Did you know we only understand the use of approximately 2% of our DNA? The remaining 98% is often written off as ‘junk’.
Thanks to AI, this might be about to change.
This ‘junk DNA’ may hold answers to understanding disease, at least that's what Google DeepMind are hoping for with the release of their new AlphaGenome model.
What is AlphaGenome?
AlphaGenome is a new AI model developed by Google DeepMind. It aims to revolutionise our understanding of the human genome.
It will help to accurately predict how genetic variants or mutations in human DNA sequences impact the biological processes that regulate genes.
This will help with identifying the genetic causes of diseases, developing new treatments, and deepening our understanding of how the human genome works.
AlphaGenome marks a significant breakthrough in comparison to previous models that focused on protein-coding regions (which make up only about 2% of our genome).
AlphaGenome makes waves by assessing the vast 98% of non-coding DNA. This is crucial for gene regulation and is where most disease-associated variants are found.
Long sequence-context
A key breakthrough of AlphaGenome is its ability to analyse very long DNA sequence while maintaining high, single-letter resolution, In simple terms this means s means AlphaGenome can look at a very long stretch of DNA and still understand the impact of even a tiny change, such as a single letter being different, no matter where it is within that long sequence.
This overcomes a major limitation of previous models that had to sacrifice one for the other, ultimately leading to a more complete picture of genetic function.
Multimodal prediction abilities
AlphaGenome also boasts comprehensive multimodal prediction abilities. This means the model AlphaGenome can predict details about how genes are controlled in every part of the body.
This includes how active a gene is, exactly where a gene's instructions begin and end. As well as if a gene's DNA is "open" and ready to be read. It also confirms where important proteins stick to the DNA to control genes. Vitally it shows how gene instructions are cut and glued together.
Mutations in this cutting and gluing process can cause diseases like spinal muscular atrophy or cystic fibrosis, and AlphaGenome helps us understand why.
Efficient variant scoring
Efficient variant scoring means the model can quickly assess the impact of a genetic variant on all different properties by comparing predictions between mutated and unmutated sequences.
Put simply, this means the model will help predict what impact a gene mutation will have on a person, often times incredibly quickly in less than a second.
What Will AlphaGenome Be Used For?
The powerful new AlphaGenome model will primarily be used in scientific research. It will particularly help scientists to understand the non-coding regions of the human genome.
The immediate value is obvious as the model will help identify genetic causes of disease. Pin-pointing how tiny genetic changes even in non coding DNA lead to various diseases, including both rare genetic disorders and more common complex diseases like cancer, is crucial step on the journey to lifesaving medications.
When the variant is first detected, AlphaGenome can quickly assess its impact. By understanding the exact molecular mechanisms disrupted by disease-causing variants, researchers can identify new points in the biological process that could be targeted by new drugs or therapies.
On a more theoretical level, AlphaGenome will also help create more detailed maps of the human genome. It will identify crucial "control elements" in DNA that tell genes how to operate.
Looking to the future, Alpha Genome may assist in the design of new, synthetic DNA sequences with specifically designed functions. For instance, researchers might want to create a DNA sequence that only activates a gene in brain cells, but not in muscle cells, for targeted therapies. AlphaGenome could be used to predict if their designed sequence will be able to achieve its goals.
How To Use AlphaGenome?
It’s important to remember that AlphaGenome is a research tool and is not currently approved for clinical use.
However, the model is now available via an API for non-commercial use to help to advance scientific understanding, which DeepMind hopes will eventually lead to clinical applications.
- Visit the AlphaGemone folder on the DeepMind Github repository.
- Obtain an API Key from Google DeepMind. This key provides you with programmatic access to the AlphaGenome model. The API is currently free for non-commercial use
- You can chose whether to clone the AlphaGenome GitHub repository to your local machine and then install it using pip or use Google Colab notebooks quickstart.
- Once installed and set up with your API key, you can make predictions by writing code typically in Python.
Comments ( 0 )