What does the mean mean? A simple test for neuroscience

Abstract

Trial-averaged metrics, e.g. in the form of tuning curves and population response vectors, are a basic and widely accepted way of characterizing neuronal activity. But how relevant are such trial-averaged responses to neuronal computation itself? Here we present a simple test to estimate whether average responses reflect aspects of neuronal activity that contribute to neuronal processing in a specific context. The test probes two assumptions inherent in the usage of average neuronal metrics: 1. Reliability: Neuronal responses repeat consistently enough across single trial that their average response template remains recognizable to downstream regions. 2. Behavioural relevance: If a single-trial response is more similar to the average template, it will be easier for the animal to identify the correct stimulus or action. We apply this test to a large publicly available data set featuring electrophysiological recordings from 71 brain areas in behaving mice. We show that single-trial responses were less correlated to the average response template than one would expect if they represented discrete versions of the template, down-sampled to the number of spikes in the single trial. Moreover, single-trial responses were barely stimulus-specific — they could not be clearly assigned to the average response template of one stimulus. Most importantly, better-matched single-trial responses did not more accurately predict behaviour for any of the recorded brain areas. We conclude that in this data set, average responses do not seem particularly relevant to neuronal computation in a majority of brain areas, and we encourage other researchers to apply similar tests when using trial-averaged neuronal metrics.

In vivo Quantification of Neural Criticality and Complexity in Mouse Cortex and Striatum in a Model of Cocaine Abstinence

“…3F) in nearly all mice, and in those few cases where avalanches were still observed, an increase in MAE was concomitant with the jitter applied, consistent with previous work (45). Thus, the functional relevance of avalanche fractal structure and sub millisecond-level timing in “signal-less” quiescent recordings 15 min long (57) is indeed curious, at the very least.…”

Investigation of neural functional connectivity in thick acute mouse brain slices with novel multi-region 3D neural probe arrays

“…Not only that, but in vivo RMS noise tends to be upwards of 40 μV, drastically limiting the number of units discoverable, due to the LFP being a summation of all electrical signals at the location one has implanted an electrode. The resultant neurons of focus, “non-dark” and “type 2 dark” neurons end up skewing all analyses(36) based on population averages, if even the murine or human brain “cares” about arbitrary average firing rates within experimenter-defined windows(45). Ultimately this “sparse” but skewed activity representation in vivo can generate essentially statistical artifacts of “encoding” such as “grandmother”-like cells (46), units explicitly tuned to concepts like numbers (47), and brain-wide reward-responsive neurons.…”

Brain signatures indexing variation in internal processing during perceptual decision-making

“…18 However, it is also possible that subsets of trials exhibit brain activity that is not well represented by the average. 19 , 20 Indeed, brain activity measured with EEG is both spatially and temporally variable during a task, with at least some of this variability likely stemming from meaningful variation in internal processing rather than simply noise. 8 , 9 , 10 , 11 Additionally, averaging across all trials might eliminate the ability to observe distinct forms of stimulus-driven activity (i.e., the ERP) that are engaged differentially on individual trials or contexts but that are, nonetheless, relevant for behavioral performance.…”