In the last episode of The Genetics Podcast, Patrick spoke with Dr. Slavé Petrovski, Vice President of AstraZeneca’s Center for Genomics Research. Slavé oversees one of the largest human genomics resources in industry and uses it to support target discovery, biomarker development, disease prediction, and other programs of work. The conversation covered his career path, how large biobanks change the questions we can ask, how to identify protective biology in populations, and why computational sustainability now matters as much as technological advances.
Slavé originally enrolled in a business information systems degree, then changed direction after a simple piece of advice from his father to pursue what he genuinely enjoyed. He switched into genetics while keeping his computing and statistics training. That combination shaped his academic work on epilepsy genetics and eventually positioned him to move into large scale human genomics.
He joined AstraZeneca just after the company launched its genomics initiative in 2016. The aim was to build a large, patient centred resource that could inform decisions throughout drug discovery. Slavé expected a short stint in industry but stayed because the science kept evolving. He also notes that the gap between academic and industry research has narrowed and that much of his work relies on collaborative science.
Working with large biobanks has made clear that scale is essential for robust rare variant analysis. Individually, rare variants do not offer enough statistical power, but gene-level analyses across hundreds of thousands of genomes do. These datasets also allow direct links between natural genetic variation and long-term health outcomes.
Diversity has been equally important. In 2021, AstraZeneca found that most of its sequenced participants were of European ancestry. The team then built partnerships in Mexico City, South Asia, East Asia and Africa to broaden representation. About 40% of participants are now of non-European ancestry. This shift reflects ethical responsibility but also improves discovery by enabling studies of variants and haplotypes that are uncommon in European populations.
Population cohorts allow researchers to look beyond disease risk and study people who remain healthy despite strong environmental or genetic pressures. These resilient individuals can reveal mechanisms worth targeting therapeutically. Classic examples include loss of function variants in PCSK9 and their effect on LDL cholesterol and cardiovascular risk.
Slavé’s team and others have identified similar patterns in MAP3K15 for diabetes, TSLP for asthma and IL23R for autoimmune conditions. In many cases, the genetic signals match what is already known from drugs in the clinic. In others, they point toward new opportunities. Slavé believes this approach shortens the path between genetic evidence and practical drug development.
MILTON, one of the group’s core tools, began as a way to improve case definitions in biobank studies. By learning biological signatures that distinguish cases from controls, the model could identify likely missed cases and improve the quality of genetic analyses. During testing, the team saw that the same models could also predict future disease.
Because UK Biobank includes long-term follow up, MILTON can identify signals in genetics, biomarkers and proteomics that appear many years before clinical diagnosis. It now predicts the onset of more than one thousand disease endpoints, sometimes up to 15 years before clinical diagnosis. Slavé sees this as a path toward more proactive care where high risk individuals can be monitored earlier and offered suitable interventions when they exist.
Modern genetic studies involve trillions of statistical tests. If executed with standard tools, they are costly and can produce significant carbon emissions. Slavé’s team addressed this by re-engineering core statistical methods for large, stable datasets and removing steps that are no longer necessary.
These redesigned tests reduced compute time by more than 450-fold and led to over 99% lower estimated carbon emissions while keeping test results highly consistent with existing methods. Slavé hopes this motivates more groups to consider sustainability when building new analytical tools as datasets continue to grow in size and complexity.
AstraZeneca releases many of its summary statistics and results publicly to help accelerate follow up research. Their PheWAS portal, MILTON, and other tools receive heavy use from academic and industry labs worldwide. Sharing results reduces redundant computation and helps identify potential collaborators for future work.
Slavé is aware of the risk that open resources can drive everyone toward the same ideas. Within his team, he encourages people to question standard assumptions and focus on simple, defensible approaches rather than unnecessary complexity. He sees this as essential for avoiding false positives and staying open to unexpected results.
Looking ten years ahead, Slavé expects genomics to become more integrated into routine care, at least for conditions where it improves diagnosis or management. As sequencing costs fall, one time germline profiling could support lifelong risk assessment and monitoring. He also expects richer longitudinal data, including proteomics, metabolomics and digital health, to shape more personalized models of health trajectories.
Genetic evidence is strong across many therapeutic areas. The next step is to convert the most meaningful findings into treatments and care pathways that reach patients earlier and deliver measurable benefit.
Listen to the full episode below.