This article describes the first application of a generic (empirical) Bayesian analysis of between-subject effects in the dynamic causal modeling (DCM) of electrophysiological (MEG) data. It shows that (i) non-invasive (MEG) data can be used to characterize subject-specific differences in cortical microcircuitry and (ii) presents a validation of DCM with neural fields that exploits intersubject variability in gamma oscillations. We find that intersubject variability in visually induced gamma responses reflects changes in the excitation-inhibition balance in a canonical cortical circuit. Crucially, this variability can be explained by subject-specific differences in intrinsic connections to and from inhibitory interneurons that form a pyramidal-interneuron gamma network. Our approach uses Bayesian model reduction to evaluate the evidence for (large sets of) nested models-and optimize the corresponding connectivity estimates at the within and between-subject level. We also consider Bayesian cross-validation to obtain predictive estimates for gamma-response phenotypes, using a leave-one-out procedure. Hum Brain Mapp 37:4597-4614, 2016. © The Authors Human Brain Mapping Published by Wiley Periodicals, Inc.
Keywords: Bayesian model reduction; classification; dynamic causal modeling; empirical Bayes; gamma oscillations; neural fields; random effects.
© 2016 The Authors Human Brain Mapping Published by Wiley Periodicals, Inc.