Independent component analysis (ICA) of resting state fMRI data

The coregistered 2D resting state functional data from 14 monkeys having 37 runs were temporally concatenated, and group ICA was performed on the rsfMRI signals from the entire brain. Runs that qualified for the analysis varied per animal as follows: 2 runs (8 animals), 3 runs (3 animals), and 4 runs (3 animals). Standard procedures in GIFT software (Calhoun et al. 2001) were followed to obtain spatial ICA maps and their corresponding time series. Next, we used a dual regression technique to obtain subject-specific component maps, along with their associated time series (Beckmann et al. 2009). In spatial ICA, each component generally denotes a network which is spatially independent from other networks. For ICA decomposition of the entire brain, we chose 15 components based on previous reports of studying the brain of other species using ICA (Beckmann et al. 2005; Damoiseaux et al. 2006; Smith et al. 2009; Hutchison et al. 2011; Belcher et al. 2013; Hori, Schaeffer, Gilbert, et al. 2020) and empirical evidence of synchronized BOLD fluctuations in specific regions of the SM brain using different range of components such as 10, 15 and 20 (more details in Supplementary Material).

The reliability of the estimated group ICs was obtained using GIFT software’s ICASSO method (Himberg et al. 2004). The algorithmic and statistical stabilities were investigated by running the algorithm 10 times with different initial values or/and with differently bootstrapped data sets, respectively. We also performed single subject ICA from individual monkeys with parameters matched to group ICA data to examine the robustness of the group ICs at an individual subject level.