Machine learning predicts the impact of antibiotic properties on the composition and functioning of bacterial community in aquatic habitats.

In the past decades, hundreds of antibiotics have been isolated from microbial metabolites or have been artificially synthesized for protecting humans, animals and crops from microbial infections. Their everlasting usage results in impacts on the microbial community composition and causes well-known collateral damage to the functioning of microbial communities. Nevertheless, the impact of different antibiotic properties on aquatic microbial communities have so far only poorly been disentangled. Here we characterized the environmental risk of 50 main kinds of antibiotics from 9 classes at a concentration of 10 μg/L for aquatic bacterial communities via metadata analysis combined with machine learning. Metadata analysis showed that the alpha diversity of the bacterial community increased only after treatment with aminoglycoside and β-lactam antibiotics, while its structure was changed by almost all tested antibiotics. The antibiotic treatment also disturbed the functions of the bacterial community, especially with regard to metabolic pathways, including amino acids, cofactors, vitamins, xenobiotics and carbohydrate metabolism. The critical characteristics (atom stereocenter count, number of hydrogen atoms in the antibiotic, and the adipose water coefficient) of antibiotics affecting the composition of the bacterial community in aquatic habitats were screened by machine learning. The key characteristics of antibiotics affecting the function bacterial communities were the number of hydrogen atoms, molecular weight and complexity. In summary, by developing machine learning models and by performing metadata analysis, this study provides the relationship between the properties of antibiotics and their adverse impacts on aquatic microbial communities from a macro perspective. The study also provides guidance for the rational design of antibiotics.