Toward Robust Anxiety Biomarkers: A Machine Learning Approach in a Large-Scale Sample

Area of Science:

• Neuroscience
• Psychiatry
• Machine Learning

Background:

• Psychiatric biomarkers are needed for objective diagnosis and risk prediction.
• Neuroimaging and machine learning show promise but often lack methodological rigor.
• Anxiety biomarker development faces uncertainty due to these challenges.

Purpose of the Study:

• To develop and validate a machine learning-based neuroimaging biomarker for trait anxiety.
• To assess the generalizability of such a biomarker in an independent sample.

Main Methods:

• Applied machine learning models to predict trait anxiety from neuroimaging data (functional connectivity, volumetric, cortical thickness).
• Utilized the Brain Genomics Superstruct Project dataset with a discovery and a held-out test sample.
• Employed k-fold cross-validation for model selection and generalization testing.

Main Results:

• The best model achieved moderate predictive accuracy within the discovery sample (R² = 0.06).
• The model failed to generalize to the independent test sample (R² = -0.04).
• Permutation tests indicated significance in the discovery sample but not in the test sample.

Conclusions:

• This study did not identify a generalizable neuroimaging-based biomarker for anxiety.
• Future research should prioritize large sample sizes and robust methodologies.
• Further investigation is needed to clarify the potential of neuroimaging for anxiety biomarkers.