Aerobic biodegradation is one of the most important elimination pathways of organic contaminant from the environment. This study mainly focuses on the aquatic environment and established a large dataset based on the standardized experimental guidelines such as OECD 301 and 302 methods.
Various machine learning algorithms, chemical representations, and categorical encoding methods were compared to obtain the best model performance. Two models were obtained, i.e., classification and regression models.
The classification model uses SMILES strings (converted to fingerprints) as the input and the class (0 or 1) as the output. Only ready biodegradation data with time of 28 and principles of closed bottle test, closed respirometer, and CO2 evolution were considered.
The regression model uses SMILES strings (converted to fingerprints), time (day), guideline (e.g., OECD 301F), principle (e.g., closed respirometer), endpoint (e.g., ready or inherent), and reliability (e.g., 1 or 2) as the inputs. The biodegradation percentages are the output.