The artificial intelligence giant DeepMind is again making waves throughout academia with its AlphaGenome.
Leveraging the world-changing breakthroughs made with AlphaFold, which so recently revolutionized protein folding, AlphaGenome will set its sights on one of the great remaining unknowns of biology: the overwhelming majority of human DNA, which we tend to forget about, yet is present even in the non-coding part of the genome.
For decades, researchers had been trying to figure out the role of 98% of our genome that doesn’t direct the production of proteins. Once dismissed as “junk DNA,” these non-coding areas of the genome are being discovered to be central to controlling gene behavior, and there’s mounting evidence that changes in them cause or contribute to hundreds of diseases, from different kinds of cancers to rare genetic disorders.
Due to the complexity of motion of these regions, conventional approaches could not properly address them, thus constituting a critical bottleneck in genomics.
That’s where AlphaGenome a sophisticated AI model, comes into play — and changes the game. It can handle DNA sequences of mind-boggling length — a megabase long, up to a good million base pairs — and predict thousands of molecular properties at a degree of resolution never before possible.
That includes identifying where genes begin and end, how RNA is spliced and which proteins bind to DNA, in a wide variety of cell types and tissues. This overall predictive ability in one, combined model represents a significant advancement in comparison to distributed, multivendor approaches.
Preliminary results demonstrate the outstanding accuracy of AlphaGenome, which outperformed or was comparable to state-of-the-art models in almost all of the considered genomic prediction benchmarks.
In one impressive use-case study, the AI was able to correctly predict how a particular non-coding mutation could switch on the TAL1 gene, which is linked to cancer, in T-cell acute lymphoblastic leukaemia – a known disease mechanism to the letter.
This testifies to the great potential of AlphaGenome to help unravel the cause-effect relationships of non-coding variants and disease parameters.
Although available by API for non-commercial research, AlphaGenome is not yet cleared for clinical diagnostic applications. But there’s no question that it is having a near-term effect on speeding up discovery in disease research, synthetic biology, and even at the level of fundamental genomic knowledge.
Scientists throughout the world will be able to use this powerful tool to reveal never-before-seen genomic landscapes that play a role in causing disease, as well as to develop more targeted treatments with a never-before-seen velocity and precision. DeepMind’s AlphaGenome heralds an era of genomics where AI shines a light on the “dark matter” of our DNA leading to a step change in human health.