Causes of mental illness
We lack biological explanations and objective tests for depression, schizophrenia and other mental disorders.
Open in the interactive tree →Diagnosis rests on symptoms described in interviews, not on blood tests or scans. Genes, early life, stress and brain circuits all contribute, with thousands of small genetic effects. Treatments work only for some patients, and most were found by accident.
As of October 2026
WHO reported in September 2025 that more than 1 billion people live with mental disorders, that 727,000 people died by suicide in 2021, and that in low-income countries fewer than 10 percent of those affected receive care; governments spend a median of 2 percent of health budgets on mental health. Large genetic studies have found hundreds of associated regions of the genome (287 loci for schizophrenia in 2022, 635 loci for depression in 2025), but none has yet produced a diagnostic test.
What is missing
- Biological markers that separate disorders or predict treatment response
- Causal models linking genes, brain circuits and symptoms
- Large long-term studies that follow people from childhood
- Animal models that capture human mental states
- Diagnostic categories based on mechanisms instead of symptom lists
Becomes possible once solved
- Mechanism-based diagnosis
- Treatments matched to the individual
- Prevention of disorders in high-risk youth
Open steps
- Biology-based subtypes Medium AI leverageBiological markers that separate disorders or predict treatment response, replacing symptom lists with mechanisms.
- From risk loci to cell types High AI leverageLink the hundreds of genome-wide risk regions to the genes and brain cell types they act in.
- Genes to circuits to symptoms Medium AI leverageCausal models that link genes, brain circuits and symptoms, tested by intervention.
- Following people from childhood Medium AI leverageLarge long-term studies that follow people from childhood to find early-life paths into mental illness.
- Animal models of mental states Low AI leverageAnimal and cell models that capture human mental states well enough to test mechanisms and drugs.
Where AI could help
Medium AI leverage. AI can mine imaging and genetic data for subtypes, but validated biomarkers need large long-term cohorts and mechanism experiments.
- Clustering imaging, genetic and wearable data into mechanism-based subtypes
- Predicting who responds to which treatment from baseline data
- Linking hundreds of genome-wide association loci to genes and cell types with variant-effect models
- Mining health records for early-warning patterns before diagnosis
Shown so far
- In June 2024 Nature Medicine reported machine-learning clustering of fMRI scans from 801 people with depression or anxiety that found six brain-activity biotypes with different treatment responses; one biotype showed no brain difference. source
Prerequisites
- Psychiatric drugs1952
- Functional MRI1992
- Dopamine and reward prediction1997Reward-prediction models are used to study dopamine in psychosis, depression and addiction
- Genome sequencing2003