QSAR classification models for predicting affinity to blood or liver of volatile organic compounds in e-health

PONZONI, IGNACIO; CRAVERO, FIORELLA; DIAZ, MÓNICA F.; MARTINEZ, MARÍA JIMENA

Granada

Conferencia; 5th International Work-Conference on Bioinformatics and Biomedical Engineering (IWBBIO 2017); 2017

Universidad de Granada

In this work, we present classification Quantitative Structure-Activity Rela-tionship (QSAR) models for characterization of molecules affinity to blood or liver for volatile organic compounds (VOCs), using information provided from log Pliver measures for VOCs. The models are computed from a dataset of 122 molecules. As first phase, alternative subsets of relevant molecular descriptors related to the target property are selected by using feature selec-tion methods and visual analytics techniques. From these subsets, several QSAR models are inferred by different machine learning methods. These models allow classifying a new compound as a molecule with affinity to blood, to liver or equal affinity to both. The model with the highest perfor-mance correctly classifies 72.13% of VOCs and has an average receiver op-erating characteristic area equal to 0.83. As conclusion, this QSAR model can predict the medium affinity of a VOC, which can help in the develop-ment of physiologically based pharmacokinetic computational models re-quired in e-health.