Fig. 7.

Identified multimodal neuromarkers and the predictability on composite cognitive scores across three cohorts. a Four identified modality-specific brain networks from FBIRN cohort that were used as regressors to predict individual cognitive scores. b Prediction of CMINDS composite scores based on linear regression of the four regressors (mean ROI values in a). A correlation of r = 0.463 was achieved between the estimated CMINDS composite scores and its true values. c Generalization of the CMINDS prediction model in b to UNM cohort (41HCs/37SZs) to predict MCCB, r = 0.231. d Generalization of the CMINDS prediction model in b to COBRE cohort (42 HCs/46SZs) to predict MCCB, r = 0.406. In both c and d, good generalizability of the proposed prediction model were validated. The gray regions in b–d indicate a 95% confidence interval was achieved between the estimated MCCB composite scores and its true values