The labels psychiatry assigns to mental illness are useful. They give patients language for what they’re feeling, give clinicians a shared vocabulary, give insurance companies something to reimburse. But they aren’t exact the way a CT scan is exact. The Diagnostic and Statistical Manual, the field’s bible since 1952, carves the mind into categories based on how patients describe their experience, not on what’s actually happening in their biology. Think of diagnostic labels as postal codes: helpful for routing, less helpful for understanding the terrain.

A massive study published in Nature, with follow-up research rolling through early 2026, has made this harder to ignore. Researchers analyzed the DNA of over one million people diagnosed with at least one of 14 psychiatric disorders, alongside five million controls. The genetic architecture underlying these supposedly distinct conditions clusters into just five broad factors. Those five factors explain roughly two-thirds of the genetic variance across all fourteen disorders.

Put differently: the genome doesn’t seem to know the DSM exists.

The Ptolemaic Problem in Psychiatry

For over a thousand years, astronomers relied on the Ptolemaic model: Earth at the center, planets moving in elaborate epicycles. The model could predict eclipses with decent accuracy. But the framework was wrong in a way that made everything harder than it needed to be. When Copernicus proposed a heliocentric model, the math got simpler because the structure finally matched reality.

The DSM might be psychiatry’s Ptolemaic system. One patient reports persistent sadness and sleep disruption: major depressive disorder. Another presents with racing thoughts and the same sleep disruption: generalized anxiety disorder. But the genetic data shows these two conditions share an “internalizing factor” so deeply intertwined at the molecular level that separating them may be creating epicycles where a simpler model would suffice.

Five Factors, Fourteen Labels

The consortium (Harvard, CU Boulder, Mass General Brigham, Virginia Commonwealth University) used genomic structural equation modeling to find latent genetic factors beneath clinical categories. The five factors tell a story the DSM can’t: one links schizophrenia and bipolar disorder through shared patterns in excitatory neurons. Another connects depression, PTSD, and anxiety through genes expressed in oligodendrocytes, the cells that insulate neural wiring. Substance use disorders clustered together. Neurodevelopmental conditions formed their own group. A fifth factor captured compulsive traits.

A person diagnosed with bipolar disorder shares more genetic real estate with someone diagnosed with schizophrenia than with someone diagnosed with depression. Clinicians have spent decades arguing about where bipolar “ends” and each of those other conditions “begins.” The borders were always political as much as they were scientific.

Why This Isn’t Just Academic

Misclassification has real clinical costs. In economic terms: misallocated capital doesn’t just fail to generate returns, it generates negative returns through opportunity cost. Every month a patient spends on a medication chosen for the wrong reasons is a month of side effects without benefit, a month of eroding trust, a month closer to being labeled “treatment-resistant” when the resistance was never theirs.

Roughly 30 to 40 percent of patients with depression don’t remit after their first medication trial. We’ve chalked this up to the inherent difficulty of treating mental illness. But what if much of that failure rate reflects the fact that we’re sorting patients into categories that don’t correspond to what’s happening in their brains?

The Borges Problem

Jorge Luis Borges once described a fictional Chinese encyclopedia that classified animals into categories like “those that belong to the Emperor,” “embalmed ones,” and “those that, from a long way off, look like flies.” Michel Foucault built an entire book around the joke: classification systems reveal as much about the classifiers as the things being classified. Viewed from the vantage point of genomic data, some DSM categories start to resemble Borges’s taxonomy. Diagnoses aren’t useless, any more than zip codes are useless. They describe how patients feel, and that matters: the experience of depression is real regardless of which genetic factor is driving it. But the danger has been treating labels as if they describe the biology of the mind when they really describe the phenomenology, the felt experience. Useful for communication. Less useful for choosing which molecule to put in someone’s body.

What Comes Next

The research points toward a biology-first classification system. Separate work at Stanford has identified six “biotypes” of depression using functional brain imaging, each responding differently to specific treatments. When treatment was matched to biotype, prediction accuracy for remission nearly doubled. That’s what happens when the map matches the territory.

Replacing the DSM is a multigenerational project, not a software update. Insurance companies need codes. Researchers need consistent categories. We’ll be living with the current system for decades while biology reshapes it from underneath, the way plate tectonics reshapes continents: imperceptibly in real time, dramatically in retrospect. The genome already knows what the textbooks are catching up to. The best treatment is the one calibrated to your biology, not your billing code.

– Delos