Connectivity analysis

In order to extract measures of the brain functional connectome we retrieved individual-level time series data for the 25 components from each participant. We calculated two measures, FC and node variance, using the FSL toolbox¹. Out of the 25 components 4 captured mostly noise. We regressed and removed the time series of these 4 components from the time series of the 21 good components. We extracted the functional connectivity measures by calculating the regularized partial correlations of the component time series with an algorithm developed by Olivier et al.². This algorithm serves to more aggressively distinguish between noise and signal via automated adjustment of the shrinkage parameter lambda (\(\lambda\))². Finally, we regressed age, age², sex, scanner, motion, SNR, and the first 20 genetic principal components from the resulting measures for each participant. For functional connectivity, this resulted in a partial correlation matrix consisting of 210 measures (\(\frac{21 * (21 - 1)}{2}\)). We obtained the measures of node variance by calculating the level of variance in the time series for each of the 21 components for each participant³ and applying the same regression as for the functional connectivity measures.

To validate the downstream results of the ICA approach, we compared the results with measures of connectivity obtained through an ROI approach. For the ROI approach we used the Schaefer parcellation with 1000 parcels⁴. We accessed the FEAT⁵ transformed data prepared by the UKB and registered all images to MNI space. We averaged the time series of the 1000 parcels in the Schaefer parcellation across their correspondent 17 brain networks defined by Yeo et al.⁶. We then extracted functional connectivity measures by calculating the Pearson’s correlation between the time series of the 17 brain networks. We calculated node variance across the 17 networks following the same procedure as the ICA-based stream.