Deep-learning-assisted spectroscopic single-molecule localization microscopy based on spectrum-to-spectrum denoising.

Area of Science:

• Biophysics
• Microscopy
• Computational Biology

Background:

• Spectroscopic single-molecule localization microscopy (sSMLM) enables multiplexed subcellular imaging by capturing spatial and spectral data.
• Extracting accurate spectral information in sSMLM is difficult due to low signal-to-noise ratio (SNR) from limited photons and electronic noise.

Purpose of the Study:

• To develop a novel deep learning framework for accurate spectral recovery in low-SNR sSMLM data.
• To significantly improve the quality of spectral information obtained from single-molecule localization events.

Main Methods:

• Introduced a self-supervised deep learning network named spectrum-to-spectrum (Spec2Spec).
• Developed a training strategy for Spec2Spec utilizing correlated spectral information from adjacent pixels with independent noise.
• Validated the framework on simulated and experimental sSMLM data.

Main Results:

• Spec2Spec significantly suppresses noise and recovers low-SNR emission spectra.
• Achieved approximately a 6-fold improvement in SNR and a 3-fold increase in structure similarity index measure (SSIM) for single-molecule spectra.
• Facilitated 94.6% spectral classification accuracy and nearly 100% data utilization in dual-color sSMLM imaging.

Conclusions:

• The Spec2Spec framework effectively enhances spectral data quality in sSMLM.
• This advancement improves the reliability and utility of sSMLM for advanced subcellular imaging applications.
• Enables more accurate spectral classification and higher data utilization in multiplexed imaging.