Evolution doesn't care about human geography
Cada año, millones de personas viven con enfermedades genéticas que la medicina no logra nombrar, atrapadas en un silencio diagnóstico que ningún secuenciador ha podido romper del todo. Un equipo liderado por Mafalda Dias ha desarrollado popEVE, un sistema de inteligencia artificial que recurre a la historia evolutiva de las proteínas —millones de años de conservación biológica— para distinguir las variantes genéticas dañinas de las inocuas. Publicado en Nature Genetics, el modelo no solo amplía las posibilidades diagnósticas para enfermedades raras, sino que desafía la desigualdad estructural de una ciencia genómica construida, en su mayor parte, sobre datos de poblaciones europeas.
- Millones de pacientes con enfermedades raras reciben secuenciaciones genómicas completas y aun así abandonan las consultas sin diagnóstico, porque identificar la variante culpable entre miles sigue siendo un problema sin solución clara.
- popEVE invierte la lógica habitual: en lugar de buscar patrones dentro de poblaciones humanas, rastrea qué cambios proteicos han sobrevivido —o no— a cientos de millones de años de evolución, convirtiendo el tiempo biológico en criterio diagnóstico.
- La sobrerepresentación de poblaciones europeas en las bases de datos genómicas hace que los algoritmos existentes fallen sistemáticamente con pacientes de otros orígenes; popEVE esquiva ese sesgo al anclar su análisis en patrones evolutivos universales.
- Colaboraciones tempranas con hospitales en Senegal ya revelan que enfermedades con presentaciones clínicas similares pueden tener raíces genéticas distintas según la ascendencia del paciente, confirmando tanto la urgencia como la pertinencia del enfoque.
- El equipo trabaja ahora para integrar el modelo en flujos clínicos reales y extenderlo a enfermedades complejas, donde no una sola variante sino la interacción de muchos factores genéticos y ambientales determina el daño.
Cada año, millones de personas portan enfermedades genéticas que la medicina no consigue identificar. Su ADN ha sido secuenciado, pero el resultado es un mapa lleno de ruido: miles de variantes genéticas, la mayoría inofensivas, algunas devastadoras, y ninguna forma fiable de separarlas. Ese silencio diagnóstico es el problema que Mafalda Dias y su equipo decidieron abordar.
Su respuesta es popEVE, un sistema de aprendizaje profundo que mira hacia atrás en el tiempo evolutivo para ver hacia adelante en la enfermedad humana. El modelo compara secuencias de proteínas a lo largo de millones de años de historia biológica —desde organismos simples hasta especies complejas— para detectar qué cambios rompen funciones esenciales. La lógica es elegante: si una secuencia se ha conservado durante eras geológicas, probablemente cumple una función crítica; si una variante la interrumpe, probablemente causa daño. La investigación, publicada en Nature Genetics, trata a la evolución como maestra.
El modelo también enfrenta un problema estructural de la genómica moderna: las poblaciones europeas dominan las bases de datos y los estudios, lo que hace que los algoritmos entrenados con esos datos funcionen peor para personas de otros orígenes. popEVE sortea ese sesgo porque las secuencias proteicas conservadas trascienden cualquier población humana. Una variante dañina se ve igual en un paciente de Estocolmo que en uno de Dakar.
El equipo ya trabaja con hospitales en Europa y África. Las primeras observaciones desde Senegal revelan algo significativo: pacientes con presentaciones clínicas similares pueden tener causas genéticas completamente distintas según su ascendencia. Lo que parece la misma enfermedad puede tener raíces genéticas diferentes en distintas poblaciones, lo que subraya tanto la promesa como la necesidad urgente de herramientas diagnósticas verdaderamente universales.
El objetivo inmediato es llevar popEVE de la publicación científica a los flujos de trabajo hospitalarios. Pero el equipo ya mira más lejos: extender el enfoque evolutivo a enfermedades complejas, donde no una sola variante sino la interacción de múltiples factores genéticos y ambientales determina el daño. Por ahora, popEVE representa un cambio de paradigma: entender lo que está roto en el presente leyendo lo que la evolución ha preservado durante millones de años.
