI-I aerobic biodegradation

An alternate representation of the biodegradation of chemicals (biochemical oxygen demand (BOD) amount instead of BOD-based degradation percentage) allows the derivation of an improved QSAR model. The structural descriptors for this model are related to biodegradation mechanisms. A linear group contribution model of 99 variables was derived from a training set of 1190 chemicals. The squared correlation coefficients for self-prediction and for 20%-off cross-validation were 0.83 and 0.69, respectively. Additional external validation with 62 reported in the literature chemicals resulted in 91% overall correct classification, although ready biodegradable molecules were too few and predicted poorly (two out of five correct).