Fig. 1: Schematic of methodology.

a Defining thalamic seeds. Throughout the thalamic volume (orange area), a set of seeds 1.75 mm apart are defined. Only those which were consistently localised (see Methods for details) across participants were used as seed points (black: consistent seeds; white: inconsistent seeds). b Thalamic seed connectivity. Probabilistic tractography was conducted from each seed to 250 left hemisphere cortical targets based on a random parcellation. Connectivity between thalamic seeds and cortical regions was averaged across participants to produce a 921-by-250 seed-by-cortical target matrix of thalamocortical connectivity. Connectivity to cortical regions was scaled to the unit interval using a sigmoid transformation. c Assigning transcriptomic data to thalamic seeds. Voxelwise estimates of post-mortem gene expression for 2228 genes with differential expression in brain tissue were extracted for the thalamus. For each gene, each seed point is assigned the expression value of the voxel it is located within to produce a 921-by-2228 seed-by-gene matrix. As above, each gene’s expression levels were normalised to the unit interval according to a scaled sigmoid. d Joint decomposition. The seed-by-cortical connectivity and seed-by-gene matrices were concatenated and decomposed into a set of orthogonal factors by Principal Component Analysis (PCA). From the resulting principal components (PCs), the first PC (PC1) explained 30.3% of the variance in the concatenated data matrix. For each PC, the scores, one per thalamic seed, describe the representation of each component in the thalamus and the loadings, describe the contribution to the PC of connectivity strength and gene expression level for each of the cortical regions and genes, respectively.