I spend many hours working with LLMs to help me support my academic writing. This post summarises what I have learned. If you are using LLMs to help write academic papers, grant applications, or technical reports, these observations might save you from some painful mistakes.

You can listen to the extended version for this blog entry either in the attached audio or in my podcast.

This is my most frequent source of errors. The models cannot write confident, detailed descriptions of figures. The problem is fundamental: LLMs generate text based on what sounds plausible in context. They have no way to verify their descriptions against your actual visual content. Every single figure caption and every sentence that references a figure needs to be checked manually against the image itself.

When I ask the model to update a document, LLMs process text in chunks and do not maintain a global view of document consistency. After any significant revision, you need to search the entire document for terms that should have been replaced. Do not assume the model caught everything.

LLMs are remarkably good at generating numbers that feel appropriate for a given context. They are not good at generating numbers that match your actual data. Every quantitative claim needs verification against your source files, your code output, or your analysis logs. This includes sample sizes, p-values, effect sizes, percentages, and any other numerical assertion.

LLMs write persuasively. They naturally reach for broader, more impressive-sounding claims. This is useful when drafting, but dangerous when finalising. For each conclusion, ask yourself: did we actually measure this? If the answer involves any hedging, the text probably needs hedging too.

LLMs tend to treat related terms as interchangeable, so you need to define your terminology clearly and then verify it is used precisely throughout.

LLMs often describe standard or common approaches rather than your specific implementation. If your Methods section was drafted or revised with LLM assistance, cross-reference every procedure against your actual analysis code.

None of this means LLMs are useless for academic writing. They are genuinely helpful for structuring arguments, improving prose flow, catching grammatical issues, and generating first drafts quickly. But they optimise for fluency and plausibility, not for accuracy against your specific data.

The practical implication is that LLM-assisted writing requires a verification pass that you might not need with purely human-written drafts. You need to check figures against their descriptions, search for outdated content after revisions, verify every number against source data, ensure claims match what was actually measured, confirm technical terms are used consistently, and cross-reference methods against code.

This takes time. But it takes less time than responding to reviewer comments about errors, or worse, correcting a published paper.

Think of the LLM as a skilled collaborator who writes well but has never actually seen your data, run your code, or looked at your figures. They are working from your descriptions and their general knowledge, doing their best to produce something coherent and polished.

Your job is to be the fact-checker. The LLM drafts; you verify. That division of labour, properly maintained, can be genuinely productive. But if you skip the verification, you will end up with a manuscript full of confident, well-written errors.

Presented at the UK–Indonesia Health Genomics Forum, London, 25 November 2025

When we speak about the future of precision medicine, it is easy to focus on the technology. We talk about sequencing platforms, AI systems, analytical pipelines and discovery engines. Yet the real foundation of precision medicine is something much more fundamental. Precision medicine cannot be precise if the global datasets that feed it continue to exclude most of humanity.

This was the central argument I presented at the UK–Indonesia Health Genomics Forum 2025. I began with a simple observation. Around 80 percent of genomic discovery data comes from individuals of European ancestry. Entire regions such as Southeast Asia, Africa, Latin America and Indigenous nations remain significantly underrepresented, even though they contain some of the richest reservoirs of human genetic diversity in the world.

This imbalance is not abstract. It shapes the tools we build. It influences which drug targets rise to the top of development pipelines. It affects the accuracy of polygenic risk scores. It affects how AI models generalise across populations. When most global datasets are drawn from a narrow subset of the world, the science built on top of them inherits the same limitations. This leads to inequitable care and lost opportunities for discovery.

A widespread misunderstanding concerns the relationship between population size and genetic diversity. The two are not equivalent. Africa represents about 17 percent of the global population but holds more than a quarter of humanity’s genetic diversity. Latin America accounts for roughly 8 percent of the world’s population but contains close to 16 percent of global genetic variation. Asia is home to almost half the planet, yet its representation in genomic databases remains low because so few communities have ever been sequenced. Indonesia alone has more than 1,300 ethnic groups. It is one of the world’s great centres of human diversity, but almost entirely missing from global datasets.

The consequences are clear. If these populations are absent, our science is not only incomplete but also unreliable. Our predictions become less accurate. Our biological insights become narrow. We describe our work as global, although it captures only a fragment of global variation.

