Fig. 4: Sparse partial least squares (sPLS) analysis on each FNC state dwell time and behaviour questionnaires for resting-state and task-fMRI sessions.

a A multiple hold-out framework was applied to the sPLS analysis using the CCA/PLS Toolkit³⁵. The statistical significance of association in the test set was determined by 5000 iterative permutation procedures (p < 0.05). An example correlation between latent variables in the training and test set was plotted on the right. b For significant sPLS models (p < 0.05), behavioural weights were plotted with a heatmap. Behavioural items were grouped into categories: internalizing symptoms, externalizing symptoms, substance use, and cognitive performances. The correlation value and significance of all sPLS models are listed in Table S10. Behavioural questionnaire abbreviations: Development and Well-Being Assessment (DAWBA); Strengths and Difficulties Questionnaire (SDQ); Adolescent Depression Rating Scale (ADRS); European School Survey Project on Alcohol and Other Drugs (ESPAD); Alcohol Use Disorders Identification Test (AUDIT); Substance Use Risk Profile Scale (SURPS); Monetary-Choice Questionnaire (MCQ); Affective Go-Nogo task (AGN) (CANTAB, www.cambridgecognition.com); Cambridge Gambling Task (CGT) (CANTAB).