Detecting causality in neural spike trains using a new technique

Understanding the brain’s functional architecture is a fundamental challenge in neuroscience. The connections between neurons ultimately dictate how information is processed, transmitted, stored, and retrieved, thus forming the basis of our cognitive functions. Scientists often study neuronal signaling by recording the brief electrical pulses they generate over time, often referred to as ‘spike trains.’

Because of their bursty and aperiodic nature, inferring causal relationships between spike trains recorded from different neurons remains a significant challenge. Traditional causality detection methods, such as Granger causality and transfer entropy, require regularly sampled time series, make assumptions of linearity, or need very large datasets. This makes them less suited for the more chaotic, nonlinear dynamics inherent in biological systems like the brain. Scientists have struggled to find effective, model-free methods to directly analyze causality relationships in neural networks and other nonlinear systems with similar characteristics.

In a recent study, a research team led by Assistant Professor Kazuya Sawada from the Department of Information and Computer Technology, Faculty of Engineering at Tokyo University of Science (TUS), Japan, successfully developed a new technique to detect causality in neural spike trains. Their paper, co-authored by Professor Tohru Ikeguchi from TUS and Associate Professor Yutaka Shimada from Saitama University, was published online in Volume 112, Issue 1 of Physical Review E on July 28, 2025.

The team’s method builds upon a known framework called convergent cross mapping (CCM), which is effective for analyzing causality between nonlinear time series data. However, conventional CCM cannot be applied to time series data with irregular sampling intervals (like spike trains). To address this, the researchers first used a technique to reconstruct a system’s state space from the interspike intervals (ISIs), which is the most usual way of storing data from spike train recordings. They then devised a new approach to establish the temporal correspondence between different ISI time series.

Combining these two methods resulted in a new way of determining causality in spike trains. The core idea is to calculate the accuracy of predictions one makes on a given spike train based on data from the others, focusing specifically on whether this accuracy increases or remains low as more data is provided. “The method proposed in our paper differs from previous ones in that it can be directly applied to spike sequences and identify causal relationships in data generated by complex, nonlinear systems that cannot be represented by simple rules,” highlights Dr. Sawada. The causality between neurons can be detected from easily observable spike trains, thereby estimating their connectivity. 

To test the efficacy of their method, the researchers applied it to a well-studied mathematical model of neurons with known causal connections. Through numerical experiments, they demonstrated that the proposed approach accurately detected bidirectional, unidirectional, and non-existent coupling between neurons. It proved effective even in the presence of weak coupling with internal noise, a common feature of biological systems.

By providing a new tool for inferring neural connectivity from spike train data, this research opens the door to a more granular understanding of how information is processed in the brain. “The connections between brain neurons are not yet fully understood, and causality detection methods can be used to estimate not only structural and anatomical connections but also effective connections,” explains Dr. Sawada. “If we could clarify the nature of such effective connections within the brain, it would contribute to a better understanding of disorders and mental illnesses caused by neuronal connections, potentially paving the way for new therapies.” The study may have implications in understanding the mechanism behind epilepsy, and in the diagnosis of schizophrenia and bipolar disorder that could be caused by an imbalance between excitatory and inhibitory neurons.

Dr. Sawada explained that the detection of causality focused only on two or three neurons in their study and emphasized that future research will focus on extending the method to larger networks. This will help in exploring the study’s applicability to more complex neural dynamics. Worth noting, given how common time series data similar to spike trains are seen in other contexts-known as ‘point processes’-the findings of this study could also guide the development of new techniques for evaluating causality in fields such as finance, seismology, and logistics.