Indonesia and Latin America share a particular challenge. Both are regions with extraordinary diversity but limited representation in international genomics. Both have experienced long histories of extractive research, where samples left the country and benefits did not return. Trust has been damaged. Local capacity has often been overlooked. Without investment in local sequencing, bioinformatics and governance, even well-meaning initiatives risk repeating these patterns.

This is why solving the diversity gap is not a matter of ethics alone. It is a scientific necessity. Genomic diversity is not a decorative add-on. It is the backbone of robust and equitable science. If we want drug development to work for everyone, if we want AI models that are safe and reliable, and if we want precision medicine that truly deserves its name, we must rethink how we build global datasets.

The solution cannot be reduced to collecting more samples. What we need is a coherent system that strengthens trust, protects sovereignty and builds capability in the regions where the data originates. This means developing sequencing and data infrastructures across Southeast Asia and Latin America. It means training local scientists and ensuring that benefits remain with the communities that contribute. It means governance frameworks that protect rights and promote responsible use. It means treating Indonesia and Latin America as equal partners in shaping the next generation of genomic science.

The future of genomics will be defined by the choices we make now. The missing pieces of our global datasets are not peripheral. They represent essential chapters of the human story. Without them, our science, our tools and our predictions will remain incomplete.

My invitation in London was clear. Let us build a future in which diversity is recognised as scientific infrastructure. Let us ensure that equity is not an aspiration but a practice. Let us move towards a form of precision medicine that reflects the full breadth of the populations it aims to serve.

The path is visible. What we need now is the collective will to follow it.

By Dr Manuel Corpas

Precision medicine is built on a contradiction. The field promises individualised care calibrated to each person’s biology, yet its foundations rest on data representing a fraction of humanity. After two decades at the intersection of genomics, artificial intelligence, and global health, I have reached a conclusion that the field must confront directly:

You cannot have precision medicine with partial data. And today, our data is profoundly partial.

This is not a diversity problem to be managed. It is a scientific validity problem that undermines the entire enterprise. Genomic medicine is making 100% claims on 80% foundations.

For most of medical history, treatment followed population averages. Clinicians asked: What works for most people? The answer was imprecise at best and harmful at worst.

Consider the evidence: for every person helped by a top-10 prescribed drug in the United States, between three and twenty-four people fail to benefit—some experiencing adverse reactions. This is what happens when treatments are calibrated to a mythical “average patient” who does not exist.

Genomics offered a different question: What works for this person? The cost of sequencing a human genome collapsed from $100 million in 2001 to under $1,000 today—a trajectory that outpaced Moore’s Law. Medicine could finally become personal.

But this promise rests on a fragile assumption: that the underlying data represents the full diversity of human biology. It does not.

Here is the defining statistic of contemporary genomic medicine: approximately 80–86% of all genome-wide association study (GWAS) data derives from individuals of European ancestry. Europeans constitute roughly 16% of the global population.

This imbalance is not incidental. It is structural, and it cascades through every layer of the precision medicine pipeline.

Reference genomes were constructed predominantly from European donors. Variants of Unknown Significance (VUS) are substantially more common in non-European ancestries, precisely because those populations lack adequate representation in the databases used to interpret clinical meaning. Polygenic risk scores—the most promising tools for predicting complex disease—show predictive accuracy that decays linearly with genetic distance from European training populations. Drug response predictions, calibrated on narrow populations, miscalculate for large regions of the world. And AI models, trained on biased data, amplify these distortions at scale.

This is not merely an equity concern. It is a scientific validity problem. When narrow data is used to build universal models, the science fails on its own terms.

Before the genomic era, Mendelian disorders—single-gene conditions—dominated the field. These remain important but rare.

The vast majority of the global disease burden—diabetes, cardiovascular disease, Alzheimer’s, cancer—is polygenic, involving hundreds or thousands of variants interacting with environment and lifestyle. This reality complicates interpretation substantially: small genetic effects accumulate; environment modulates risk; no variant determines destiny; and critically, all of this depends on the population in which the model was trained.

Precision medicine constructed on narrow ancestry cannot deliver precision for global populations. What we have instead is statistical inference masquerading as clinical truth—confident predictions that work well for some populations and fail systematically for others.

My research centres on Peru, a country whose genomic and environmental diversity challenges the assumptions embedded in current databases.

