Paulo Paixao
Universidade de Lisboa
Portugal
Title: Prediction of plasma protein binding and the corresponding determination of the applicability domain by using an artificial neural networks ensemble
Biography
Biography: Paulo Paixao
Abstract
Purpose To develop a QSAR model, based on calculated molecular descriptors and an Artificial Neural Networks Ensemble (ANNE), for the estimation of plasma protein binding (as fraction of unbound drug in plasma - fup) of drugs in human, rat, dog and monkey plasma, as well as the assessment of the applicability domain (AD) of the model. Methods A total of 680 individual fup values (75% train and 25% validation), were collected in the literature from human, rat, dog and monkey plasma concentrations. A correlation between simple molecular descriptors for lipophilicity, ionization, size and hydrogen bonding capacity and fup data was attempted by using an ANNE. Results A degradation of the correlations was observed for predicted values with high uncertainty, as judged by the standard deviations of the ANNE outputs. Based on this, a “cut-off” SD<0.0857 was establish to consider that a particular drug is inside the AD of the model. Similar statistics were observed between the train and validation group of data, when inside the AD, with correlations between the observed values and the predicted average ANNE values, of 0.951 and 0.854, respectively. 82% of the drugs were well predicted with diference of less than 0.2 in the validation group of data, again when inside the AD (93% in the train dataset). Conclusions This model may be a valuable tool for simulation and prediction in early drug development, allowing the insilico estimation of fup in different pre-clinical models and in the human, that may be used for PBPK purposes.