Deep learning-based spectroscopic single-molecule localization microscopy.

Area of Science:

• Biophysics
• Microscopy
• Computational Biology

Background:

• Spectroscopic single-molecule localization microscopy (sSMLM) combines nanoscopy and spectroscopy for sub-10 nm resolution and multicolor imaging.
• Deep learning offers a promising avenue for reconstructing sSMLM data to visualize subcellular structures at the nanoscale.

Purpose of the Study:

• To develop a novel deep learning computational approach for reconstructing both label-free and fluorescence-labeled sSMLM imaging data.
• To enhance the resolution and accuracy of sSMLM imaging through advanced computational methods.

Main Methods:

• A two-network-model deep learning algorithm, named DsSMLM, was developed for sSMLM data reconstruction.
• The algorithm's effectiveness was validated using diverse samples: label-free single-stranded DNA (ssDNA) fibers, fluorescence-labeled histone markers in cells, and multicolor DNA origami nanorulers.

Main Results:

• Achieved 6.22 nm spatial resolution for label-free ssDNA fiber imaging.
• Revealed chromatin distribution using histone markers and enabled multicolor imaging of nanorulers with 40 nm separation.
• Observed an 8.8% increase in single-color and up to 5.05% increase in two-color localization detection.

Conclusions:

• Demonstrated the feasibility of deep learning-based reconstruction for sSMLM imaging across various sample types.
• DsSMLM is a valuable tool for high-quality super-resolution imaging, aiding the study of DNA photophysics and cellular nanostructures.