Peru comprises over 55 Indigenous populations, deep Amazonian ancestries, and high-altitude groups who have lived above 3,000–4,000 metres for millennia. Roughly 30% of the population resides at extreme altitude, where oxygen availability drops to approximately 60% of sea-level concentrations. Over thousands of years, these conditions have reshaped immune pathways, cardiovascular responses, and metabolic processes in ways absent from global genomic resources.

The findings are instructive: high-altitude immunity operates through distinct pathways compared to low-altitude populations; protective variants in one environment can become vulnerabilities in another; drug metabolism patterns diverge significantly from European-centred assumptions.

This work is not only about Peru. It reveals biological mechanisms that would remain invisible if the genomic lens remained Eurocentric. The diversity that has been excluded from genomic research is precisely where novel scientific insights reside.

Addressing these structural imbalances requires tools that do not yet exist. To this end, my team is developing the Health Equity Informative Markers (HEIM) framework, designed to quantify the equity and representativeness of genomic datasets, establish diversity standards, embed data sovereignty and benefit-sharing principles, provide equity metrics for AI models, and translate representation directly into clinical guidelines.

The field speaks frequently about diversity. It rarely measures it. Without quantification, diversity remains aspiration rather than methodology. HEIM is designed to change that equation—to move equity from rhetoric to rigour.

Precision medicine is being constructed on incomplete data while making universal claims. The field asserts individualised accuracy while ignoring the populations for whom that accuracy does not hold.

The evidence of this contradiction is already visible. Polygenic risk scores fail when applied to African, Latin American, or South Asian populations. Variant interpretation becomes unreliable outside European reference frames. Drug dosing guidelines misfire in Indigenous communities. Clinical tools optimised for European ancestry may work adequately for that population—but they do not generalise.

The global spread of precision medicine, absent correction, risks deepening the very inequities it claims to address. We cannot build equitable healthcare on inequitable foundations.

Delivering on the original promise of precision medicine requires a fundamental reorientation of practice.

First, representation in genomic research must expand—not as a compliance exercise, but as core scientific methodology. Populations currently underrepresented in databases are not supplementary to science; they are essential to it. The variants, adaptations, and disease mechanisms present in these populations hold keys to understanding human biology that Eurocentric data cannot provide.

Second, AI models must be designed and validated to work across populations, not merely tested on convenient cohorts and then deployed globally. Technical solutions exist; the barrier is institutional will.

Third, data sovereignty must become standard practice. Communities contributing genomic data should hold governance rights over that data. The extractive model—where data flows from the Global South to Northern institutions without reciprocal benefit—is scientifically limiting and ethically untenable.

Fourth, collaboration must replace extraction. Underrepresented populations should be partners in research design, interpretation, and benefit, not subjects from whom data is harvested.

Fifth, clinicians and scientists must be educated that genetic risk is probabilistic, not deterministic, and that the confidence intervals widen dramatically when tools are applied outside their training populations.

Precision medicine cannot remain a privilege available to those who fit the reference genome. It must become infrastructure that works for everyone.

The potential of genomic medicine remains substantial. But that potential cannot be realised on fractured foundations. The decisions the field makes now will shape healthcare for the next century.

My work—and increasingly, our collective work—must ensure that precision medicine serves all of humanity, not merely those whose ancestors happened to be recruited into early studies. If we want the future of medicine to be truly precise, it must first become truly equitable.

The science demands it. The ethics require it. And the populations currently excluded deserve nothing less.

Dr Manuel Corpas is a Senior Lecturer in Genomics, AI, and Data Science at the University of Westminster, focused on advancing equitable genomics for global health.

A recap of my presentation to Britcham Brazil, 7 October 2025

This morning I had the privilege of presenting to a distinguished audience at the British Chamber of Commerce and Industry in Brazil (Britcham), as part of a webinar titled “Health Data Equity in Latin America in the Age of AI and Genomics.” I was joined by Mr. Adam Patterson (Deputy Director and Honorary Consul in Curitiba), Mr. Luciano Moraes (President of the Technology and Innovation Committee), and Mr. Luiz Felipe Di Sessa (Vice-President of the same committee), who moderated and contextualised the discussion with clarity and insight.

My presentation focused on a simple but powerful idea: genomics and AI can transform healthcare but only if the data that powers them represents us all.

