For the serotonin 1A receptor, correlation between PET binding potentials and predicted comprehensive mRNA expression was markedly higher (Spearman ρ = 0.72 for cortical surface, ρ = 0.84 for subcortical regions) than correlation of PET and discrete samples only (ρ = 0.55 and ρ = 0.63, respectively). In subcortical regions and cerebellum, different sets of genes were implicated by high spatially structured variability. 816 out of 18,686 genes exhibited a high spatial dependence accounting for more than 50% of variance in the difference of gene expression on the cortical surface. Across genes, the correlation of predicted and observed mRNA expression in leave-one-out cross-validation correlated with the strength of spatial dependence (cortical surface: r = 0.91, subcortical regions: r = 0.85, cerebellum: r = 0.84). Prediction results showed minimal bias ranging within ☐.016 (cortical surface), ☐.12 (subcortical regions) and ☐.14 (cerebellum) in units of log2 expression intensity for all genes. For validation, predicted whole-brain transcription of the HTR1A gene was correlated with WAY-100635 positron emission tomography data collected from 30 healthy subjects. Variogram models were employed in Gaussian process regression to calculate best linear unbiased predictions for gene expression at all locations represented in well-established imaging atlases for cortex, subcortical structures and cerebellum. For each of 18,686 genes, spatial dependence of transcription was assessed using variogram modelling. Microarray data of 3702 samples from 6 brain donors was registered to MNI and cortical surface space using FreeSurfer. To resolve this issue, a method allowing for the prediction of mRNA expression in the entire brain based on microarray data provided in the Allen Human Brain Atlas was developed. The quantification of big pools of diverse molecules provides important insights on brain function, but is often restricted to a limited number of observations, which impairs integration with other modalities.

Human brain mapping 2018