[NMF Analysis] Using a sliding window approach, we can identify time-varying functional connectivity, using simple correlation across the anatomical regions estimated within each time window.
We can then concatenate unique edges over time into a n by k array, where n is the number of unique edges and k the number of time windows sampled. Using sparsity constrained non-negative matrix factorisation, we can decompose the full set of connectivity edges into variably expressed subnetworks, and their time varying expression profiles.

[Left] We can represent factors derived from NMF as subnetworks that are variably expressed over time. Networks derived from a four-factor factorisation are shown here.

[Middle] The factors' subnetwork expression varies over time, but shows persistent differences across genotypes.

[Right] Full time varying connectivity matrices are shown here, with time points corresponding to the dashed line in the middle plot.