Background: Deep brain stimulation targeting the subcallosal cingulate area (SCC-DBS) has emerged as a promising therapy for treatment-resistant depression (TRD). However, only half to two thirds of patients experience meaningful clinical response, highlighting the need for biomarkers that could help to optimize SCC-DBS outcomes. Our group previously showed that a support vector machine (SVM) incorporating pre-operative FDG-PET glucose metabolism values from frontal pole, anterior cingulate cortex, and temporal pole could retrospectively classify treatment response in 21 TRD patients with 81.0% accuracy. Here, we assessed the out-of-sample performance and wider applicability of this putative biomarker. Methods: Baseline FDG-PET data were preprocessed and fed into an SVM classifier. This model, which employed the three aforementioned regional inputs, was trained and tuned using the familiar 21-patient cohort and tested on an unseen TRD validation set (n=35). Within the combined cohort, we also explored glucose metabolism’s potential influence on previously demonstrated relationships between white matter tract stimulation and clinical outcome. Results: Our model classified out-of-sample response status with 77.1% accuracy (80.0% precision; 87.0% recall; 0.83 F1 score).This performance proved statistically significant in permutation testing (ppermute=0.008) and exceeded that of an alternative, clinically informed SVM. In addition, we found that patients with lower temporal pole metabolism showed stronger coupling between uncinate fasciculus engagement (approximated using electrode localization and activation modelling) and clinical outcome (p=0.027). Conclusions: These results corroborate the validity of FDG-PET models as a tool for predicting SCC-DBS outcomes and underscore their value in refining patient selection and further personalizing DBS treatment.