Multiplexed Identification of Bacterial Biofilm Infections Based on Machine-Learning-Aided Lanthanide Encoding

Pathogenic biofilms are up to 1000-fold more drug-resistant than planktonic pathogens and cause about 80% of all chronic infections worldwide. The lack of prompt and reliable biofilm identification methods seriously prohibits the diagnosis and treatment of biofilm infections. Here, we developed a machine-learning-aided cocktail assay for prompt and reliable biofilm detection. Lanthanide nanoparticles with different emissions, surface charges, and hydrophilicity are formulated into the cocktail kits. The lanthanide nanoparticles in the cocktail kits can offer competitive interactions with the biofilm and further maximize the charge and hydrophilicity differences between biofilms. The physicochemical heterogeneities of biofilms were transformed into luminescence intensity at different wavelengths by the cocktail kits. The luminescence signals were used as learning data to train the random forest algorithm, and the algorithm could identify the unknown biofilms within minutes after training. Electrostatic attractions and hydrophobic-hydrophobic interactions were demonstrated to dominate the binding of the cocktail kits to the biofilms. By rationally designing the charge and hydrophilicity of the cocktail kit, unknown biofilms of pathogenic clinical isolates were identified with an overall accuracy of over 80% based on the random forest algorithm. Moreover, the antibiotic-loaded cocktail nanoprobes efficiently eradicated biofilms since the nanoprobes could penetrate deep into the biofilms. This work can serve as a reliable technique for the diagnosis of biofilm infections and it can also provide instructions for the design of multiplex assays for detecting biochemical compounds beyond biofilms.