The relationship between structural connectivity (SC) and functional connectivity (FC) in the human brain can be studied using magnetic resonance imaging (MRI). However many of the underlying physiological mechanisms and parameters cannot be directly observed with MRI. This limitation has motivated the recent use of various computational models meant to bridge the gap. However their absolute and relative explanatory power and the properties that actually drive that power remain insuf␣ciently characterized. We performed an extensive comparison of seven mainstream computational models predicting FC from SC. We investigated the extent to which simulated FC could predict empirical FC. We also applied graph theory to the entire set of simulated and empirical FCs in order to further characterize the relationships between the models and the MRI data. The com- parison was performed at three different spatial scales. We found that (i) there were signi␣cant effects of scale and model on predictive power; (ii) among all models, the simplest model, the simultaneous autoregressive (SAR) model, was found to consistently perform better than the other models; (iii) the SAR also appeared more ‘central’ from a graph theory perspective; and (iv) empirical FC only appeared weakly correlated with simulated FCs, and was featured as ‘peripheral’ in the graph analysis. We conclude that the substantial differences existing between these computational models have little impact on their predictive power for FC and that their capacity to predict FC from SC appears to be both moderate and essentially underlined by a simple core linear process embodied by the SAR model.
