Artificial intelligence is increasingly being used for something much broader than generating text or images.
Scientists are using AI to search through enormous biological datasets, identify patterns and generate predictions that would be difficult to obtain manually.
One of the most striking examples is genomics.
The human genome contains billions of DNA letters.
Understanding what each sequence does is an enormous scientific problem.
AI is beginning to provide researchers with tools capable of exploring that space at unprecedented scale.
The complexity of DNA
The human genome contains roughly three billion base pairs.
Only a small portion directly codes for proteins.
Much of the remaining DNA regulates when and where genes are activated.
For decades, scientists have understood that this regulatory DNA is important.
But determining exactly how individual genetic variants affect biological processes can be difficult.
Traditional experimental approaches can take substantial time.
AI offers another route.
AlphaGenome
Google DeepMind’s AlphaGenome is designed to predict how genetic variants affect molecular processes.
In September 2026, Google introduced AlphaGenome Atlas, a database containing predictions for the effects of every possible single-nucleotide variant in the human genome.
That represents approximately 9 billion possible single-letter changes.
Google says the resulting dataset is about one petabyte in size. (blog.google)
The scale is extraordinary.
Instead of examining variants one by one, researchers can use the atlas to prioritize potentially important mutations.
Why non-coding DNA matters
Scientists understand many protein-coding genes relatively well.
But the majority of the genome does not directly encode proteins.
These regions can act like regulatory switches.
They can influence gene expression.
A mutation in one of these regions may alter how a gene behaves without changing the protein itself.
That makes non-coding DNA extremely important to understanding disease.
AlphaGenome is designed to analyze these regulatory effects.
Turning enormous search spaces into manageable problems
Scientific research often involves searching through enormous possibilities.
Drug discovery is one example.
Genomics is another.
Researchers may have tens of thousands of potential variants.
Testing every one experimentally can be impractical.
AI can rank candidates.
That does not replace experiments.
Instead, it helps scientists decide which experiments are most informative.
Google describes AlphaGenome Atlas as a tool for variant prioritization. (Google DeepMind)
Research applications
Google reports that researchers are already using AlphaGenome to investigate rare variants and complex traits.
One example involved researchers at the University of Exeter analyzing data from more than 54,000 UK Biobank participants to investigate non-coding variants associated with traits. (Google DeepMind)
Another example examined genetic signals associated with Alzheimer’s disease and their activity across immune and metabolic tissues. (Google DeepMind)
These examples demonstrate a broader trend.
AI models can connect genomic variation with biological systems rather than treating DNA as a simple sequence.
AI as a scientific collaborator
AI is increasingly becoming part of scientific workflows.
Researchers can ask models to analyze data, generate hypotheses, identify relationships and prioritize experiments.
Google’s broader scientific AI work includes systems designed to assist with scientific reasoning and research workflows. (Google DeepMind)
The role is similar to that of a research assistant, although the underlying technology is fundamentally different.
The scientist remains responsible for interpreting evidence.
Prediction is not proof
An important distinction must remain clear.
An AI prediction is not automatically a biological fact.
Models can make mistakes.
Predicted relationships need experimental validation.
AlphaGenome can identify potentially important variants, but researchers still need laboratory and clinical evidence to determine what those variants actually do.
This distinction will remain critical as AI enters scientific research.
The scale advantage
One of AI’s greatest scientific advantages is scale.
A human researcher can examine a limited number of hypotheses.
A computational system can evaluate millions or billions of possibilities.
That changes the economics of discovery.
Instead of asking:
“Which possibility should we test?”
scientists can increasingly ask:
“Which of these millions of possibilities deserve experimental validation first?”
From genomics to medicine
The long-term implications could be substantial.
If researchers can better understand genetic variants, they may improve diagnosis and disease research.
AI could help identify biological mechanisms involved in rare diseases.
It could also contribute to drug discovery.
But there is a long path between identifying a biological signal and developing a safe treatment.
Clinical development still requires extensive testing.
Open scientific access
Another notable aspect of AlphaGenome Atlas is accessibility.
Google says the atlas is available to academic researchers through a free portal and provides API access for computational workflows. (Google DeepMind)
That could allow researchers without large infrastructure budgets to explore the dataset.
Scientific innovation increasingly depends on access to computational resources.
The rise of scientific foundation models
AlphaGenome is part of a larger trend toward foundation models designed specifically for scientific domains.
Instead of training AI only on human language, researchers are training models on:
- DNA
- proteins
- molecules
- weather systems
- scientific literature
- physical simulations
These models can become computational instruments.
A new relationship between AI and science
The most important change may not be that AI becomes a scientist.
Instead, scientists gain a new type of instrument.
A microscope allowed researchers to see structures they could not see before.
A particle accelerator allowed them to probe matter at extraordinary energies.
AI can provide another form of scientific visibility.
It can identify patterns across datasets that are too large or complex for humans to analyze directly.
The next stage
The future of AI-driven science will depend on the combination of models, experimental systems and human expertise.
AI can generate hypotheses.
Robotic laboratories can perform experiments.
Scientists can interpret results.
The cycle can repeat much faster than traditional research workflows.
That possibility could transform biology.
AlphaGenome Atlas is an early example of what happens when an AI system is used not merely to answer questions but to map an enormous scientific search space.
The broader innovation is therefore bigger than one genomic database.
It represents a shift toward AI-powered scientific discovery, where computation becomes an active part of the experimental process.