Understanding Life at Every Scale

Discovering how life works has been one of humanity’s grand scientific pursuits, and has led to people’s quality of life improving everywhere. We now better understand disease, physiology, and human health. This quest has required deciphering life’s rules across a vast spectrum — from the microscopic language of life, DNA, to its fundamental building blocks, proteins, to the assembly of these elements into cells, tissues, and entire organisms. This is an immense level of scale and complexity, and untangling it requires analyzing data and mechanisms far beyond what a single human mind can comprehend.

We believe AI is uniquely suited for overcoming these challenges and advancing our understanding of life’s fundamental questions. In the decade since we began our work on using AI to decipher biology, we have seen enormous success. Yet, our vision for the future of biological research is even more expansive.

We must not only map how life functions in its entirety, knitting together our knowledge of DNA, genomes, proteins, and cells into one coherent view. We must also use this complete understanding to advance human health and work on critical challenges, such as complex diseases. As AI’s capabilities for biological research advance in the coming months and years, they will have a profound impact on both life sciences research and society.

Soon, we will share our next major steps: scaling up and widely distributing our knowledge of the human genome, releasing novel systems to control and design biology itself, and establishing new ways to trace and authenticate AI-designed molecules.

Our research teams have some of the broadest and deepest expertise in both AI and biology. We first began our investigation into the science of life almost eight years ago, by tackling the grand challenge of predicting the 3D structure of proteins. AlphaFold, our protein structure prediction system, made a historic breakthrough on this problem — a milestone recognized by the 2024 Nobel Prize in Chemistry. Today, more than 3.5 million researchers have accessed AlphaFold’s 260 million protein predictions, which have proven vital for efforts ranging from combating antibiotic resistance to understanding heart disease. AlphaFold was an important first step, yet we still needed a full recipe book for life.

Since then, we have expanded our family of domain-specific AI models to operate across multiple biological scales, broadening our knowledge of how life functions. AlphaMissense predicts how single genetic mutations cause disease, while AlphaFold 3 maps how complex biomolecules dock and interact. 

One of our most significant steps was AlphaGenome, which deciphers how the genomic code affects biology and human health. The model works at the vast scale of the human genome, taking DNA sequences of up to 1 million letters, also known as base-pairs, and predicting thousands of molecular properties that dictate genetic effects. It can analyze every possible combination of base-pairs, which runs to billions of combinations. Soon, we will make these predictions even easier to access, just as we did with the AlphaFold database.

Beyond advancing biology, our general AI tools like Gemini and our agentic systems can help researchers across every domain of science: scanning literature and processing data at speed, generating hypotheses and finding needle-in-a-haystack solutions. A unique feature of Google DeepMind’s work is that we have developed systems across all these categories, from general to narrow AI, empowering researchers to do work across the biological spectrum.

There are certainly challenges to overcome, and it would be foolish to underestimate life’s complexity and the enduring importance of real world experimental validation, but this represents a profound shift. We are moving toward a world in which we will have a comprehensive understanding of the language of life, alongside increasingly fine control over its function, such as allowing us to improve proteins and produce them from scratch.

This capability clearly will have incredibly positive consequences for human health and biomanufacturing. However it also brings distinct risks, which is why we are working on ensuring that our systems are beneficial and are not used to cause harm, like helping DNA synthesis providers screen for risks and partnering with biosecurity experts to utilize models like AlphaGenome for faster pathogen detection. 

Our success would not be possible without the support of the research community, from providing public datasets which help to power our models, like AlphaFold, to working in close collaboration with research partners to determine how we can protect the entire ecosystem from bad actors. Through our upcoming work on tracing biological authenticity, we aim to support shared governance frameworks for responsible biological design.

Consequential technology must also benefit as many people as possible. This has been central to our mission from the start — we, and our partners like the EMBL’s European Bioinformatics Institute, have ensured foundational tools like the AlphaFold Database remained widely accessible — and we will continue to democratize knowledge in this way. 

Already, Google DeepMind’s models are at the frontier of the life sciences, and the technology we have today is undeniably powerful. But soon, we will have much more capable models. The task of mapping life’s complexity at every level — of joining together scales from the microscopic to the whole human — is difficult, and we must walk this path carefully. But rather than an insurmountable barrier, this complexity is a code waiting to be fully deciphered. By unlocking it, we will enter a transformational era of biological research.