Dynamic clustering of genomics cohorts beyond race, ethnicity—and ancestry

Hey there! Let’s chat about something pretty cool happening in the world of genomics. You know how sometimes, when we talk about human genetic variation, we fall back on old, familiar labels like race, ethnicity, or even broad geographic ancestry? Well, it turns out those labels, while sometimes used for important social or administrative reasons, aren’t always the best fit when we’re trying to understand the nitty-gritty details of our genes and what they do, especially when it comes to complex traits like diseases.

The Old Ways and Why They’re Tricky

For a long time, science, like society, has grappled with classifying people. Way back in the 17th and 18th centuries, folks like Francois Bernier and Carl Linnaeus tried to put humans into neat little boxes. But honestly, their systems were pretty arbitrary, often based on limited travel, anecdotal stories, and unfortunately, a whole lot of prejudice and political agendas – think colonialism and the horrific Atlantic slave trade. These early attempts weren’t just flawed; they actively created hierarchies and justified discrimination.

Fast forward a bit, and even into the 19th and 20th centuries, the concept of “race” in science remained messy and confused, famously tied up with the deeply problematic eugenics movement. Even after eugenics thankfully lost favor, using racial categories in genetic studies persisted.

But here’s the thing: human variation isn’t neat and tidy. It’s a beautiful, complex continuum. Take skin color, for example. It’s not about distinct color lines; it’s a spectrum influenced by climate, genetics, and UV radiation. People in different parts of the world can have similar skin tones not because they’re closely related, but because they adapted to similar environments. Trying to force this continuum into a few fixed categories just doesn’t capture the reality.

Even genetic ancestry, which sounds more scientific because it’s about shared genetic material, has its limitations as a *classification* tool for studies. What defines ancestry? Geography? Politics? Culture? It’s not always clear-cut, and categories can be broad (like “continent-based”) and accidentally imply a “purity” that doesn’t exist. Plus, most of us have ancestry linked to multiple groups, which broad, single labels completely miss. Using these predefined categories, especially broad ones, can risk overlooking important genetic signals specific to the trait we’re studying. It’s like trying to understand a detailed map using only country borders – you miss all the interesting local features!

A Smarter Way to Group Genes

So, what do we do? Given the incredible amount of genomic data we have now and the sheer complexity of human variation, sticking to those fixed, predefined categories feels… well, a bit reductionist, doesn’t it?

This is where a new idea comes in, building on some earlier thoughts but really taking off with today’s technology. Instead of grouping people based on those old, broad labels, we can group them dynamically based on the genomic variation *relevant to the specific trait* we’re interested in. Imagine you’re studying handedness. Instead of grouping people by their continent of origin, you group them by the variations in genes known to influence whether you’re left or right-handed. This is the core of the dynamic clustering approach.

The beauty of this is that it doesn’t use any predefined boxes. The clusters emerge directly from the data, based on similarity in the specific genetic regions tied to the trait under study. The number of clusters isn’t fixed beforehand; it depends on the data and the trait itself. It’s a much more flexible and, frankly, more biologically sensible way to look at things.

Testing the Waters with Cancer

To see if this dynamic approach really works, we decided to test it on something really complex: cancer. Cancer isn’t just one disease; it’s many, with intricate genetic underpinnings. We used publicly available data from The Cancer Genome Atlas (TCGA), looking at germline (inherited) genetic variants in genes known to be associated with cancer predisposition across ten different cancer types.

We applied our dynamic clustering method, focusing on the variations in these cancer-relevant genes for each specific cancer type. We wanted to see how the individuals in these cancer cohorts would cluster based *only* on these trait-specific genes, ignoring their reported ancestry or other broad labels.

What We Discovered: Leaving Categories Behind

The results were pretty eye-opening! When we clustered individuals based on their germline variants in cancer-specific genes, the groupings we saw consistently *transcended* the old continent-based categories. People labeled as having “African,” “East Asian,” “European,” or “Other” ancestry often ended up in the *same* dynamic cluster if their cancer-relevant genes were similar. This happened across all ten cancer types we studied.

