Imaging fluorescence correlation spectroscopy (FCS) measures molecular dynamics at each pixel of a camera, providing diffusion coefficient maps over an entire cross-section of a cell, tissue, or organism. It supplements the structural data obtained by imaging and provides new insights into the relation between structure and dynamics in live samples. However, until today, imaging FCS was data hungry and slow.

In this article, we demonstrate that convolutional neural networks (CNNs) can address both issues, that the required data amount can be reduced by up to a factor of 20, and that data evaluation can be sped up by orders of magnitude, rendering imaging FCS a real-time tool.

In the cover image of the March 19 issue of Biophysical Journal, we show the diffusion coefficient map of eGFP-tagged bicoid, a morphogen that controls anterior-posterior axis formation during Drosophila embryogenesis. The map was measured at nuclear cycle 14 of the embryo and shows clear differences in nucleus and internuclear spaces, thus providing information on bicoid dynamics and potentially the mechanisms of bicoid gradient formation. In our study, we demonstrate that we can calculate these maps of over 16,000 points by using CNNs within £10 s for a 2.5–50-s measurement.

The CNNs are trained on simulated data and can be applied to any measurement geometry because no specific fit models for data analysis are required. In addition, data treatment is fast and does not require expert input. In the future, CNNs will be extended to different measurement geometries and samples with more complex dynamics. You can find more information on our work at https://www.dbs.nus.edu.sg/lab/BFL/index.html.

— Wai Hoh Tang, Shao Ren Sim, Daniel Ying Kia Aik, Ashwin Venkata Subba Nelanuthala, Thamarailingam Athilingam, Adrian Röllin, and Thorsten Wohland