A New Framework Infers Causal Brain Circuits Without Direct Neuronal Intervention - General science

Researchers have introduced Interventional Dynamical Causality, a framework that uses delay-embedding analysis and Interventional Embedding Entropy to estimate intervention-level causal relationships from observational time-series data.

A research team from Fudan University, the Chinese Academy of Sciences and the University of Tokyo has proposed a mathematical framework for identifying cause-and-effect relationships in complex systems without physically manipulating them. Published in The Innovation, the approach is called Interventional Dynamical Causality, or IntDC.

The method addresses a limitation of established techniques such as Granger causality, transfer entropy and convergent cross mapping. Those methods evaluate whether the history of one variable helps predict another, but can produce misleading links when an unmeasured factor influences both variables.

IntDC instead seeks to determine whether an intervention on one variable would alter another. To estimate this from observational data, the researchers used delay-coordinate embedding, a technique from nonlinear dynamics that reconstructs a system’s state space from time-series observations. They argue that the recorded history of an affected variable retains a mathematical imprint of the perturbations caused by a genuine driver.

The framework quantifies this relationship with a metric called Interventional Embedding Entropy, or IEE. According to the study, IEE can estimate the direction and strength of causal information flow without requiring prior knowledge of the system’s equations or a real experimental intervention.

In computational tests involving coupled logistic maps, Hénon maps and chaotic neural networks, IEE rejected connections created by shared drivers that did not directly link the tested variables. The metric also remained stable under changes in recording length, measurement noise, embedding dimensions and coupling strength.

The researchers applied the method to several real-world datasets. In calcium-imaging recordings from the nematode Caenorhabditis elegans, it was used to reconstruct directed functional relationships among motor, sensory and interneurons. The team also applied it to regional COVID-19 infection records from Japan to estimate transmission routes and to transcriptomic data to examine regulatory relationships involving circadian genes.

The authors say that IntDC and IEE could support causal-network reconstruction in neuroscience, systems biology and computational medicine when direct intervention is impractical or cannot be performed. The study, “Deciphering interventional dynamical causality from non-intervention complex systems,” was published in The Innovation.

Sources: Neuroscience News.