Every year, millions of people carry genetic diseases that medicine cannot name. They move through clinics with symptoms that don't fit standard categories, their DNA sequenced but unreadable—thousands of genetic variations present in every human genome, most of them harmless, a few catastrophic, and no reliable way to tell them apart. This diagnostic silence is the problem that Mafalda Dias and her team set out to solve.
Their answer is popEVE, a deep learning system that does something counterintuitive: it looks backward through evolutionary time to see forward into human disease. Rather than trying to predict which genetic variants cause illness by studying human populations alone, popEVE compares protein sequences across millions of years of biological history—from simple organisms to complex species—to identify which changes break essential functions. The logic is elegant. If a particular genetic change has been preserved across vast stretches of evolution, it likely does something important. If it disrupts that conservation, it likely causes harm. The research, published in Nature Genetics, treats evolution itself as a teacher.
The practical stakes are enormous. Rare diseases remain undiagnosed in countless patients despite the routine availability of genomic sequencing in clinical practice. A person can have their entire genetic code read and still leave the hospital without answers, because distinguishing pathogenic variants from benign ones remains a formidable technical challenge. PopEVE doesn't solve this overnight, but it offers a new lens—one that works across human populations rather than within them.
This matters because genetic research has a visibility problem. European populations are dramatically overrepresented in genomic databases and studies, which means the algorithms trained on that data tend to work better for people of European descent. PopEVE sidesteps this bias by anchoring itself to evolutionary patterns that transcend any single human population. The same protein sequences have been conserved in organisms across the globe for hundreds of millions of years. A harmful variant looks the same whether it appears in a patient in Stockholm or Dakar.
Dias's team is already testing this principle in practice. They are collaborating with hospitals across Europe and Africa, and early observations from Senegal have revealed something important: patients with similar clinical presentations sometimes carry entirely different genetic causes depending on their ancestry. What looks like the same disease can have distinct genetic roots in different populations. This finding underscores both the promise and the necessity of the work. A diagnostic tool that works only for Europeans is not a diagnostic tool—it is a tool that works for some people and fails others.
The immediate goal is to refine popEVE's clinical application, to move it from research publication into hospital workflows where it can actually change patient outcomes. But the team is already thinking beyond rare diseases. They are working to extend the evolutionary approach to complex diseases—conditions caused not by a single catastrophic variant but by the interaction of many genetic and environmental factors. The next frontier is not just identifying which variants cause disease, but understanding how and where they act within the body, how they interact with other genes, and what interventions might work.
For now, popEVE represents a shift in how we think about genetic diagnosis: not as a problem of pattern-matching within human populations, but as a problem of reading the deep grammar that evolution has written into our proteins. It is a reminder that sometimes the best way to understand what is broken in the present is to look at what has been preserved in the past.
Notable Quotes
The team observed in Senegal that diseases with similar clinical presentations can have entirely different genetic bases depending on the patient's ancestry— Mafalda Dias's research team
The Hearth Conversation Another angle on the story
Why does it matter that evolution is being used here instead of just comparing human genomes to each other?
Because human genetic databases are skewed. Most of them come from European populations. If you train an algorithm only on that data, it learns to recognize disease patterns in European genomes but misses them in others. Evolution doesn't care about human geography—the same protein has been doing the same job in organisms across the world for millions of years.
So popEVE is looking at, say, a fruit fly and a human and asking what changed?
Exactly. If a protein sequence has barely changed between a fruit fly and a human, that means it's doing something critical. If a genetic variant disrupts that conserved sequence, it's probably harmful. That logic works regardless of which human population you're looking at.
The article mentions they found different genetic causes for similar diseases in Senegal. What does that tell us?
It tells us that the same disease can have multiple genetic roots depending on ancestry. If you only trained your algorithm on European data, you might miss the genetic cause in a Senegalese patient entirely. PopEVE's evolutionary approach is more likely to catch it because it's not looking for European patterns—it's looking for broken evolutionary conservation.
Is this ready to use in hospitals now?
Not quite. It's published and being tested in collaborations with hospitals, but moving from research to routine clinical practice takes time. They're working on making it faster, more reliable, and easier for doctors to actually use. But the foundation is there.
What happens next?
They want to move beyond rare diseases to complex ones—conditions caused by many genetic factors at once. That's much harder, but if they can crack it, the impact would be enormous. Most common diseases are complex, not rare.