Power in network neuroscience

Abstract

Connectomics has become a prime method for studying brain circuitry. The success of these investigations hinge on the capacity to detect the effects present in the data – that is, statistical power. Here, we discuss four main facets of power in connectomics: sample size, variance, effect size and network topology. We discuss how these factors (1) shape the overall power of connectome studies and (2) give rise to ‘differential power’ within individual studies, rendering some network effects easier to detect than others. We discuss how power impacts our understanding of brain networks and their circuitry and review strategies to optimally work with – not around – power in connectomics.

The Genetic Architectures of Functional and Structural Connectivity Properties within Cerebral Resting-State Networks

“…First, the definition of RSN used in this univariate GWAS study reduces voxel-level diffusion and functional information to one phenotype by averaging potentially variable connectivity patterns. This could unequally affect more variable higher-order RSNs compared to less variable unimodal RSNs, which would lead to differential statistical power across the RSNs studied here (Helwegen, Libedinsky, & van den Heuvel, 2022). This could explain why the most significant results are observed in the visual and somatomotor network.…”

Structural and functional connectivity reconstruction with CATO - A Connectivity Analysis TOolbox

“…Such strict connectome reconstructions are beneficial as false positive connections are estimated to have more impact on network measures than false negatives (Zalesky et al, 2016). They do however have a negative effect on the level of sensitivity of finding more complex fiber bundles, i.e., at the price of having more false negatives (Helwegen et al, 2022). Exploring the effect of reconstruction parameters showed various ways in which researchers can tune the sensitivity and specificity profile of their reconstructions (Supplementary Figure 3).…”