Code
Works on 2D and 3D data without needing 3D labels. The GUI handles manual labeling and curation, can train new models on your own images, and can denoise, deblur or upsample them. Developers have integrated it into CellProfiler, ImagePy, ImJoy, aPeer and Napari. Try it on hugging-face.
Cellpose
Generalist deep-learning segmentation that works across a wide range of image types.
GPU-accelerated for speed and accuracy. It detects spikes, uses them for drift correction (essential in most recording setups), then clusters them with novel graph-based methods. Kilosort4 is written in Python with an easy-to-use GUI.
Kilosort
The most widely used spike sorter for Neuropixels probes, and it works on others too.
Runs on standard workstations, faster than real time, recovering about twice as many cells as previous methods. Routinely detects ~25,000 cells at once from standard resonant-scanning microscopes, including axons, boutons, dendrites and spines. Extensive GUI for exploring the data.
Suite2p
Complete two-photon pipeline: registration, cell detection, traces and spike times.
Applies SVD to facial motion for a 500-dimensional summary of the mouse’s movements, and tracks keypoints with a deep network. A 1D convolutional network then predicts neural activity from either, twice as accurately as previous approaches.
Facemap
Extracts orofacial behavior from face videos and predicts neural activity from it.
Combines manifold discovery with clustering, using both instantaneous and cross-correlations between clusters, then upsamples the sorting so each neuron sits at its most correlated position. Finds VR-corridor sequences that t-SNE and UMAP miss. Runs in a notebook, its GUI, or inside Suite2p.
Rastermap
Visualizes large-scale neural activity with a one-dimensional manifold embedding.
Datasets
Visual response dataset
29,000 neurons in mouse primary visual cortex responding to up to 65,000 natural images.
Visual learning dataset
50,000+ neurons recorded simultaneously in mouse visual cortex as mice undergo unsupervised and task learning in virtual reality.
Facemap dataset
Spontaneous activity from 50,000+ neurons in mouse visual and sensorimotor cortex, with simultaneous face camera recordings and a keypoint tracking training set.
Neural responses to oriented stimuli
20,000+ neurons in mouse primary and higher-order visual cortex.
Spontaneous neural activity in V1
10,000 neurons in visual cortex during spontaneous behaviors.
Neural responses to natural images
10,000 neurons in primary visual cortex responding to 2,800 natural images.
Eight-probe Neuropixels recordings
Recordings during spontaneous behaviors, collected by Nicholas Steinmetz, from Stringer, Pachitariu et al 2019.
V1 responses to drifting gratings
10,000 neurons in mouse V1 during drifting gratings.
Cellpose training data
70,000 segmented cells and other objects, the training set behind the Cellpose algorithm.
Kilosort4 ephys simulations
Simulated electrophysiology recordings used to benchmark spike sorting in the Kilosort4 paper.
