Back

Breaking the Norm: Population-Scale Normative Modeling of Brain Structure in Depression and Anxiety

Wiegert, J.; Marty-Lombardi, S.; Oweda, J.; Lenz, E.; Ahnert, P.; Berger, K.; Brenner, H.; Frank, J.; Grabe, H.; Greiser, K. H.; Klinger-Koenig, J.; Karch, A.; Leitzmann, M.; Meinke-Franze, C.; Mikolajczyk, R.; Nees, F.; Niendorf, T.; Sander, O.; Schmidt, C. O.; Riedel-Heller, S. G.; Ritter, K.; Peters, A.; Pischon, T.; Witt, S.; Nitsche, J.; Naamanka, J.; Volkmer, S.; Mai, A.; Abas, A.; Li, X.; Meyer-Lindenberg, A.; Gradinger, T.; Streit, F.; Braun, U.; Schwarz, E.

2025-09-28 psychiatry and clinical psychology
10.1101/2025.09.26.25336528 medRxiv
Show abstract

We applied deep normative modeling to structural MRI data from two large cohorts (German National Cohort, N {approx} 29,000 and UK Biobank, N {approx} 25,000) to characterize individual-level brain deviations along symptom dimensions of depression, anxiety, and alcohol use. Each brain was embedded into a 256-dimensional latent space, allowing us to quantify both the magnitude and direction of deviation from a normative reference trained on the non/lowsymptomatic subpopulation. Deviation magnitude increased with symptom severity, and directional patterns separated mood-anxiety and alcohol-use tendencies. These deviation axes generalized across cohorts and supported individual-level classification of symptomatic group membership, especially at higher symptom levels. Combining deviations with polygenic risk scores improved classification performance, particularly for depressive and anxiety measures, indicating complementary contributions of imaging and genetics. Our findings demonstrate that structural brain deviations reflect meaningful, continuous variation in affective and behavioral symptoms.

Published in Molecular Psychiatry (predicted rank #1) · training set

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.