Implementation of Neural Networks for the Identification of Single Molecules

Analyzing complex single-molecule emission patterns with deep learning

A fluorescent emitter simultaneously transmits its identity, location, and cellular context through its emission pattern. We developed smNet, a deep neural network for multiplexed single-molecule analysis to enable retrieving such information with high accuracy. We demonstrate that smNet can extract three-dimensional molecule location, orientation, and wavefront distortion with precision approaching the theoretical limit and therefore will allow multiplexed measurements through the emission pattern of a single molecule.

The deep neural network smNet enables extraction of multiplexed parameters such as 3D position, orientation and wavefront distortion from emission patterns of single molecules.

Monitoring dynamic systems with multiparameter fluorescence imaging

A new general strategy based on the use of multiparameter fluorescence detection (MFD) to register and quantitatively analyse fluorescence images is introduced. Multiparameter fluorescence imaging (MFDi) uses pulsed excitation, time-correlated single-photon counting and a special pixel clock to simultaneously monitor the changes in the eight-dimensional fluorescence information (fundamental anisotropy, fluorescence lifetime, fluorescence intensity, time, excitation spectrum, fluorescence spectrum, fluorescence quantum yield, distance between fluorophores) in real time. The three spatial coordinates are also stored. The most statistically efficient techniques known from single-molecule spectroscopy are used to estimate fluorescence parameters of interest for all pixels, not just for the regions of interest. Their statistical significance is judged from a stack of two-dimensional histograms. In this way, specific pixels can be selected for subsequent pixel-based subensemble analysis in order to improve the statistical accuracy of the parameters estimated. MFDi avoids the need for sequential measurements, because the registered data allow one to perform many analysis techniques, such as fluorescence-intensity distribution analysis (FIDA) and fluorescence correlation spectroscopy (FCS), in an off-line mode. The limitations of FCS for counting molecules and monitoring dynamics are discussed. To demonstrate the ability of our technique, we analysed two systems: (i) interactions of the fluorescent dye Rhodamine 110 inside and outside of a glutathione sepharose bead, and (ii) microtubule dynamics in live yeast cells of Schizosaccharomyces pombe using a fusion protein of Green Fluorescent Protein (GFP) with Minichromosome Altered Loss Protein 3 (Mal3), which is involved in the dynamic cycle of polymerising and depolymerising microtubules.