Lead. Sleep problems are among the most common health complaints during adolescence, affecting up to 40% of teenagers. The relationship between poor sleep and emotional disorders is well established but poorly understood at the neural level. Not every adolescent with insomnia develops depression or anxiety, and clinicians have lacked reliable biological markers to distinguish those who are most vulnerable. A new study published in Biological Psychiatry: Cognitive Neuroscience and Neuroimaging takes a step toward solving this problem by using generative deep learning to identify distinct neuroanatomical biotypes among adolescents with insomnia symptoms.
What they found. Researchers led by Qianhui Jin and Yongbin Wei from Beijing University of Posts and Telecommunications applied a generative adversarial network (GAN) to structural MRI data from 179 adolescents who reported insomnia symptoms. The model analyzed gray matter volume across the whole brain and uncovered two reproducible neuroanatomical subtypes.
Biotype A showed gray matter volume deviations concentrated in temporoparietal and occipital regions, primarily involving areas linked to sensory processing and spatial attention. Biotype B displayed a different pattern of volume alterations that overlapped partially with these regions but carried markedly different clinical implications.
The behavioral and genetic differences between the two groups were striking. Compared with Biotype A, Biotype B adolescents scored significantly higher on DSM-5 depression ratings (p = 0.021), showed greater internalizing behaviors (p = 0.041), and had elevated anxious/depressed syndrome scores (p = 0.048). These differences extended beyond self-report measures to the genomic level: Biotype B carried higher polygenic risk scores for both major depressive disorder (p < 0.037) and anxiety disorders (p < 0.033).
Longitudinal follow-up confirmed that the heightened vulnerability in Biotype B was not a snapshot effect. These adolescents continued to exhibit elevated internalizing behaviors and anxious/depressed symptoms over time (p < 0.041), suggesting that the neuroanatomical differences track with a persistent trajectory of affective risk rather than a transient phase.
Importantly, the researchers validated their findings in an independent adult sample of 83 participants with self-reported insomnia symptoms. The same two biotypes emerged, and the association between Biotype B and elevated depressive and anxiety measures held, demonstrating that the neuroanatomical subtypes are not limited to adolescent development.
Why it matters. This study reframes insomnia-related affective risk as a problem rooted in measurable brain structure differences rather than a uniform consequence of poor sleep. The use of a generative deep learning model is noteworthy because traditional clustering methods often struggle with high-dimensional neuroimaging data. GANs can learn the underlying distribution of gray matter patterns and generate synthetic data to stabilize training, which may yield more robust subtypes than conventional approaches.
If replicated and refined, a biotype-based stratification could have direct clinical consequences. Instead of treating all adolescents with insomnia as a homogeneous group, clinicians might one day use neuroimaging-derived biotypes to identify those most likely to benefit from early, targeted mental health interventions. An adolescent classified as Biotype B, for instance, might be prioritized for cognitive behavioral therapy for insomnia combined with mood monitoring, while Biotype A adolescents could be managed with sleep hygiene alone.
The study also adds weight to the view that sleep and affective disorders share overlapping neurobiology. The fact that Biotype B carries elevated polygenic risk scores for depression and anxiety suggests that the link between insomnia and mood disorders is not purely behavioral or psychological. It may be partially encoded in brain structure from an early age.
Limits. The study has several important caveats. The sample sizes are modest, with 179 adolescents and 83 adults. Larger, multi-site cohorts will be needed to ensure the biotypes are generalizable rather than dataset-specific. The cross-sectional design for the primary biotype identification does not establish whether the gray matter differences precede insomnia, follow it, or arise from a common third factor. Although the longitudinal follow-up supports the predictive value of the biotypes, the follow-up duration and sample retention should be expanded in future work.
The reliance on self-reported insomnia symptoms rather than objective sleep measures such as polysomnography or actigraphy introduces potential reporting bias. And while the GAN-based approach is methodologically sophisticated, deep learning models can produce stable but clinically irrelevant subtypes if the latent features do not map onto meaningful biological variation.
Finally, the study does not address whether the biotypes are specific to insomnia or might overlap with neuroanatomical subtypes seen in primary depression or anxiety disorders. Comparative studies that include psychiatric control groups would help clarify specificity.
Bottom line. Adolescents with insomnia symptoms are not a monolithic group. Two distinct neuroanatomical biotypes can be identified using generative deep learning on gray matter volume data, and one of these biotypes consistently predicts higher depression and anxiety risk at both behavioral and genetic levels. The findings open a pathway toward personalized risk stratification and targeted intervention for insomnia-related affective disorders in young people, though larger and more diverse samples will be needed to translate the biotypes into clinical tools.
Source. Jin Q, Zhao S, Wang M, Wang Z, Qi T, Zhong S, Li A, Liu Y, Wei Y. Neuroimaging-derived biotypes of self-reported insomnia symptoms are differentially associated with affective symptoms. Biol Psychiatry Cogn Neurosci Neuroimaging. 2026. doi:10.1016/j.bpsc.2026.06.014. PMID: 42398825.