Neuroscientists have long been able to record the electrical chatter of neurons with remarkable precision, but making sense of the resulting mountain of timing data has remained a stubbornly manual, subjective exercise. A team of researchers has now built an automated tool that could change that, converting a half-century-old analytical technique into a fast, quantitative pipeline for probing how neural networks respond to injury, drugs, and disease. The work, published in the journal Neuroinformatics, centers on a deceptively simple question: what can the shape of a correlogram tell us about the brain, and how can we extract that information without bias?
Correlograms have been a cornerstone of neuroscience analysis since the 1960s. The concept is straightforward. When researchers record from neurons, whether through microelectrode arrays (MEAs), electroencephalography (EEG), or calcium imaging, they reduce the raw signals to lists of firing times called rasters. To understand the relationship between two neurons or signal channels, scientists construct a correlogram: a histogram of the relative timing between events in one signal and events in another, with one signal designated as the reference. When normalized, the correlogram describes the probability of observing a comparison spike at time t, given that a reference spike occurred at t = 0. A flat, uniform distribution suggests the two signals are independent of one another. Peaks indicate temporal relationships, whether from direct synaptic communication, shared input, or synchronized network activity.
The problem, according to the team led by Casey Erin Adam, Shatha J. Mufti, and colleagues working with Riyi Shi, is that interpreting these distributions has traditionally required a human eyeballing each plot. The challenges compound quickly: a recording with N signals produces N² cross- and autocorrelograms per analysis window, meaning researchers frequently face thousands of plots per experiment. In practice, most labs either examine a small subset by eye, summarize correlogram shapes qualitatively, or skip shape analysis entirely in favor of inferring functional connectivity from correlogram peaks—a practice that carries a well-known statistical hazard. Correlation in spike timing, after all, does not establish causation. Two neurons can appear linked because one drives the other, but they can also appear linked because both receive common input, because a drug induced synchronization, or because of shared network rhythms.
The new MATLAB algorithm sidesteps the connectivity question altogether. Instead of trying to guess which neuron causes which to fire, it quantifies three shape properties of every correlogram in a recording and tracks how those properties shift over time or across experimental conditions. The first metric is uniformity: the algorithm applies a chi-squared test, well suited to binned distributions, to determine whether each correlogram is statistically flat. A uniform correlogram signals independence between the two signals; a nonuniform one flags a temporal relationship worth noting, without any assumption about its mechanism. The second metric is peak count, which serves as a proxy for the diversity of firing patterns. A comparison signal with highly stereotyped timing relative to the reference produces one dominant peak; richer, more varied interspike structure yields multiple peaks. To count peaks reliably, the algorithm first smooths each correlogram using quadratic regression, then identifies peaks with MATLAB’s findpeaks function, accepting only peaks whose prominence exceeds the value expected from a uniform distribution.
The third metric addresses firing order. By calculating the area of the correlogram lying left of t = 0, the algorithm quantifies whether the comparison signal tends to fire before or after the reference signal—a directional signature that persists even though the analysis never claims one causes the other. Together, the three metrics capture signal dependence, firing pattern complexity, and firing order, dimensions of network dynamics that the authors argue are common to all recording techniques, from single-cell imaging to population-level EEG. That generality is a key selling point: because the metrics do not depend on the biophysical specifics of a particular recording modality, they can in principle be compared across length scales and platforms that are otherwise difficult to bridge.
To validate the approach, the researchers turned to a platform called TBI-on-a-Chip, developed in their laboratory, in which primary neuronal networks are cultured on microelectrode arrays and subjected to controlled injuries. This in vitro system allows the team to apply blast pressure waves or impact injuries to living neural cultures while continuously monitoring how cell firing, protein expression, and morphology change, with a degree of experimental access impossible in a living brain. Traumatic brain injury is known to enhance neuronal excitability, alter connectivity, and raise long-term risks of neurodegeneration, depression, and epilepsy, making it an ideal test case for a tool designed to detect subtle shifts in network dynamics.
The algorithm was tested against three distinct perturbations in MEA recordings: exposure to bicuculline methiodide, a drug that blocks inhibitory GABA receptors and typically triggers bursts of synchronized, seizure-like activity; pH shock, an acute chemical stressor; and physical impact injury. In all three cases, the automated metrics described meaningful changes in signaling. The outputs correctly bridged changes observed in individual signal pairings to changes across the entire population of recorded signals, and the results agreed with previously published findings. Critically, the approach preserved information about heterogeneous network responses rather than averaging them away—a common casualty of population-level summaries.
