In a significant leap for medical genetics, a novel artificial intelligence model has emerged with the capacity to discern which mutations in human proteins are most prone to causing disease—even when those mutations have never appeared before in any individual on Earth.

Developed through an ambitious integration of evolutionary biology and modern genomics, this model, known as popEVE, stands poised to reshape the landscape of rare disease diagnosis and genetic medicine.

The ingenuity behind popEVE lies in its harnessing of data from hundreds of thousands of species combined with genetic variation observed across the human population. Unlike previous approaches that mostly relied on limited patient cohorts or family genetics, popEVE draws upon the vast evolutionary history embedded in the biological record.

By analysing how proteins have changed—or resisted change—over billions of years, the model pinpoints which segments of the roughly 20,000 human proteins are indispensable for life and which can tolerate mutation. This evolutionary perspective offers a lens far broader than any clinical dataset could provide.

But why does this matter? For countless individuals living with rare diseases, the journey to a diagnosis is often fraught with uncertainty. Genetic mutations—especially those that alter just one amino acid in a protein—can either be benign quirks or the root cause of debilitating conditions.

The challenge facing clinicians is not simply to identify mutations, but to distinguish which ones are truly harmful, and among those, which are most severe.

Until now, many artificial intelligence tools have provided binary answers: dangerous or not. Yet, the reality is more nuanced. Some mutations may result in mild symptoms, others in severe disability, and a few may even prove fatal during childhood. The lack of a sliding scale for severity has hampered efforts to prioritise variants during diagnosis—particularly in cases where a patient presents with a mutation never before documented.

This is where popEVE marks a departure from convention. Supported by research published in Nature Genetics by scientists at Harvard Medical School and the Centre for Genomic Regulation (CRG) in Barcelona, popEVE can not only flag disease-causing mutations but also rank their severity across the entire body.

The implications are profound. Doctors can now focus on the most damaging mutations first, streamlining the diagnostic process for rare diseases and reducing time spent on less consequential variants.

Another striking advantage of popEVE is its ability to operate with only the patient’s genetic information. In many healthcare settings—particularly those with limited resources or where family DNA is unavailable—this model can accelerate diagnosis, making it more accessible and affordable.

The coauthor explains, many patients seeking genetic evaluation arrive at clinics alone, without parental samples. PopEVE’s independence from familial data enables clinicians to pinpoint pathogenic mutations regardless of background.

To appreciate how popEVE functions, it’s essential to understand its evolutionary underpinnings. Evolution has essentially conducted countless “experiments” over eons, testing which protein changes are survivable and which lead to extinction. By comparing protein sequences across an astonishing diversity of species, computational models can infer which amino acid positions are vital and which are flexible.

This principle formed the basis for an earlier algorithm called EVE (Evolutionary model of Variant Effect), which classified mutations as benign or harmful using evolutionary patterns. EVE proved remarkably effective—often matching or surpassing laboratory methods—and quickly found utility in clinical genetics.

However, EVE had its limitations. While it could assess the impact of mutations within a single gene, its severity scores were not directly comparable between different genes. For clinicians needing to rank several variants across a patient’s genome, this shortcoming was critical. PopEVE addresses this gap by merging evolutionary data with vast human genetic datasets like those from the UK Biobank and gnomAD. These repositories catalogue variants present in healthy individuals, allowing popEVE to calibrate its predictions specifically for humans.

This dual foundation—evolutionary comparison and human population data—enables popEVE to provide severity rankings that span the entire human proteome.

Now, a mutation in one gene can be directly compared on a unified scale with another mutation elsewhere in the genome. The result is a powerful triage tool for clinicians: instead of sifting through dozens or hundreds of variants without clear guidance, they can prioritise those most likely to cause serious disease.

Validation studies underscore popEVE’s utility. Researchers analysed genetic data from over 31,000 families with children affected by severe developmental disorders. In an impressive 98 percent of cases where a causal mutation was already known, popEVE correctly identified that variant as the most damaging in the child’s genome.

It also outperformed leading competitors such as AlphaMissense from DeepMind. Moreover, popEVE uncovered 123 new candidate disease genes never previously linked to developmental disorders—many active in brain development and interacting with established disease proteins.

The model’s relevance extends beyond technical prowess; it also addresses longstanding biases in genetic research. Many databases disproportionately represent individuals of European ancestry, leaving other populations underrepresented and at risk of misclassification. Some tools mistakenly flag variants as dangerous simply because they’re absent in these skewed datasets—a particular concern for patients from diverse backgrounds.

PopEVE circumvents this issue by treating all human variants equally, regardless of their frequency or population origin. By doing so, it significantly reduces false positives and ensures that no patient receives an alarming result purely due to lack of representation.

Of course, popEVE is not without limitations. It focuses exclusively on DNA changes that alter proteins—missense mutations—leaving other types of genetic variation outside its scope. It also does not replace clinical judgement; diagnosis still requires careful consideration of medical histories and symptoms alongside genetic findings.

The advent of popEVE marks a watershed moment for medical genetics and rare disease diagnostics. For healthcare systems stretched thin by resource constraints, the ability to rapidly and accurately identify disease-causing mutations using only patient DNA could prove transformative.

The model’s capacity to rank mutation severity across the human proteome offers unprecedented clarity for clinicians navigating complex cases, especially those involving unique or previously unseen variants.

The study highlights popEVE’s performance and potential impact. Experts involved emphasise that ongoing collaborations with clinics are already demonstrating real-world benefits, accelerating diagnoses and reducing uncertainty for patients who might otherwise face years without answers.

Looking ahead, further refinement and expansion of tools like popEVE may broaden their utility beyond protein-altering mutations. As genetic medicine evolves, integrating models that balance evolutionary insight with robust human data will be key to unlocking new frontiers in personalised healthcare.

For now, the arrival of popEVE signals hope for millions worldwide living with rare diseases—a future where diagnosis is faster, fairer and more precise than ever before.