Google DeepMind and Google.org have announced a major partnership with the UK's Wellcome Sanger Institute to create high-quality genomic datasets that will serve as training material for next-generation AI models. The consortium, funded at $5 million per year for five years, aims to address a critical gap in the life sciences: while AI tools have transformed protein prediction and drug discovery, many areas of genomics lack the curated, indexed datasets needed to train powerful new algorithms.

Why Does the Life Sciences Community Need Better Genomic Datasets?

The challenge facing AI researchers in biology is straightforward but significant. DeepMind has already released powerful tools like AlphaGenome, which can predict the function of DNA sequences and go beyond simple expression patterns to forecast DNA accessibility and transcription-factor binding. The company also unveiled Co-Scientist, a multiagent AI platform that can scan existing literature and generate new scientific hypotheses. However, all of these tools depend on open-access datasets that are properly organized and curated.

Not every corner of the life sciences has such resources. This consortium directly addresses that gap by creating datasets that will be shared widely with the scientific community, enabling researchers worldwide to build and train new AI models without starting from scratch.

"The consortium aims to create resources that will be shared widely with the community to enable transformative scientific discoveries and deliver broad impact across the life sciences," stated Julia Wilson, chief innovation and impact officer at the Sanger Institute.

Julia Wilson, Chief Innovation and Impact Officer at the Wellcome Sanger Institute

What Tools Has DeepMind Already Built on These Datasets?

DeepMind's track record in genomics and protein science provides context for why this partnership matters. The company is best known for AlphaFold, the protein structure prediction software that won widespread acclaim for its ability to predict how amino acids fold into three-dimensional shapes. More recently, DeepMind released AlphaGenome in January, a publicly available model that predicts DNA sequence function with remarkable detail.

According to Žiga Avsec, a Google DeepMind researcher and lead author of the AlphaGenome paper, the model can predict more than just whether a gene is expressed. It can forecast detailed aspects like DNA accessibility, which determines whether proteins can access specific DNA regions, and transcription-factor binding, which controls how genes are turned on and off.

How Will This Partnership Accelerate AI Biology Research?

• Standardized Datasets: The consortium will create indexed and curated genomic datasets that researchers can use immediately, eliminating months of data preparation work before training AI models.
• Democratized Access: By sharing these resources widely with the scientific community, the partnership removes barriers that currently prevent smaller labs and institutions from developing their own AI tools for genomics.
• Foundation for New Models: High-quality training data is the foundation for building better AI models; this consortium provides that foundation across multiple areas of genomics that currently lack such resources.

The five-year timeline reflects the scale of the undertaking. Creating datasets that meet the standards required for training modern AI models requires careful curation, quality control, and ongoing maintenance. The $5 million annual funding from Google.org and DeepMind signals serious commitment to this infrastructure work, which often goes unnoticed but underpins major scientific breakthroughs.

This partnership also reflects a broader shift in how AI companies approach biology. Rather than keeping proprietary datasets locked behind closed doors, DeepMind and the Sanger Institute are betting that open, shared resources will accelerate discovery across the entire field. The Sanger Institute, a world-leading genomics research center, brings decades of expertise in sequencing and data curation, while DeepMind brings cutting-edge AI capabilities and resources.

The timing is significant. As AI tools become more powerful and more researchers want to apply them to biological questions, the bottleneck is increasingly not the algorithms themselves but the quality and availability of training data. This consortium directly addresses that constraint, potentially unlocking new discoveries in gene function, disease mechanisms, and therapeutic targets that depend on understanding how DNA sequences actually work in living cells.