iiinova — where your digital self keeps evolving

Carsen Stringer @computingnature.bsky.social · 12/02/2026

4) Suite2p has a GUI with many visualizations and tools - we recommend starting there to run the algorithm and explore the data.

Suite2p GUI features including correlation visualization, automated ROI merging, motion-correction movie viewing, registration metrics, and Rastermap visualization of neural activity

210

Carsen Stringer @computingnature.bsky.social · 12/02/2026

3) We developed an ROI detection algorithm which implements matrix decomposition with L0-sparsity constraints in both space and time.

ROI detection. a, Frames are binned in time (left) and then high-pass filtered in space and time (right). b, Time-averaged thresholded activity from convolved square templates of different sizes. c, Initialization of new ROI with best fitting square template from b. d, Initial time trace corresponding to the dot product of the template in c with each movie frame. e, Iterative template refinement for four example ROIs. f, Example region with many cells detected and their spatial overlaps (top left), temporal overlaps (top right) and active frames crossing the threshold (bottom). g, Detected ROIs divided by the sizes of their initial square templates.

210

Carsen Stringer @computingnature.bsky.social · 12/02/2026

2) Motion correction is very important in recordings of awake, behaving animals. We perform *fast* non-rigid motion correction (170 frames per sec).

a, Reference image is computed by iterative re-alignment of a random subset of frames. b, Rigid motion offsets are computed by correlation of whitened reference and frame, i.e. phase correlation. c, Calculated X/Y rigid motion offsets. d, Non-rigid motion offsets computed by dividing image into overlapping blocks, computing phase correlation in each block, smoothing across blocks, taking the optimal shifts per block and upsampling them via bilinear interpolation. e, Timing as a function of number of frames in recording - Hz estimated from slope

130

Carsen Stringer @computingnature.bsky.social · 12/02/2026

1) Suite2p is a complete pipeline for processing large optical recordings of neurons, including motion correction, ROI detection, classification, and manual curation.

Suite2p pipeline. a, Example volume imaging data. b, X/Y motion correction (left: before, right: after). c, ROI detection (left) and trace extraction with neuropil extraction based on surrounding pixels (right). d, Classification of ROIs into somatic and non-somatic (i.e. dendrites and other neuropil). e, The graphical user interface (GUI) can be used for manual curation and results visualization. f, Multi-plane recording of over 100,000 neurons using a standard Thorlabs Bergamo2 scope (see Methods for exact configuration). Somatic ROIs for example plane shown in different colors. g, Neural activity raster from recording in f, with neurons sorted by Rastermap, colored by the corridor that the mouse is in

130

Carsen Stringer @computingnature.bsky.social · 12/02/2026

🔬🧠 Releasing the 1.0 version of #Suite2p and THE PAPER w/ @marius10p.bsky.social! Now with GPU acceleration. Want to use Suite2p but don’t have 100,000 neuron recordings? We show you how to get those with a standard 2p microscope #neuroscience #imaging #neuroAI www.biorxiv.org/content/10.6...

210136

Carsen Stringer @computingnature.bsky.social · 19/11/2025

#sfn2025 interested in how eye movements 👀 versus orofacial movements influence 🐭 visual cortex activity 🧠? Check out poster R8 by @atika-syeda.bsky.social NOW!

0211

Carsen Stringer @computingnature.bsky.social · 12/02/2025

Also includes a super-generalist segmentation model, "cyto3" which works well out-of-the-box on many image types.

"cyto3" model segmentations for fluorescent microscopy images, brightfield and other techniques, on cells from various tissues and cell lines, nuclei, yeast cells, and bacteria; model segmentation in yellow, ground truth in purple

140

Carsen Stringer @computingnature.bsky.social · 12/02/2025

#Cellpose 3 paper now out. Not all images are perfect. Restore your images with Cellpose3 to get better segmentations, w/ @marius10p.bsky.social www.nature.com/articles/s41...

613749

Carsen Stringer @computingnature.bsky.social · 18/01/2025

"Robustness of working memory to prefrontal cortex microstimulation" - impressive expts and interesting findings from @jsoldadomagraner.bsky.social et al www.biorxiv.org/content/10.1...

Multi-electrode electrical microstimulation to probe dlPFC delay period activity during working memory. A, Electrical microstimulation (uStim) protocol. An Utah array was implanted in the dlPFC, area 8Ar. Subthreshold stimulation was applied to either individual electrodes or pairs of electrodes in the array, creating a variety of spatial microstimulation patterns (see Methods). Neural population activity was simultaneously recorded from all channels in the array after microstimulation. B, Memory-guided saccade task. After the monkey acquired fixation, a target was briefly presented (for 100ms) at one out of four possible spatial locations. This was followed by a variable delay period (1.25-2s, see Methods), after which the goe cue signaled the monkeys to saccade to the remembered target location. Microstimulation was applied during the delay period, using 3-4 different spatial patterns on each session (see Methods). tpre: 50 ms before uStim onset; tpost: 50 ms after uStim ends; tend: time of go cue. C, Neural activity of one representative unit for each of the four target angle conditions. Shown are firing rates (FR) computed from no-microstimulation trials (n=22) in an example session. This unit is spatially tuned to the different target locations during the delay period (activity is stronger for the rightward targets, and in particular for the 315° target, than for the other targets). Tuning strength at tpost = 15 spikes per second (sp/s), computed as max FR (target 315°) - min FR (target 225°). In no-microstimulation trials, tpost marks the same time as in microstimulation trials, but no stimulation was delivered in this case. Responses have been smoothed for visualization with a Gaussian filter with a standard deviation of 40 ms. Monkey W, session 20220810, unit 30. D, Tuning strengths across all units at tpost (monkey W, n=10 sessions; monkey S, n=1 session). Triangles indicate mean tuning strength across all units and sessions (7±6 sp/s, mean±std, monkey W; 7…

