Validity and validation of expert (Q)SAR systems

Abstract

At a recent workshop in Setubal (Portugal) principles were drafted to assess the suitability of (quantitative) structure-activity relationships ((Q)SARs) for assessing the hazards and risks of chemicals. In the present study we applied some of the Setubal principles to test the validity of three (Q)SAR expert systems and validate the results. These principles include a mechanistic basis, the availability of a training set and validation. ECOSAR, BIOWIN and DEREK for Windows have a mechanistic or empirical basis. ECOSAR has a training set for each QSAR. For half of the structural fragments the number of chemicals in the training set is >4. Based on structural fragments and log Kow, ECOSAR uses linear regression to predict ecotoxicity. Validating ECOSAR for three 'valid' classes results in predictivity of > or = 64%. BIOWIN uses (non-)linear regressions to predict the probability of biodegradability based on fragments and molecular weight. It has a large training set and predicts non-ready biodegradability well. DEREK for Windows predictions are supported by a mechanistic rationale and literature references. The structural alerts in this program have been developed with a training set of positive and negative toxicity data. However, to support the prediction only a limited number of chemicals in the training set is presented to the user. DEREK for Windows predicts effects by 'if-then' reasoning. The program predicts best for mutagenicity and carcinogenicity. Each structural fragment in ECOSAR and DEREK for Windows needs to be evaluated and validated separately.

Predictive Capability of QSAR Models Based on the CompTox Zebrafish Embryo Assays: An Imbalanced Classification Problem

“…Only few QSAR methods are available to evaluate developmental toxicity [11,12,[48][49][50]] and the general lack of quantitative models further justifies our scope in assessing the quality and applicability of CompTox ZF model to predict outcomes on developmental endpoints. The main advantage of models such as those presented is the rapid and simultaneous toxicity prediction of numerous chemicals based on their action on development even though its applicability domain is restricted to only organic compounds.…”

Introducing Spectral Structure Activity Relationship (S-SAR) Analysis. Application to Ecotoxicology

“…As such, while, for instance, a certain set of parameters has been identified for environmental studies, i.e. bioaccumulation, chemical degradation (aqueous and gas phase), biodegradation, soil sorption, and ecotoxicity, two major aspects have been identified for QSAR analyses, namely the quality and the chemical domain of the QSAR [ 69 , 71 , 72 ].…”