DeepCINAC: a deep-learning-based Python toolbox for inferring calcium imaging neuronal activity based on movie visualization

Abstract

Two-photon calcium imaging is now widely used to indirectly infer multi neuronal dynamics from changes in fluorescence of an indicator. However, state of the art computational tools are not optimized for the analysis of highly active neurons in densely packed regions such as the CA1 pyramidal layer of the hippocampus during early postnatal stages of development. Indeed, the reliable inference of single cell activity is not achieved by the latest analytical tools that often lack proper benchmark measurements. To meet this challenge, we first developed a graphical user interface allowing for a precise manual detection of all calcium transients from detected neurons based on the visualization of the calcium imaging movie. Then, we analyzed our movies using a convolutional neural network with an attention process and a bidirectional long-short term memory network. This method reaches human performance and offers a better F1 score than CaImAn to infer neural activity in the developing CA1 without any user intervention. Overall, DeepCINAC offers a simple, fast and flexible open-source toolbox for processing a wide variety of calcium imaging datasets while providing the tools to evaluate its performance. Highlights:-A user-friendly GUI allows experimentalists to label easily cell activity from the visualization of calcium imaging movies -A neural network trained on labeled calcium imaging data can faithfully infer neuronal activity of few hundreds of neurons.-The toolbox offers a simple, fast and flexible method for processing a wide variety of calcium imaging data.

The rapid developmental rise of somatic inhibition disengages hippocampal dynamics from self-motion

“…Cell type prediction was done using the DeepCINAC cell type classifier (Denis et al, 2020). Briefly, a neuronal network composed of a convolutional neuronal network (CNN) and Long Short Term Memory (LSTM) was trained using labelled interneurons, pyramidal cells and noisy cells to predict the cell type using 100 frames long movie patches centered on the cell of interest.…”

“…We examined 4 age groups, postnatal days 5 to 6 (P5-6, n=6 mice, 11 movies, see Movie 1), P7-8 (n=9 mice, 17 movies, see Movie 2), P9-10 (n=7 mice, 14 movies, see Movie 3) and P11-12 (n=5 mice, 7 movies, see Movie 4). The contours of imaged neurons and calcium events were determined using Suite2P and DeepCINAC respectively (Denis et al, 2020;Pachitariu et al, 2017).…”

“…To this aim, we identified GABAergic neurons by the expression of a red reporter tdTomato in GAD67Cre pups virally infected with AAV9-FLEX-tdTomato and AAV1.hSyn.GCaMP6s (Figure 3A). In addition, we used these imaging experiments to train a cell classifier that could infer interneurons in the absence of any reporter with a precision of 91.2% (Denis et al, 2020) "labelled" and "inferred" GABAergic neurons were combined into a single group referred to as "interneurons'' in the following (Figure S4). As illustrated in a representative raster plot from a P5 mouse, both pyramidal cells (green) and interneurons (red) seemed activated during movement (vertical grey lines, Figure 3A).…”

Database and deep learning toolbox for noise-optimized, generalized spike inference from calcium imaging

“…The reliability of spike inference obviously depends on the recording quality of the calcium imaging data. To improve data quality of ΔF/F signals, future work should focus on the reduction of movement artifacts and neuropil contamination by both experimental design 50,57 and extraction methods [36][37][38][39] , including the correct estimation of the F 0 baseline despite unknown background fluorescence. In the long term, the development of more linear calcium indicators 58 and especially the acquisition and integration of more specific ground truth, e.g., for inhibitory interneuron subtypes 59 , will enable quantitative spike inference for an even broader set of experimental conditions.…”

“…S9d-e). Neuropil contamination is often very difficult to distinguish from somatic calcium signals and particularly severe when neurons are tightly packed and densely labeled 1,[36][37][38][39] (Fig. S9b).…”