Rastermap: a discovery method for neural population recordings

Abstract

Neurophysiology has long progressed through exploratory experiments and chance discoveries. Anecdotes abound of researchers setting up experiments while listening to spikes in real time and observing a pattern of consistent firing when certain stimuli or behaviors happened. With the advent of large-scale recordings, such close observation of data has become harder because high-dimensional spaces are impenetrable to our pattern-finding intuitions. To help ourselves find patterns in neural data, our lab has been openly developing a visualization framework known as "Rastermap" over the past five years. Rastermap takes advantage of a new global optimization algorithm for sorting neural responses along a one-dimensional manifold. Displayed as a raster plot, the sorted neurons show a variety of activity patterns, which can be more easily identified and interpreted. We first benchmark Rastermap on realistic simulations with multiplexed cognitive variables. Then we demonstrate it on recordings of tens of thousands of neurons from mouse visual and sensorimotor cortex during spontaneous, stimulus-evoked and task-evoked epochs, as well as on whole-brain zebrafish recordings, widefield calcium imaging data, population recordings from rat hippocampus and artificial neural networks. Finally, we illustrate high-dimensional scenarios where Rastermap and similar algorithms cannot be used effectively.

Pan-cortical 2-photon mesoscopic imaging and neurobehavioral alignment in awake, behaving mice

“…4c (2), bottom of Rastermap), followed by transient suppression (see Suppl Movie S4). Examination of individual “raw” dF/F traces confirmed that these differences were not due to artifacts of z-scoring or superneuron averaging (Stringer et al, 2023, bioRxiv) in the PC and Rastermap sorted displays (not shown).…”

“…This method for one-dimensional nonlinear embedding works by sorting and clustering of ROIs (i.e. neurons) by similarity of neural activity patterns across multiple timescales, using k-means clustering, sorting by asymmetric similarity as defined by peak cross-correlations at non-negative timelags, and upsampling of individual cluster activities in principal components space to allow sorting within clusters (Stringer et al, 2023, bioRxiv). Here, sorting with Rastermap revealed stark clustering of neurons with clearly distinguishable activity patterns that exhibited heterogeneous relationships to the behavioral primitive arousal/movement variables of walk speed, whisker motion energy, and pupil diameter (Fig.…”

“…simultaneously imageable, either contiguous or non-contiguous, regions; Allen et al, 2017; Hattori et al, 2019; but see Yu et al, 2020). Recent advancements in cranial window preparations have enabled imaging over large portions of the dorsal cortex (Kim et al, 2016; Stringer et al, 2023, bioRxiv). However, simultaneous 2-photon imaging over a large cortical area including both dorsal and lateral regions, particularly in a preparation that allows for simultaneous imaging of the major primary sensory cortices (auditory, visual, and somatosensory) and frontal choice/motor areas (M1, M2) in awake, behaving mice has not been previously shown.…”

“… a) Rastermap (top), first principal component (PC1; middle), and second principal component (PC2, bottom) sorting of normalized, rasterized, neuropil subtracted dF/F neural activity for 3096 cells from a single 20 min duration example dorsal mount 2-photon imaging session. Each row in the display corresponds to a “superneuron”, or average activity of 50 adjacent neurons in the Rastermap sort (Stringer et al, 2023, bioRxiv). Activity for each superneuron is separately z-scored, then displayed as low activity in blue, intermediate activity in green, and maximum activity level in yellow (standard viridis color look-up table).…”

“…Numbers at left (R0-R7) indicate Rastermap motif numbers, selected manually “by-eye” for this session, and horizontal dashed white lines on raster indicate separation between neighboring Rastermap motifs. Each row in the display corresponds to a “superneuron”, or average of 50 adjacent neurons in the Rastermap sort (Stringer et al, 2023, bioRxiv). Colored, dashed boxes indicate alignment of high-level behaviors with neural activity epochs from the same example session.…”