Prediction of biopersistence of hydrocarbons using a single parameter

Aerobic biodegradation is an important attenuation process for petroleum hydrocarbons (PHCs) in the natural environment. It has also been frequently used in engineered systems to remediate PHC-contaminated sites. A model such as a quantitative structure property relationship (QSPR) that can predict the biodegradation rate of PHCs would be helpful prior to implementing any extensive environmental measurements and bioremediation strategies. Existing QSPRs either have a large number of predictor variables that may cause overfitting or are based on a small dataset of PHCs. The goal of this study is to develop a simple, portable QSPR that has only a few predicator variables but can accurately predict the biodegradation half-lives of a large group of PHCs. To this end, more than 500 molecular variables were screened, and candidate variables were refined by a feature selection method and fitted to biodegradation data of a group of structurally heterogeneous PHCs (n = 173). The model was established by means of hierarchical clustering and classification and regression tree algorithms, which was optimized by an internal validation procedure and validated by an external dataset. The optimal QSPR model, containing only one predictor variable (the number of bonds that do not contain hydrogen), was able to accurately predict biodegradation half-lives for a wide variety of PHCs. The internal validation test indicated an overall prediction accuracy of 93%, and predictions applied to an independent external set of 64 PHCs yielded 95% accuracy. The new model is transparent and easily portable from one user to another. [Display omitted] •A new QSPR was built for the biodegradation half-lives of hydrocarbons.•This new QSPR contains only one predictor variable.•Number of bonds that do not contain hydrogen is the most significant predictor.•This model was validated by an external test set of 64 hydrocarbons.•The model shows a prediction accuracy of 95% over the external validation set.