Biomarker sensing has been traditionally limited by large sample requirements, limited sensitivity, and low degree of multiplexing. Multiplexing issues arise even with a single-molecule fluorescence approach due to factors such as spectral crowding. Previously, we developed FRETfluors, novel DNA-based FRET constructs with differing spectroscopic signatures that offer high-order multiplexed sensing at the single-molecule level, up to 27 at once. By rationally designing the constructs, we utilize the sensitivity of fluorophore photophysics to the local physicochemical environment to access characteristic emission properties which enrich the feature space in which we classify these tags. We present advancement to the biomarker sensing problem by pairing FRETfluors with affinity-based sensing and data-driven inference of biomolecular identity. FRETfluors are covalently conjugated to affinity binders such as antibodies to report single-molecule binding events to biomarkers. Because the tags themselves encode optical “barcodes,” many different biomarkers can be monitored at the same time while remaining agnostic to the choice of binder, presenting scalability and versatility. To further unlock the information latent inflorescence time traces, we develop a machine learning pipeline that takes in single-photon data to learn classifying features. The resulting classifiers identify tags and binding states, suppress background, and triage artifacts automatically, increasing accuracy and opening a path toward higher throughput biosensing at the single-molecule level.
Febryanto et al. (Sun,) studied this question.