Spectroscopic Single-Molecule Discrimination of BODIPY Fluorophores with Deep Learning

Abstract

Photoactivatable dyes with spectrally-resolvable single-molecule fluorescence must be developed to enable simultaneous imaging of structurally-distinct biomolecules with nanometer precision in densely-labeled samples. To fill this crucial gap, we synthesized a library of 16 borondipyrromethene (BODIPY) derivatives sharing a common functional group compatible with photoactivation but differing in a single substituent required for spectral tuning. We demonstrated that the nature of this particular substituent controls the spectral position of the emission maximum and the relative intensity of a vibronic shoulder in their single-molecule emission spectra. We implemented a convolutional neural network (CNN) to discriminate individual molecules with different substituents from these subtle spectral differences. This deep-learning algorithm can distinguish the two components of up to 41 pairs of dyes and the three components of up to 16 triads of dyes with a classification accuracy greater than 0.7 from the unique spectral signatures encoded in their single-molecule fluorescence.

Keywords: BODIPY, deep learning, fluorescence, photoactivatable fluorophores, single-molecule localization microscopy, single-molecule spectroscopy