The arrow‐of‐time in neuroimaging time series identifies causal triggers of brain function

Abstract

Moving from association to causal analysis of neuroimaging data is crucial to advance our understanding of brain function. The arrow‐of‐time (AoT), that is, the known asymmetric nature of the passage of time, is the bedrock of causal structures shaping physical phenomena. However, almost all current time series metrics do not exploit this asymmetry, probably due to the difficulty to account for it in modeling frameworks. Here, we introduce an AoT‐sensitive metric that captures the intensity of causal effects in multivariate time series, and apply it to high‐resolution functional neuroimaging data. We find that causal effects underlying brain function are more distinctively localized in space and time than functional activity or connectivity, thereby allowing us to trace neural pathways recruited in different conditions. Overall, we provide a mapping of the causal brain that challenges the association paradigm of brain function.

Multilevel irreversibility reveals higher-order organisation of non-equilibrium interactions in human brain dynamics

“…However, the highlighting of individual tuples is most meaningful when there is a strong intuition about the nature of the interaction, which can be only be expected in low-dimensional parcellations where ROIs are clear, functionally segregated brain areas. Finally, we note that our measure is undirected within the tuple, meaning we cannot identify the direction of information flow as one can with classical measures of causality [60, 61] or some approaches to the AoT [7, 8]. However, we note that the AoT represents directed flow between states and not variables, meaning it is not a direct measure of causality, but instead capturing a distinct, but related, phenomena in interacting dynamics.…”

“…As discussed in the main manuscript, the entropy production rate of a system that is out of equilibrium is equal to the information-theoretic evidence for the AoT quantified by, where Γ is a trajectory, Γ , is its time-reversal and P (Γ) is the ‘path-probability’, the probability of observing that specific trajectory [34, 18]. Calculating this divergence quantifies the distance from equilibrium [8, 25, 33].…”

“…First, the hierarchical organisation of positions in state-space, that results from asymmetrical transition probabilities, has been linked to the dynamic hierarchical organisation of brain regions [7, 25, 26]. Second, the AoT has been interpreted as inducing a ‘causal flow’ in the system where some regions emerge as information ‘sources’ and others as ‘sinks’ with these relationships identifiable from irreversibility analysis [7, 8]. These studies for quantifying non-equilibrium in the brain approximate the global evidence for the AoT in time-series using techniques such as estimating transitions between coarse-grained states [4], with time-shifted correlations [5], machine learning [6] or with model-based approaches [7–10].…”

Static and dynamic fMRI-derived functional connectomes represent largely similar information

“…Measures or models that are sensitive to the temporal order of time points are called dynamic , while measures that are insensitive to temporal order are measures of static FC. Given that the information flow in the brain is causally organized in time ( Bolton, Van De Ville, Amico, Preti, & Liégeois, 2023 ; Cole, Ito, Bassett, & Schultz, 2016 ), dynamic connectivity models could be more informative in terms of understanding brain function and investigating brain-behavior associations.…”

Constructing Causal Knowledge Representation with the Common Model of Cognition

“…Causal inference has become increasingly relevant in science due to a formalism known as do-calculus developed by Judea Pearl (2008Pearl ( , 2021Pearl & Mackenzie, 2018) to represent statistical results as probabilistic causal outcomes. Currently, causal inference techniques are being used in the fields of statistics (Tian & Pearl, 2000;Barenboim & Pearl, 2014;Pearl, 2019), artificial intelligence (Pearl & Mackenzie, 2018;Scholkopf, Locatello, Bauer, Ke, Kalchbrenner, Goyal, & Bengio, 2021), physics (Wolfe, Spekkens, & Fritz, 2019;Wolfe, Schmid, Sainz, Kunjwal, & Spekkens, 2020;Chaves, Moreno, Polino, Poderini, Agresti, et al, 2021), neuroscience (Weichwald, Meyer, Özdenizci, Schölkopf, Ball, & Grosse-Wentrup, 2015;Bolton, Van De Ville, Amico, Preti, & Liegeois, 2022), and cognitive science (Weichwald et al, 2015;Thagard, Larocque, & Kajić, 2021), but have yet to be used extensively in cognitive modelling. This thesis presents a new theoretical framework for cognitive representations of causation with five cognitive models of causal reasoning based upon Pearl's work in causal inference (Pearl, 2000(Pearl, , 2001(Pearl, , 2008Pearl & Mackenzie, 2018).…”