Quantitative structure-activity relationship (QSAR) modeling of human blood: Air partitioning with proper statistical methods and validation

Blood: Air partition coefficient (BA_pc) is important in assessing toxicokinetics of chemicals. Since very few experimental data are available, quantitative structure-activity relationship (QSAR) models based on calculated molecular descriptors can be useful in estimating BA_pc. Since all descriptors used in the analysis are computed strictly from structure, they can be applied to any chemical, real or hypothetical. In this article, we report models for BA_pc estimation using four methods, including stepwise ordinary least-squares regression, which is commonly used in QSAR studies but often results in an inflated 'naïve' q², over-representing the predictive ability of the model. The models developed using proper statistical techniques had q² values of 0.825 and 0.841, and may be used to reliably predict BA_pc values for new compounds that are structurally similar to those upon which the models are based. The models developed using improper techniques had associated q² values, as computed using naïve methods, of 0.920 and 0.934, severely overstating their actual quality.