Fig. 4: Effect of targeted parameter changes to restore frontostriatal brain dynamics in OCD.

A Schematic of the virtual intervention approach: Sets of \({n}_{t}=1\ldots 6\) posterior distribution(s) were systematically permuted between parameter sets for OCD and healthy controls, simulating the restoration of healthy parameter ranges in OCD. The statistical outcome of each intervention was estimated using the difference in distance between the simulated healthy functional connectivities (virtual control cohorts) and the simulated OCD connectivities computed after the neural parameters were permuted (virtual interventions, Fig. 1C, D, Methods, and Supplementary Fig. 5). B Statistical outcome of the targeted virtual interventions. The five most effective interventions to normalise changes in OCD functional connectivity for each number of targets \({n}_{t}\) (permuted parameters listed on the y-axis and highlighted by colour bands) are shown. The area under the receiver operating characteristic curve (AUC) is derived from the one-sided Mann-Whitney U statistic and denotes the probability that a virtual intervention improves frontostriatal functional connectivity (sample size \(n=400\)). \({n}_{t}=0\) corresponds to the baseline OCD and controls simulated functional connectivity. The vertical dashed line indicates the AUC at which p_FWE = 0.05. Average AUC \(\langle {AUC} \rangle\) per number of intervention (\({n}_{t}\)) indicates a logarithmic scaling (inset). The grey shaded area corresponds to a CI of 95% estimated by bootstrapping with 1000 resamples. C Dot-product between normalised parameter changes and the improvement of functional connectivity across statistically significant virtual interventions (p_FWE< 0.05), indicating the association between a modulation of parameter (increased or decreased values) and the resulting improvement in functional connectivity. The colour code relating to the number of targets \({n}_{t}\) is shared across panels.