I began by presenting the current landscape: despite Latin America accounting for over 650 million people, it contributes less than 2% of global genomic data. This underrepresentation creates a knock-on effect. AI models trained on Eurocentric datasets inevitably perform worse for populations like Brazil’s, leading to misdiagnoses, overlooked risks, and missed opportunities for precision treatments.

To ground this problem, I shared data from multiple studies (Popejoy & Fullerton 2016; Fatumo et al. 2023; Skantharajah et al. 2023) showing how clinical trials, genomic repositories, and pharmacogenomic datasets disproportionately reflect individuals of European descent: mostly men, mostly from a handful of high-income countries .

Brazil’s population is one of the most genetically diverse in the world, with layered ancestries tracing Indigenous, African, and European roots. I presented evidence of Brazil’s complex Native American admixture, which occurred long before European contact, highlighting the scientific and clinical importance of this diversity.

Yet this diversity is largely invisible in current AI and genomic infrastructures. We risk creating diagnostic tools and therapeutic models that simply don’t work for large segments of Brazil’s population, especially its Indigenous and Afro-descendant communities.

I underlined how biobanks can be powerful enablers of equity, offering researchers large, structured datasets that include genetic, environmental, and social variables. But to serve as global resources, these biobanks must be inclusive from the start. If not, they replicate existing biases at scale .

AI learns from data, and when that data is skewed, so are the outputs. I walked the audience through examples of how AI systems can misclassify risk in underrepresented populations, or entirely miss pathogenic variants because they’re rare in European datasets but common in Latin American ones .

But I also highlighted the positive potential: with the right guardrails, AI can help correct bias. Tools like:

can be deployed to make models more fair and more robust.

I closed by emphasising that Brazil is uniquely positioned to lead in equity-aware AI and genomics. Its national health system (SUS), combined with its genomic richness, creates the conditions for building world-leading, inclusive infrastructure. If Brazil chooses to invest in its own biobanks, equity dashboards, and genomic AI models, it could emerge not just as a regional leader but as a global standard-bearer for responsible innovation.

My message was clear: inclusion drives accuracy. The more representative our data, the more effective our AI. Latin America’s diversity is not a problem to be solved, but a resource to be celebrated. It holds the key to scientific discoveries with global impact.

These weeks I will be putting here the lectures I am currently delivering for Biobanking for Data Science, a module that I lead at the University of Westminster for the MSc AI Digital Health course I also lead. I will be providing a summary and the recordings for these lectures.

The research presented here is available via this preprint, which is currently in revision.

Biobanks have become one of the most powerful infrastructures in modern biomedical research. By systematically collecting, storing, and cataloguing biological materials alongside rich clinical and lifestyle data, they enable discoveries that shape the way we understand disease, predict risk, and develop personalized treatments.

Over the past two decades, biobanking has evolved from static repositories into dynamic digital ecosystems. What began in the early 2000s with standardization efforts has now entered an era of high-throughput data generation, ethical regulation, and most recently artificial intelligence (AI) integration . Today, biobanks are not just about sample storage; they are about data-driven health intelligence.

Biobanks underpin critical advances in both research and healthcare:

From the UK Biobank’s half a million participants to the U.S. All of Us initiative and the Estonian Biobank, these large-scale resources are shaping how science approaches both common and rare diseases .

The 2020s have brought a decisive shift: the convergence of biobanking with AI and digital health. Machine learning tools now allow us to sift through immense, heterogeneous datasets—genomes, imaging, electronic health records—to generate insights that were previously inaccessible .

This is where Large Language Models (LLMs) enter the stage. Trained on massive text corpora, LLMs like GPT and Claude are proving their value in mining biobank-related literature, summarising biomedical findings, and even benchmarking patterns across thousands of studies .

In recent work, I explored how LLMs perform when applied to UK Biobank research outputs. The findings are illuminating:

Yet challenges remain—coverage does not guarantee accuracy, and LLMs still struggle with clinical reasoning, multimodal integration, and hypothesis generation .

The promise of AI-enabled biobanking is vast, but so are the challenges:

If addressed properly, the integration of biobanking with LLM-driven analytics could revolutionise our ability to link genetics, lifestyle, environment, and health outcomes.

The future of biobanking lies in its digital transformation. AI and LLMs are not replacing the scientific process but augmenting it, helping us navigate complexity, accelerate discovery, and bring equitable precision medicine closer to reality.