What’s more, the number of clusters wasn’t fixed. It varied from one to eight depending on the cancer type, which makes total sense because different cancers have different levels of genomic complexity and heterogeneity. This dynamic approach naturally reflected that biological reality, unlike a fixed-category system.

Perhaps one of the most exciting findings was that this dynamic clustering helped us identify potential *novel* driver genes for cancer – genes that might be important in cancer development but were *overlooked* when the analysis was done using the traditional continent-based categories. This suggests that the old way of grouping people might actually be hiding important biological signals.

We also found that standard algorithmic clustering methods (like K-means, DBSCAN, HClust) didn’t always do a great job of finding these dynamic clusters on their own, highlighting that sometimes a little human insight is still needed to interpret the patterns in complex data.

Connecting Genes to Real Life

These dynamic clusters weren’t just abstract groupings on a chart. We dug deeper and found that they were biologically meaningful. The genes identified within these clusters were associated with fundamental biological processes known to be involved in cancer, like cell cycle regulation, apoptosis (programmed cell death), and various signaling pathways. Including the novel genes we found helped paint a more complete picture of the underlying biology.

Even cooler, these dynamic clusters showed associations with clinical factors. For example, in one lung cancer cohort, a specific dynamic cluster was associated with a lower age of cancer onset. In liver cancer, certain clusters were linked to higher tumor grade or more advanced tumor stage. We also saw differences in gene expression patterns between clusters – genes being turned on or off differently – that are known to be involved in things like tumor growth, metastasis, and patient survival. This really drives home that these dynamic groupings based on trait-specific genes capture relevant biological and clinical differences among individuals.

Why History Matters in Science

It’s crucial to remember *why* moving beyond categories like race and ethnicity in genomics is so important, both technically and socially. As we touched on earlier, the history of using “race” in science is deeply intertwined with systems of power, discrimination, and enforced social hierarchies. Race isn’t a biological discovery; it’s a social construct, an idea used to control and disenfranchise. Using these categories in genetic studies, even if not intended maliciously, risks perpetuating harmful associations and stigma onto entire communities.

Ethnicity categories, while sometimes based on culture or language, also don’t always align with genetic ancestry and can be malleable. Even genetic ancestry, while reflecting shared inheritance, can be problematic when forced into broad, fixed categories that ignore individual complexity and the continuous nature of human variation across geography and time.

The scientific understanding of human variation has progressed dramatically. We see continuums, not discrete boxes. Alleles (gene variants) that increase disease risk aren’t confined to single “racial” or “ancestral” groups; they can be found across the globe.

Relying on broad, predefined stratification, whether it comes from discriminatory legacies or just normalized data collection practices, can actually *hide* the very patterns we need to see – the ones that transcend those old boundaries and are truly relevant to the biological trait under study.

Painting a Fuller Picture

So, what’s the takeaway? With the vast amounts of genomic data available today, we need approaches that match the complexity of human variation. Dynamic clustering, by focusing on trait-specific genetic similarity and letting the data define the groups, offers a powerful alternative to fixed, broad categories.

This approach, coupled with the essential need for more diverse data collection from around the world, has the potential to give us a much more complete and accurate picture of human genomic variation and its role in health and disease. It helps us see the continuums and the specific genetic patterns relevant to a trait, rather than being limited by potentially misleading, historically loaded boxes.

It’s about using new and existing quantitative tools that can look at variation in different ways – focusing on specific genes, looking at continuous clines, or using measures of genetic relatedness – depending on the research question. This helps us tackle complex challenges in genomics and ultimately, to navigate the intricate relationship between science and society more responsibly.

We think this dynamic approach is a step towards a future where we study human genomics in a way that is both technically rigorous and socially aware, unlocking new insights that were previously hidden by outdated classification systems.

Source: Springer