Metal-Ion Optical Fingerprinting Sensor Selection via an Analyte Classification and Feature Selection Algorithm

Abstract

Accurate analyte classification remains a significant challenge in sensor technologies. We present the Analyte Classification and Feature Selection Algorithm (ACFSA), a computational tool designed to identify optimal sensor combinations from unique fingerprint patterns for analyte classification. We applied the ACFSA to a library of peptide-corona-functionalized single-walled carbon nanotubes (SWCNTs), developed as a near-infrared fluorescent, semiselective fingerprinting sensor set for detecting heavy metal ions. Inspired by natural metal-ion complexation sites, each SWCNT sensor in this library features a unique peptide sequence containing various amino acids for metal binding, revealing diverse optical response patterns to the various metal ions tested. The sensor library was further diversified using different SWCNT chiralities and photochemical modifications of the peptide coronae. The ACFSA was applied to the screening data of the fluorescence response of the 30 resulting SWCNT-peptide sensors to five metal-ion analytes. Through iterative dimensionality reduction and rational sensor selection, the algorithm identified the optimal fingerprinting sensors as a minimal two-sensor set with a 0.02% classification error. The final output of the ACFSA is thus an analyte classifier that serves as a unique analyte fingerprint pattern for the selected sensors. The developed peptide-SWCNT system serves as an effective proof-of-concept, illustrating the potential of our platform as a generally applicable tool for fingerprinting analytes and optimal sensor set selection in other sensor-analyte screening experiments.