The question now is not whether we will use these tools, but how responsibly and effectively we can embed them into global health research infrastructures.

When I’m asked how I became a scientist, I usually smile, because the truth is, I’ve been aspiring to be one for over 30 years. My career didn’t follow the straight path I once imagined. I failed to get into medical school, so I turned to biomedical sciences instead. During my undergraduate studies, something unexpected happened, I caught the “research bug.” I became obsessed with understanding nature, driven by curiosity and the thrill of discovery.

At first, I was far from a stellar biologist. I contaminated more samples than I care to admit. But I found myself fascinated by data, and this was the dawn of the internet era, when the Human Genome Project and the first waves of digital transformation in health were reshaping science. That curiosity led me to a PhD in bioinformatics at the University of Manchester. Suddenly, I was at the forefront of marrying computer science with biology, a field still in its infancy, and I’ve been working at that intersection for 25 years.

My postdoctoral years at the Wellcome Sanger Institute in Cambridge immersed me in large-scale genomic data. I worked on DECIPHER, a database that became the backbone of genomic sequencing across UK and Irish paediatric centres. But I realised genomics wasn’t just for rare childhood conditions: it was for everyone.

In a move that would shape my career, my family and I sequenced our genomes and made them freely available online. At the time, we were likely the first family to do so. It sparked a community of citizen scientists and crowdsourced analyses, culminating in a widely cited 2015 paper.

But something caught my attention: when I compiled thousands of voluntarily shared genetic datasets from around the world, 95% were of European ancestry. This was 2017, before health data equity became a mainstream discussion, and it was a wake-up call. Who had access to genomic testing? Who was being represented in the data? And crucially, who was being left out?

The underrepresentation of global populations in genomics isn’t just about cost, though tests remain prohibitively expensive in much of the world. Many companies don’t even ship to the Global South. In countries like Peru, where I work frequently, lab reagents cost twice as much, and specialist equipment is scarce. Often, samples must be shipped abroad for analysis, stripping local scientists of agency and missing opportunities to train local talent and strengthen local economies.

For me, the solution is clear: whenever possible, genomic analyses should be performed locally. It’s about capacity-building, empowerment, and ensuring that scientific progress benefits the communities it studies.

Based in the UK, I saw a unique opportunity. The UK is a leader in genomic technology and implementation, but it has limited ties with Latin America, my cultural and linguistic home. That’s why I launched the first Spanish-speaking congress dedicated to genomic medicine, creating a bridge between cutting-edge UK science and the Spanish-speaking world.

The challenges are stark. Many genetic tests are built on European genetic markers, limiting their accuracy for non-European populations. In pharmacogenomics, this can be dangerous. For example:

These are not abstract disparities, they are real, life-and-death consequences of inequitable data.

Scientists, clinicians, and even many pharmaceutical companies don’t want these inequities. The real barrier is the business case: the financial return for tailoring interventions to small or underserved populations is often unclear. But there are models to draw from, like those used in rare disease research, where “n=1” cases have inspired novel, scalable approaches.

The truth is, diversity is no longer optional. Global migration means that every country now hosts a rich mix of populations. Medicine must adapt not just for fairness, but for accuracy.

I envision a future where it doesn’t matter where you’re born, what you look like, or your socioeconomic status, everyone can benefit from the best that science offers.

That means:

The COVID-19 vaccine rollout showed what’s possible when the world aligns around a single goal. If we can do that for a pandemic, we can do it for health equity.

As a technologist, I believe we spend too much time on tools and too little on their human and ethical context. Innovation is essential, but so is remembering that every data point has a face, a story, and a community behind it. True progress in genomics and in medicine comes when technology serves people, not the other way around.

Final Thought: Health equity in genomics isn’t a niche goal. It’s essential for the validity, safety, and universality of medical science. We’ve identified the problem. We have the tools. Now we need the will, the business models, and the partnerships to make equitable precision medicine a reality.

by Dr. Manuel Corpas

Despite breathtaking advances in genomics over the past two decades, we are failing to answer a fundamental question: who benefits from this progress?

The sobering truth is that the benefits of precision medicine are overwhelmingly skewed toward people of European ancestry. In 2016, 81% of all genome-wide association studies (GWAS) were conducted in Europeans. Shockingly, by 2021 that number had increased to 86% . Rather than diversifying, the field has become even more homogenous.