Technically, the script divides a recording into user-defined analysis regions, in the validation experiments using seven-minute segments that balance event density against temporal resolution. Regions that are too short produce sparse, unreliable correlograms; regions that are too long inflate computational cost and blur dynamics over time. Users control the correlogram range and bin width—for the study recordings, ±1 second with 1-millisecond bins—and can set a sparsity threshold to exclude correlograms built from too few events, whose metrics would reflect statistical artifacts rather than biology. The algorithm then computes all three metrics for every signal pair in every region, plots the resulting distributions, and classifies peak counts and peak timings on user-selected scales, including conversion of peak locations to frequencies when rhythmic activity is of interest.
The researchers emphasize that their tool is not intended to replace connectivity analyses but to complement them with an alternative that avoids mistaking correlation for causation. Functional connectivity inferred from correlogram peaks can identify likely synapses, but nonuniform correlograms can arise from synapse-free synchronization, shared input, or other correlated activity, and no amount of timing analysis alone can guarantee a physical connection without additional measurements such as imaging. By quantifying uniformity, peak structure, and distribution asymmetry instead, the algorithm extracts information that is robust across recording types while remaining agnostic about underlying mechanisms. The trade-off, the authors acknowledge, is less direct insight into specific connections unless the analysis is paired with additional quantification or measurements.
The implications extend beyond traumatic brain injury research. Because raster-based analysis underlies virtually every technique for measuring neuronal firing, the algorithm could be applied to drug screening, studies of epilepsy and neurodegeneration, and comparisons between in vitro cultures and in vivo recordings. The team notes that fundamental responses are often similar between in vitro and in vivo platforms, but that linking results across the two has been hampered by the lack of analysis frameworks that adapt to different recording techniques and length scales. A common currency of correlogram shape metrics offers one possible bridge.
The MATLAB script has been released openly on GitHub and in the supplementary materials of the publication, reflecting a broader push toward reproducible, automated analysis in neuroscience. For a field drowning in high-dimensional electrophysiology data, a tool that turns thousands of manual visual judgments into seconds of computation—and does so without importing hidden assumptions about causality—may prove a quietly transformative addition to the neuroscientist’s toolkit. What began as a technique pioneered on paper histograms in the 1960s has now been brought squarely into the era of automated, quantitative network science.
Subject of Research: Automated quantification of correlogram shape to analyze neuronal firing dynamics in microelectrode array recordings from a traumatic brain injury (TBI)-on-a-Chip platform
Subject of Research: Medicine
Article Title: Quantifying Correlogram Shape to Analyze Neuronal Firing Dynamics Recorded in TBI-on-a-Chip
Article References: Adam, C. E., Mufti, S. J., Martinez, J., Rogers, E. A., Dalolio, M., Krishnan, N., Beauclair, T., & Shi, R. (2026). Quantifying Correlogram Shape to Analyze Neuronal Firing Dynamics Recorded in TBI-on-a-Chip. Neuroinformatics, 24(2), Article 24. https://doi.org/10.1007/s12021-026-09770-9
Image Credits: AI Generated
DOI: 10.1007/s12021-026-09770-9
Keywords: correlogram, crosscorrelation, neuronal firing dynamics, microelectrode array, TBI-on-a-Chip, traumatic brain injury, MATLAB algorithm, network dynamics, chi-squared test, firing order
Cite Scienmag News
APA MLA Chicago
Cassandra Pierce. (September 7, 2026). Measuring Correlogram Shape Reveals Neuronal Firing Patterns in TBI-on-a-Chip. Scienmag. https://scienmag.com/measuring-correlogram-shape-reveals-neuronal-firing-patterns-in-tbi-on-a-chip/
Cassandra Pierce. “Measuring Correlogram Shape Reveals Neuronal Firing Patterns in TBI-on-a-Chip.” Scienmag, 7 September 2026, https://scienmag.com/measuring-correlogram-shape-reveals-neuronal-firing-patterns-in-tbi-on-a-chip/. Accessed 7 September 2026.
Cassandra Pierce. “Measuring Correlogram Shape Reveals Neuronal Firing Patterns in TBI-on-a-Chip.” Scienmag. September 7, 2026. https://scienmag.com/measuring-correlogram-shape-reveals-neuronal-firing-patterns-in-tbi-on-a-chip/
Copy citation Download RIS
Tags: automated neuroscience data analysisautomated neuroscience data processingbrain injury simulationcorrelogram shape analysiselectrophysiological data processingelectrophysiological data visualizationmicroelectrode array dataneural firing pattern analysisneural network connectivityneural network response to injuryneural signal analysisneuroinformatics toolsneuron firing synchronizationneuronal connectivity metricsneuronal firing pattern analysisneuronal response to injuryquantitative neural signal analysisspike timing correlationspike train correlationTBI-on-a-chip models
