Methodology & reproducibility

This project is built to be checked and extended. Findings are graded, sourced, and accompanied by the data and scripts that produce them.

How this synthesis was made

Twelve parallel literature reviews (across biology, environment, diet/gut, lifestyle, psychosocial, treatment, plus gaps, cross-references, pipeline and datasets) gathered current peer-reviewed evidence, prioritising meta-analyses, large cohorts, Mendelian-randomization studies and randomized trials. Findings were consolidated into one evidence-graded matrix, with association and causation kept explicitly separate.

The world map

The homepage choropleth shows estimated population prevalence of depression from the WHO Global Health Observatory (Global Health Estimates, 2015) for ~185 countries — a single, redistributable, internally consistent source. Greenland and Palestine use GBD/field-study estimates (flagged). For the most recent age-standardized figures, use the IHME GBD Results Tool. Country comparisons reflect detection and reporting differences as well as true burden.

The evidence ladder

Every claim is pushed upward: cross-sectional screen → prospective cohort → causal triangulation (Mendelian randomization, natural experiments) → randomized trials.

The AI discovery engine

The forward plan realises "connections no one has made" via exposome-wide association studies (ExWAS) scaled with machine learning across pooled cohorts: screen thousands of exposures with multiple-testing correction; mine interactions and non-linearities (gradient boosting, causal forests, SHAP); layer in multi-omics; and treat every AI-found pattern as a hypothesis that must replicate and survive a causal test.

Reproduce it

The matrix, source register, map data and machine-readable JSON are on the Data & datasets page. Limitations: confounding is severe, causation is hard, reverse causation is pervasive, and measurement for the newest angles (microplastics) is immature.