This imbalance is not just academic, it has real consequences for global health. The lack of diversity in genomic datasets means we’re missing key genetic variants that influence disease risk, drug response, and treatment efficacy in underrepresented populations. In pharmacogenomics, for example, African populations suffer disproportionately high rates of adverse drug reactions, in part due to the near absence of population-specific variants in prescribing algorithms .

Take CYP2D6, a gene that affects how the body metabolizes more than 25% of all prescribed drugs. In Ethiopia, the high prevalence of a specific duplication variant means that standard codeine use can be life-threatening, yet this information is largely invisible to current clinical guidelines . Similarly, warfarin dosing algorithms (routinely used for stroke prevention) perform well in Europeans, but fail to account for genetic variation in African and mixed ancestry populations .

Underrepresentation isn’t limited to GWAS or pharmacogenomics. Across clinical trials and biobanks, individuals from Latin America, Africa, Asia, and Indigenous communities remain largely excluded. This lack of inclusion leads to medical practices that work well for some, but leave many behind.

We must do better.

We must ensure that every person, regardless of their ancestry, has an equal chance to benefit from genomic insights. That means investing in globally inclusive research, building equitable reference databases, and rethinking the scientific standards that currently reward sample size over representation.

Health equity in the genomic era isn’t a luxury, it’s a necessity.

Only by confronting the diversity gap head-on can we unlock the full potential of precision medicine for all.

Genomics holds the promise to revolutionize healthcare—offering tailored diagnoses, treatments, and prevention strategies. But as I shared in my recent talk, this vision remains dangerously incomplete without one critical ingredient: diversity.

At the heart of my presentation, delivered on 26 June 2025, was a stark reminder from Popejoy and Fullerton’s landmark 2016 Nature paper: over 80% of genomic data used in research comes from individuals of European ancestry . Nearly a decade later, while awareness has improved, the numbers have barely shifted. This lack of representation is not just a scientific flaw—it’s a humanitarian one.

Genomic diversity matters because it directly influences the accuracy and efficacy of medical interventions. When reference panels are built on narrow population groups, they fail to capture the genetic variants that are common—or even critical—in other ancestries . This can lead to misdiagnoses, ineffective treatment plans, and serious adverse drug reactions.

Take, for instance, CYP2D6, a gene involved in the metabolism of over a quarter of all registered drugs. A specific duplication variant of this gene affects nearly 30% of Ethiopians, leading to toxic outcomes with codeine—a commonly prescribed painkiller. Similarly, certain alleles impact how women with African ancestry respond to tamoxifen, a key breast cancer drug . Without this knowledge, clinicians are left in the dark—and patients bear the consequences.

The issue is compounded by sex bias in pharmacogenomics. Women, especially older women, experience adverse drug reactions almost twice as often as men. Yet gender and sex differences are often sidelined in drug trials and genomic datasets . With the rise of polypharmacy in aging populations, this oversight poses a growing threat to public health.

My presentation also touched on the right to science. If genomic medicine is to fulfill its promise, then all populations must have equal access—not only to its benefits but also to participation in its construction.

The takeaway is clear: we must challenge the systemic biases in genomic research. We need broader, deeper, and more inclusive data. And we need to do it now—not later. Because when only a fraction of humanity is represented in the data, only a fraction can truly benefit from the science.

In this episode of Personal Genomics Zone, I sit down with Professor Yves Moreau from KU Leuven for a deep and thought-provoking conversation on the intersections of artificial intelligence, genomics, and society.

Our connection goes back to our shared contributions to the DECIPHER database — a foundational project in rare disease research that has shaped how genetic data is shared across borders. With over 2,000 citations, it’s a landmark we’re both proud of.

But this chat goes far beyond our technical roots. Yves shares his journey from engineering and neural networks to bioinformatics and drug discovery, and his growing preoccupation with the social consequences of technology. From his work with Carta Academica — a Belgian academic network engaged in public debate — to his activism against the misuse of forensic genetics, Yves offers a compelling vision of what it means to be a scientist with conscience.

Whether you’re a researcher, policymaker, or just curious about the ethical frontiers of genomic science, this conversation is not to be missed.

Subscribe for more interviews at the cutting edge of genomics and society.

#Genomics #AI #Bioethics #PersonalGenomics #YvesMoreau #DECIPHER #ForensicGenetics #TechnologyAndSociety #Bioinformatics #PrecisionMedicine

Imagine two people walk into a hospital. Same symptoms. Same diagnosis. One receives a treatment that works like magic. The other? Nothing.

