Clinical characteristics of each subgroup in the discovery dataset.

Clinical characteristics of each subgroup in the external independent dataset.

The results of WM signature extraction of the three subgroups.

a), b) and c) parts presented the coronal brain plane of subgroup 1, 2 and 3, respectively.

The results of clinical symptomatology and mean FA values across MDD subgroups in the discovery dataset.

a) Five symptom dimensions were visualized using radar chart. b-f) Between-group differences in scores for five symptom—anxiety/somatization, cognitive impairment, retardation, sleep disturbance and feeling of hopelessness—across the three subgroups were shown in panels b to f, respectively. h) Comparison of the mean FA values of the three WM signature patterns between each subgroup and healthy controls.

The results of clinical symptomatology and mean FA values across MDD subgroups in the external validation dataset.

a) Five symptom dimensions were visualized using radar chart. b-f) Between-group differences in scores for five symptom—anxiety/somatization, cognitive impairment, retardation, sleep disturbance and feeling of hopelessness—across the three subgroups were shown in panels b to f, respectively. h) Comparison of the mean FA values of the three WM signature patterns between each subgroup and healthy controls.

Brain WM networks associated with major clinical symptoms at the average group-level.

The thickness of each connection reflected the proportion of edge presence across subgroups. TempPar - temporal parietal, DefaultC - default C, DefaultB - default B, DefaultA - default A, ContC - control C, ContB - control B, ContA - control A, LimbicA - limbic A, LimbicB - limbic B, SalVentAttnB - salience/ventral attention B, SalVentAttnA - salience/ventral attention A, DorsAttnB - dorsal attention B, DorsAttnA - dorsal attention A, SomMotB - somatomotor B, SomMotA - somatomotor A, VisPeri - peripheral visual, VisCent - central visual, Subcor - subcortical network.

Prediction of antidepressant treatment outcomes using degree centrality of affected WM network in the discovery dataset.

Columns represented six antidepressant treatment outcome measures: percentage reduction in 24-HAMD total score and five core symptom domains—anxiety/somatization, cognitive impairment, retardation, sleep disturbance, and feeling of hopelessness. Rows indicated subgroups (i.e., 1, 2 and 3). * p≤0.05, ** p≤0.01, *** p≤0.001

Prediction of antidepressant treatment outcomes using degree centrality of affected WM network in the external validation dataset.

Columns represented six antidepressant treatment outcome measures: percentage reduction in 24-HAMD total score and five core symptom domains—anxiety/somatization, cognitive impairment, retardation, sleep disturbance, and feeling of hopelessness. Rows indicated subgroups (i.e., 1, 2 and 3). N/A indicated the absence of a valid SVR model applicable to the external dataset. * p≤0.05, ** p≤0.01, *** p≤0.001

Demographic characteristic and clinical information of depressions and healthy controls in the current study.

The pipeline of stratifying subgroup in the discovery dataset.

First, the TBSS method was used to extract FA values within the WM skeleton. NMF bi-clustering was then performed across a range of predefined ranks k (e.g., 2 ∼ 8) to generate corresponding clustering results (e.g., 2 ∼ 8 subgroups). Finally, the cophenetic correlation coefficient was employed to determine the optimal number of subgroups.

The flowchart of WM signature extraction.

a) Feature occurrence mapping was illustrated how to generate the frequency matrix F. b) The split of matrix F involved two parts: consensus-driven partitioning and signature thresholding.