Machine learning assisted single-molecule sensing towards standard-free quantification of per- and polyfluoroalkyl carboxylic acids

Abstract Per- and polyfluoroalkyl carboxylic acids (PFCAs) are of global concern for their ubiquitous presence in the environment. However, precise quantification of PFCAs remains challenging due to the shortage of standards. Herein, with the aid of machine learning, a probe-directed nanopore based single-molecule electrochemical sensor is developed towards standard-free digital quantification of PFCAs. To correctly predict the signal without standards, a strict linear relationship (R2 > 0.9998) is established between current blockades and molecular volumes of PFCAs up to C14. Leveraging high-resolution multi-feature classification, identification accuracy reaches 100% for a broad range of PFCAs including isomers. Reliable, multiplexed quantification of PFCAs is verified in various environmental matrices, with a state-of-the-art detection limit of 0.1 nM for trifluoroacetic acid (an ultrashort-chain PFCA). The double-barriers of probe-pore interaction suggest that capture rates can be independently tuned, without comprising identification. As a proof-of-concept, a universal probe-determined calibration curve is realized experimentally for short- and medium-chain PFCAs, which is theoretically extendable to all PFCAs for standard-free quantification via nanopore engineering.