Home > AI
27 views 5 mins 0 comments

AlphaGenome 2 Launches to Predict DNA Changes Across Human Cell Types

In AI, Global
November 17, 2026
AlphaGenome 2 Launches to Predict DNA Changes Across Human Cell Types

AlphaGenome 2 was announced by Google DeepMind on November 11, 2026, as a new research model for predicting how DNA sequences and genetic variants influence gene activity in different human cell types. The company says it trained the model using more than 100 million individual cell profiles covering 68 tissues.

Unlike systems that predict average activity across an entire tissue, AlphaGenome 2 can condition its predictions on a chosen cell type or cellular state. Google DeepMind says the approach helps researchers investigate why a genetic change may affect one group of cells differently from another.

Read the Google DeepMind’s official AlphaGenome 2 announcement for the original details.

How AlphaGenome 2 Models Cell-Specific Gene Regulation

Every cell in the body contains essentially the same underlying DNA, but cells use different genes depending on their function. An immune cell and a heart muscle cell can respond differently to the same genetic sequence because their regulatory environments differ.

AlphaGenome 2 combines a DNA input with information identifying the relevant cell type. The company says the model can use a reference catalog or information derived from a researcher’s single-cell experiment. Its outputs predict aspects of gene activity, allowing scientists to generate hypotheses about how a genetic variant might operate.

Google DeepMind reports that the model identifies nearly twice as many gene-expression-changing variants as its previous state-of-the-art system under the company’s evaluations. This is a reported research comparison, not a claim that the system has been validated for clinical diagnosis.

Training Data Covers More Than 100 Million Cell Profiles

The scale of the training material is a notable part of the release. Google DeepMind says AlphaGenome 2 learned from profiles of more than 100 million individual cells drawn from 68 tissues. That breadth is intended to help the model represent biological differences that are lost in averaged measurements.

The company describes the system as a way to convert static cell atlases into a predictive tool. Researchers could examine candidate variants, compare predicted effects across cell populations and prioritize experiments. These outputs remain computational predictions that need verification in biological studies.

Potential Applications in Genetics and Disease Research

The model may help researchers explore the regulatory mechanisms behind inherited disease risks or differences between cellular states. It could also support work on identifying which cell types matter for a particular genetic association. Such possibilities depend on whether experiments confirm the predicted relationships.

AI models in biological research can fail when training data do not represent a relevant population, tissue or disease condition. Interpreting predictions therefore requires context about the available datasets and experimental limitations. A high score on a research benchmark is not equivalent to a diagnostic test suitable for patients.

Research Access and Commercial Availability

AlphaGenome 2 is available through an API for noncommercial research, while commercial research access is offered in private preview. Those access terms matter for universities, biomedical companies and research institutions planning to evaluate the model on their own datasets.

The announcement does not say that AlphaGenome 2 is a medical device or approved to make patient-care decisions. It should be assessed as a scientific research system until independent validation and applicable regulatory processes support a different use.

Researchers should also consider how cell profiles were collected and processed. Differences in sequencing methods, tissue sampling or cell annotation can affect the usefulness of a prediction, particularly for rare or underrepresented cellular states.

The ability to compare predictions across cell types is potentially valuable because disease-associated variants often lie outside protein-coding regions. Understanding where a regulatory effect occurs can guide which laboratory experiment to prioritize, but the model cannot independently establish causation.

What Happens Next

Researchers will test how well AlphaGenome 2 predictions replicate across independent cell experiments and difficult genetic questions. Comparisons with established laboratory techniques and other sequence models should help identify where the system is useful and where uncertainty remains.

The confirmed development is a cell-aware DNA prediction model with broad training data and research access. Its ultimate value will depend on reproducible discoveries, transparent evaluation and careful separation of research hypotheses from clinical conclusions.

Read Also