CSLogWS Home

CSLogD Home

CSLogP Home

CSBBB Home

CSPB Home

CSGenoTox Home

CSHIA Home

Download a Pre-print
about CSGenoTox

CSGenoTox

     ...Predictor Comparison

Comparison of Commercial MI Predictors

Statistical Comparison of Ames Mutagenicity Predictors:

An approximate comparison of validation results from three different commercially available GenoTox models and CSGenoTox was made.  It was impossible to make a direct compound to compound comparison since CSGenoTox was not available in 2001, the time of a published study by Pearl et al (1). Nonetheless, the relative degree of robustness of these predictors can be assessed from results found on drugs and non-drugs.  There are two in silico methods used to compute a predictive Ames mutagenicity result: a rule-based or expert system and the QSAR approach.  The former approach, as used by Derek, is low filtering technique.  Topkat and MCase employ QSAR with the latter also using rule-based methods.  A comparison of results from the CSGenoTox validation set was made for both drugs and non-drugs as shown in the Table below. Although there were only there were only 39 drugs in CSGenoTox set, the large difference in concordance between the latter and the first three models in the Table probably does reflect the robustness of CSGenoTox.  The cross-validation (a rough measure of the expected results in external validation) on 290 drugs with CSGenoTox gave a similar high level robustness for drugs. With respect to non-drugs, the same holds true; CSGenoTox is more robust overall compared to the other models with 361 compounds used here as well as cross-validation results on 2673 compounds.