Figure 2.

(A) Scatter plot between the actual and predicted energy-manic symptom scores for the model based on patterns of whole-brain activation during uncertain reward expectancy using a threefold cross-validation scheme. The correlation coefficient (r) and the mean squared error between the actual and predicted energy-manic symptom scores were .42 (p = .001) and 9.93 (p = .001), respectively. For visualization purposes, subjects were color coded according to the categorically defined diagnoses to stress the transdiagnostic nature of these results. Some subjects, however, presented with symptoms that did not meet the threshold for a DSM-5 diagnosis. (B) Weight map showing the contribution of the different brain regions for predicting the energy-manic symptom score from patterns of whole-brain activation during uncertain reward expectancy. The region with the highest contribution according to the multiple kernel learning predictive model was the left ventrolateral prefrontal cortex.