Dual deep learning network enables data-efficient two-color single-molecule localization microscopy with colorimetry camera

Multicolor single-molecule localization microscopy (SMLM) enables analysis of nanoscale spatial relationships of multiple targets but suffers from technical complexity. CC-STORM simplifies optics by integrating a colorimetry camera with MLE-Locator and NCI-Classifier, yet its low signal-to-noise ratio and undersampled point spread function necessitate >70% data rejection to maintain <2% crosstalk, severely limiting efficiency. To overcome these limitations, we developed CC-DeepSTORM, a deep learning framework combining CC-DeepLoc for localization and CC-DeepSeparator for color-separation. In simulations, CC-DeepLoc enhanced the Jaccard index and localization precision by 1.8× and 2× over MLE-Locator. For single-color experimental data, CC-DeepSeparator reduced data rejection from ∼70% to ∼40% while achieving 1% crosstalk (matching state-of-the-art). In two-color experiments, CC-DeepSTORM significantly outperformed CC-STORM in localization and color-separation, enabling efficient, high-resolution multicolor SMLM for broader biological applications.