A\nUniversal Method for Fingerprinting Multiplexed\nBacteria: Evolving Pruned Sensor Arrays via Machine Learning-Driven\nCombinatorial Group-Specificity Strategy

Array-based\nsensing technology holds immense potential for discerning\nthe intricacies of biological systems. Nevertheless, developing a\nuniversal strategy for simultaneous identification of diverse types\nof multianalytes and meeting the diagnostic needs of a range of multiclassified\nclinical diseases poses substantial challenges. Herein, we introduce\na combination method for constructing sensor arrays by assembling\ntwo types of group-specific elements. Such a method enables the rapid\ngeneration of a library of 100 sensing units, each with dual bacterial\ntargeting capabilities. By employing a three-step screening strategy\noptimized by machine learning algorithms, various optimal five-element\narrays were rapidly obtained for diverse clinical infectious models.\nMoreover, the pruned arrays successfully identified disparate mixing\nratios and quantitative detection of clinically prevalent bacterial\nstrains. Optimized through nine multiclassification algorithms, the\ntop-performing multilayer perceptron (MLP) model demonstrated impressive\nrecognition capabilities, achieving 100% accuracy for diagnosing clinical\nurinary tract infection (UTI) and 99.4% accuracy for clinical sepsis\ndetection in the test models we collected. Such a combinatorial library\nconstruction and screening process should be standard and provides\ninsights into successfully generating powerful high-recognition sensor\nelements and configuring them into highly discriminative mini-sensor\narrays.