The American Society of Human Genetics held its 2025 annual meeting October 14-18 in Boston. The dominant theme was clear: AI is transforming genomics, but the promise remains unevenly distributed across populations and geographies.
If you work in computational biology or genomics, ASHG revealed both the near-term trajectory and the harder problems ahead.
What Happened at ASHG 2025
The conference brought together thousands of geneticists, genetic counselors, lab directors, and computational biologists for the field’s annual summit. Three themes stood out across the sessions.
AI is moving from research to production. The Distinguished Speakers Symposium on “AI-Powered Genomics” showed machine learning accelerating disease discovery and personalizing treatment. But the hard question surfaced repeatedly: how do you validate models across populations different from your training data? The field increasingly recognizes that algorithms trained on European ancestry samples often fail on other populations.
Rare disease diagnosis just got faster. Sessions showcased how long-read sequencing and multi-omics are cutting diagnosis time for rare genetic disorders. Long-read technology (think Oxford Nanopore, PacBio) enables detection of complex variants that short-read sequencing misses. If you work on rare disease genomics, these workflows are production tools now, not research pilots.
Population genetics is driving precision medicine. The Presidential Symposium highlighted how DNA from Neanderthals and Denisovans shapes modern disease risk. The implication: accurate disease prediction requires understanding human genetic diversity, not just individual variants. This reshapes how you think about reference genomes and variant interpretation.
What This Means for You
If you’re a PhD student or postdoc in computational biology, the ASHG conference message is clear: AI literacy is no longer optional. You don’t need to be a machine learning researcher, but you need to understand model validation, bias detection, and generalization across populations. The field is asking harder questions about whether your carefully tuned algorithm actually works when applied to data it wasn’t trained on.
If you work in clinical genomics or rare disease diagnosis, the acceleration in long-read and multi-omics workflows means your sequencing pipelines and annotation workflows need to handle greater data complexity. The tools are more powerful. The expectations are higher.
If you’re interested in precision medicine, the unspoken subtext at ASHG was unavoidable: most genomic databases represent European ancestry individuals. The field knows this is a problem. Funding is flowing toward population-specific studies. But if you’re developing tools, algorithms, or diagnostic pipelines, you need to think early about generalization and validation across ancestry groups. This isn’t just an equity issue. It’s a technical issue that will affect your product.
The biosecurity policy forum also signaled that computational biologists should expect increased scrutiny and guidelines around how genomic tools are built and deployed. If you’re publishing genomic analysis code or building open-source bioinformatics tools, understanding the emerging biosecurity landscape will matter.
The Takeaway
ASHG 2025 laid out the next 18 months of genomics research. If you’re developing bioinformatics tools, consider where you stand on three fronts: Are you handling long-read data? Are you validating across diverse populations? Do you have a plan for AI integration or deployment?
Complete Genomics and SOPHiA GENETICS announced in November a partnership bringing AI-powered sequencing to production oncology testing. This is ASHG’s vision becoming commercial reality.