0193

Carsen Stringer @computingnature.bsky.social · 12/01/2025

Simulations with spatial connectivity replicated several of the properties we observed in the neural recordings, such as a spatial dependence of top correlated neuron pairs, and top PCs which were globally spread across cortex.

Simulation with local connectivity and neural recordings had similar properties such as: little-to-no relationship of average correlations of neurons with distance; top-correlated neurons are nearby spatially; top PCs have weights which are approximately uniformly spread out throughout the simulation / recording window.

190

Carsen Stringer @computingnature.bsky.social · 12/01/2025

Global emergent activity modes persisted in simulations with sparse, clustered or spatial connectivity.

Simulations of connectivity with sparse binary connections, clustered connections, and locally-structured connections. These simulations resulted in power-law exponents around 0.7 when the global connectivity was sufficiently high.

190

Carsen Stringer @computingnature.bsky.social · 12/01/2025

An estimate of the dynamics matrix of the data (using DMD) revealed mostly real eigenvalues, further supporting symmetric dynamics.

schematic of dynamic mode decomposition (DMD), a way to estimate the dynamics matrix empirically from data. DMD finds real eigenvalues for the dynamics from a symmetric random matrix and complex eigenvalues from a non-symmetric random matrix, as expected. In the neural data, we see primarily real eigenvalues, similar to the symmetric random matrix.

150

Carsen Stringer @computingnature.bsky.social · 12/01/2025

The model also predicted that higher PCs have longer timescales, which was true in the data.

auto-correlograms of principal components in V1, brainwide and hippocampus recordings; in each the timescales decay as a function of the PC index

160

Carsen Stringer @computingnature.bsky.social · 12/01/2025

But in hippocampus, we observed exponents around 0.5 that did not change after shuffling, suggesting that hippocampal activity is closer to completely independent neurons.

hippocampal neural activity with a power-law decay of 0.5, which is smaller than the exponent from a symmetric matrix => the variances are more similar across PCs, suggesting more independent neural activity

1120

Carsen Stringer @computingnature.bsky.social · 12/01/2025

In V1 and brainwide ephys recordings we observed that the PC variances decayed as a power-law with exponents of 0.7-0.85, consistent with symmetric, critically-normalized simulations.

neural recordings from V1 and whole-brain, along with the estimated variance of the principal components of the neural activity, which decay with a power-law of exponent 0.7-0.85, which is close to the exponents from symmetric random matrices.

170

Carsen Stringer @computingnature.bsky.social · 12/01/2025

Long timescales and large principal components (PCs) can be produced by a dynamical system with random connectivity and independent stochastic inputs, if the connectivity matrix is critically-normalized.

Simulations of a linear dynamical system with symmetric random connectivity, and non-symmetric random connectivity, produce covariance structure which has an eigenspectrum with a powerlaw decay

1141

Carsen Stringer @computingnature.bsky.social · 12/01/2025

What if… spontaneous neural activity 🧠 reflects the baseline rumblings of a brainwide dynamical system initialized for learning? We find that the rumblings have macroscopic properties like those emerging from linear symmetric, critical systems 🧵 #neuroscience #neuroAI www.biorxiv.org/content/10.1...

schematic of neural recordings from mouse V1, whole-brain, and hippocampus; neural activity traces from the population, showing more correlated activity in V1 and whole-brain recordings versus more decorrelated activity in hippocampus

931083

Carsen Stringer @computingnature.bsky.social · 15/11/2024

We wrote a review on analysis methods for large-scale neural recordings www.science.org/stoken/autho... @marius10p.bsky.social #neuroscience 🧪🧠 Anything we missed? Reply w/ your fav method!

large-scale neural recording of 50,000 neurons, showing many diverse activity patterns, and also what the cells look like at various zooms of the maximum projection image of the recording FOV

317260

Carsen Stringer @computingnature.bsky.social · 14/11/2024

Janelia is recruiting AI scientists and theory fellows to collaborate with experimental scientists on hard biological problems - these positions have a considerable level of independence and very competitive salaries. Join our great community! Ai.hhmi.org janelia.org/theoryfellow #neuroscience #AI

AI at hhmi, link ai.hhmi.org; theory fellows, link janelia.org/theoryfellow

17046

Carsen Stringer @computingnature.bsky.social · 12/11/2024

👋 calcium imaging + image/data processing lab here

Calcium imaging in mouse visual cortex, processed data with time traces of neural firing

110