Why? Because what’s written in our DNA is as unique as our fingerprint. And the hard truth is: the science that informs healthcare—its research, data, and trials—has been built on just a handful of fingerprints, while billions of others remain invisible.

As genomics rapidly becomes the foundation of medicine—from diagnosing diseases to developing life-saving drugs—we must confront an urgent question: who is being left out?

When I first analyzed open-access genetic datasets back in 2017, I stumbled upon a troubling fact: 95% of those datasets came from individuals of European ancestry. That was a wake-up call.

It wasn’t just a statistical imbalance—it was a systemic flaw. Populations across Africa, Latin America, Asia, and Oceania are missing from the very data that’s driving our genomic breakthroughs. That absence leads to real-world consequences: treatments that don’t work, adverse drug reactions, and diagnostic tools that fail.

One example: warfarin, the world’s most prescribed anticoagulant. In people of African ancestry, one in four experience adverse drug reactions. Why? Because dosing algorithms are trained predominantly on European data. That’s not just a data problem. It’s a healthcare injustice.

My journey began at the Sanger Institute in Cambridge, where I worked on rare developmental diseases. Back then, we lacked the maps and tools to diagnose many of these conditions. But sequencing the human genome changed that. Suddenly, we could read the blueprint of life.

That experience convinced me of genomics’ potential—not just for the few, but for everyone. Every human carries mutations. Every genome holds clues about disease susceptibility, drug response, and more. Genomics should be a universal right. But that can only happen if our science includes everyone.

We talk a lot about “precision medicine.” But how precise can it really be if entire populations are absent from the data? Precision for some is not precision at all.

The lack of diversity isn’t always intentional. It’s often the result of where research happens: in the Global North, among predominantly white populations, with infrastructures that don’t reach remote or under-resourced regions. But the outcome is the same—a system skewed toward a narrow slice of humanity.

One of the most humbling parts of my work has been engaging directly with indigenous communities in Latin America. These encounters changed me. They reminded me that science isn’t just about data—it’s about people.

It’s one thing to run algorithms. It’s another to sit in a village deep in the Amazon and hear a mother say, “Please help us. We want access to medicine. We want our children to be healthy.”

Science must begin with listening. If we want to build trust, we have to show up—not as experts, but as humans. Ethical genomic research must be culturally sensitive, community-driven, and grounded in genuine relationships.

We also need to address how ancestry and sex biases creep into research. Women are underrepresented in trials. Pregnant women are often excluded altogether. Africa, the most genetically diverse continent, is still treated monolithically in many studies.

Even our definitions of “diversity” vary wildly across datasets—from FDA classifications to pharmacogenomics to clinical genomics. There’s no consensus on what counts as representative, and that makes equity nearly impossible to measure.

We need standards. We need metrics. And we need leadership willing to prioritize inclusion—not just as a checkbox, but as a scientific imperative.

I often say that genomics is waiting for its “ChatGPT moment.” That moment when a powerful, complex technology becomes accessible and impactful for everyone—not just specialists.

We’re not there yet. Biology is complex. The genome is only one layer. Add the epigenome, transcriptome, proteome, and microbiome, and the picture becomes exponentially more intricate. But I believe AI can help us untangle that complexity. If—and only if—we include the full diversity of human biology in the training data.

If we do nothing, the future of medicine will be one of stark inequality. Those with wealth and access will benefit from personalized treatments and longer lives. Those without will be left behind.

But that’s not inevitable. There is hope.

The UK, for example, through initiatives like Genomics England and Our Future Health, has made real strides in democratizing access to genomic data—even if its datasets still reflect a predominantly white population. The openness to global researchers is a step forward.

More than technology, what gives me hope is the spirit of collaboration, integrity, and shared purpose I see in communities around the world. The real future of healthcare will be shaped not by algorithms alone, but by people—scientists, clinicians, advocates, and listeners—committed to equity.

To every young researcher, policymaker, or citizen reading this: don’t chase prestige. Chase truth. Chase justice. Your work matters, not just for citations or promotions, but for the lives it can improve.

Science that leaves people behind isn’t science at all. It’s time to build something better.

Because the future of healthcare belongs to all of us. And it starts with listening.