QSAR modelling of adipose:blood partition coefficient : from single models to a consensus approach
Abstract
Adipose:blood partition coefficient is a key-input for models predicting the bioaccumulation and pharmacokinetics in living beings. Indeed, other organ:blood affinities can be estimated as a function of this parameter. Existing QSAR models predict adipose:blood partition coefficient by considering in vitro or in vivo rat and human data, or a combination of them. Therefore, the modeled datasets are often heterogeneous in terms of endpoint definition. To enhance the available knowledge, we performed an ex-novo selection of in vivo data to obtain a well-defined endpoint and to increase the size of existing datasets. These new datasets were used to develop QSARs as a function of an extensive exploration of chemical descriptors and algorithms based on linear and non-linear techniques. All the models were associated to applicability domains and their prediction errors were assessed with respect to experimental uncertainty. In addition, we integrated individual QSARs into a consensus model to exploit the synergy among models and improve predictivity. Our QSAR results will be implemented in VEGA (hnps://www.vegahub.eu/) and interfaced with MERLIN-Expo (hnp://merlin-expo.eul) to characterize human and environmental fate of chemicals in the framework of chemical risk assessment.