3D single-molecule detection using semiconductor nanowires and deep learning

Semiconductor nanowires are used in biosensing due to their ability to enhance the fluorescence of bound fluorescently labeled molecules. This enhancement is influenced by nanowire diameter, refractive index, and the fluorophore’s wavelength. With a large surface-to-volume ratio and field of view, nanowires also enable quantifying molecular concentrations as low as 10 fM and single-molecule binding dynamics. However, while the position of a bound molecule along the nanowires’ z-axis has not yet been available, assessing it could enable the use of nanowires to probe molecular distribution in 3D.

Here, we extend nanowire-based single-molecule detection to include the molecule’s axial position along the nanowire length (2-3 µm). We use nanowires to engineer the fluorophore’s point-spread function, relying on diffraction and fluorescence enhancement dependence on binding position. We utilize numerical solutions of Maxwell's equations to simulate the fluorescence enhancement, followed by image creation. These images were used to train convolutional neural networks to predict binding positions with sub-100-nm resolution. High prediction accuracy suggests applicability for real microscopy data, while advanced neural networks could enable tracking of 3D molecular motion.