CITS: Nonparametric Statistical Causal Modeling for High-Resolution Neural Time Series

A team of researchers including Rahul Biswas, SuryaNarayana Sripada, and Somabha Mukherjee has developed CITS (Causal Inference in Time Series), a novel nonparametric statistical framework designed to uncover causal relationships in complex, high-resolution neural time series data. Published on arXiv, this method addresses a fundamental limitation in neuroscience and computational science: distinguishing correlation from causation. While traditional tools like Granger causality and the Peter-Clark algorithm rely on restrictive assumptions, CITS employs a flexible structural causal model of arbitrary Markov order and statistical tests for lagged conditional independence. The researchers proved the model's consistency under mild assumptions and demonstrated its superior accuracy across simulated benchmarks involving linear, nonlinear, and recurrent neural network dynamics.

In a significant practical application, the team used CITS to analyze large-scale neuronal recordings from the mouse visual cortex, thalamus, and hippocampus. The framework successfully identified stimulus-specific causal pathways and inter-regional hierarchies that aligned with known neuroanatomy while also revealing new functional insights. A key strength highlighted is CITS's ability to accurately map conditional dependencies within small inferred neuronal motifs. This breakthrough establishes CITS as a theoretically grounded and empirically validated tool for discovering interpretable causal networks directly from neural activity, moving beyond mere functional connectivity. The framework's design makes it broadly applicable beyond neuroscience to any domain involving causal discovery in complex temporal systems, from finance to climate science.