Fig. 6: Simulating implications of sex differences for rsfMRI analyses and diagnostic classifier performance.

a rsFC abnormalities in the DMN vary with the sex composition of the sample. rsFC abnormalities are summarized in terms of the mean t-statistic in mass univariate contrasts of each rsFC within the 77-node DMN, with boxplots depicting the distribution of the mean t-statistic over 1000 bootstrapped subsamples of n = 140 MDD subjects and n = 70 healthy controls, selected at random without replacement from the full study sample, across seven different sex compositions (x-axis). b Color map showing the neuroanatomical distribution of the connectivity abnormalities summarized in (a). For each functional parcel, significant t-statistics (P < 0.05, FDR q < 0.1) were summed across the 77 DMN nodes, averaged across 1000 iterations, and plotted on brain surfaces, with warm colors indicating significant connectivity increases. c Schematic for 2:1 hold-out paradigm for training and testing elastic-net regularized general linear models (GLMs) in males only, females only, or in all subjects. d, e Area under the ROC curve (AUC) for EN-GLMs trained and tested on all subjects (black lines) compared to (d) males only (blue line) and to (e) females only (red line). f Color map showing summed absolute loading weight for all connectivity features related to a given ROI, mapped onto all 360 cortical ROIs, averaged over 100 iterations of each EN-GLM for the male-only models (left), the female-only models (center), and the sex-blind models (right).