Deep learning reveals a neurocomputational mechanism predicting depression risk in adolescents

Area of Science:

• Neuroscience
• Computational Psychiatry
• Genetics

Background:

• Early detection of psychiatric disorders like depression is a global health challenge.
• Current tools lack reliability in identifying individuals at risk before symptom onset.

Purpose of the Study:

• To identify a mechanistic biomarker for depression risk using functional neuroimaging and computational modeling.
• To investigate the link between neural representations of emotion and depression development.

Main Methods:

• Utilized functional neuroimaging in a large adolescent cohort (IMAGEN, N=1332).
• Employed computational modeling, including brain-aligned deep learning, to analyze neural data.
• Conducted perturbation experiments to understand emotion perception mechanisms.

Main Results:

• Weakened neural representations of emotional signals correlated with depressive symptoms.
• A neurocomputational signature predicted depression onset up to 4 years later.
• This signature was linked to genetic risk variants and improved depression classification in a patient cohort (STRATIFY, N=411).

Conclusions:

• A potential mechanism links genetic vulnerability to altered emotion perception and future depression.
• A predictive computational marker shows promise for early depression detection and prevention.
• This neurocomputational approach offers a novel tool for identifying at-risk individuals.