Spatial collinearity constrains multivariate molecular-enriched network estimation.
bioRxiv : the preprint server for biology June 12, 2026 Timothy Lawn, Johan Nakuci, Steve Cr Williams et al.
Spatial overlap among brain receptor maps derived from PET imaging can distort analyses that model multiple receptors together. Using test-retest fMRI data, the authors show that as more receptors are included in a multivariate model, the reliability of the resulting functional connectivity networks decreases, and this degradation is driven by collinearity among the receptor maps. A univariate approach, modeling each receptor independently, produces more reliable networks and, in a study comparing LSD to placebo, better captured the known role of the 5HT-2A receptor. Spatial collinearity is a fundamental constraint on multivariate molecular-enriched network estimation, and univariate modeling is recommended as a more robust default.