Fig 1.
A) Matlab-based graphic user interface. B) Overview of MENGA’s outputs. Left: Printed result report on Matlab workspace. Right: summary statistics and scatter analysis plot for gene vs. image cross-correlation and image auto-correlation.
Fig 2.
MENGA workflow and functional blocks.
MENGA requires as input an image to process and a list of genes of interest to compare with. The image needs to be a 3D matrix already in standard MNI format (1mm, 181x217x181 or 182x217x182). The genes are selected from a pre-processed version of the Allen human brain database, where each protein is identified with a unique transcript profile. Once the image is resampled into the mRNA donor spaces, MENGA performs between-subject correlation analysis for both mRNA and image data. As last step, the cross-correlation analysis between mRNA and imaging is calculated for all the selected genes.
Fig 3.
Tested neuroimaging modalities.
A) Serotonin system—[11C]CUMI and [11C]WAY100635 PET imaging are used as 5HT1A receptor markers (agonist and antagonist respectively). The protein density maps are images of non-displaceable specific binding, BPND [unitless] and volume of distribution, VT [ml/cm3], respectively. B) Myelin system–MR-based absolute myelin water content modality (, [unitless]) is used as myelin density marker. C) Dopamine system–[11C]Raclopride PET imaging is used as D2/D3 receptor maker; [123I]FP-CIT SPET imaging is used as dopamine transporter (DAT) marker; [18F]FDOPA PET imaging refers to dopamine synthesis capacity (DDC, dopa decarboxylase enzymatic rate). The parametric maps are images of non-displaceable specific binding, BPND [unitless], index of uptake IU [unitless] and rate of net uptake, Ki [ml/cm3], respectively.
Table 1.
MENGA applications in representative imaging modalities.
Fig 4.
MENGA applications in representative imaging modalities with matched genomic data.
The figure reports the genomic-imaging cross-correlation values along with genomic auto-correlation. In the results matrix, each element represents the correlation between couples of donors (for the auto-correlation) or between one donor’s mRNA levels and the image sampled in the same donor’s space (for the cross-correlation). A) Serotonin system: [11C]CUMI and [11C]WAY100635 PET imaging vs. 5HT1A receptor (HTR1A) mRNA expression; B) Myelin system: myelin water content MR imaging vs. myelin-associated oligodendrocyte basic protein (MOBP) mRNA expression C) Dopamine system: [11C]Raclopride PET imaging vs. dopamine D2 receptor (DRD2) mRNA expression; [123I]FP-CIT SPET imaging vs dopamine transporter (DAT) mRNA expression; [18F]FDOPA PET imaging vs. dopa decarboxylase (DDC) mRNA expression.
Fig 5.
Summary statistics of MENGA applications in representative imaging modalities with matched and mismatched genomic data.
Cross-correlation results as A) univariate Pearson’s R2, B) multivariate correlation coefficient and C) chance-likelihood associated to the R-squared are reported for all the sets of image vs. genomic applications, both for matched (green bars) and mismatched (red bars) cases. Univariate correlations are presented as mean and variability across donors.
Fig 6.
MENGA applications in representative imaging modalities with mismatched genomic data.
The figure reports the genomic-imaging cross-correlation values along with genomic auto-correlation. In the results matrix, each element represents the correlation between couples of donors (for the auto-correlation) or between one donor’s mRNA levels and the image sampled in the same donor’s space (for the cross-correlation) in three different cases of images vs. mismatched genomic data integration. A) [11C]CUMI PET imaging vs. D2 receptor (DRD2) mRNA expression; B) [123I]FP-CIT SPET imaging vs 5HT1A receptor (HTR1A) mRNA expression; C) myelin water content MR imaging vs. dopa decarboxylase (DDC) mRNA expression.