Historically, genetics has had a big problem with our body and the instruction manual that we have in each of the cells and that gives us the possibility of living: DNA. Until now, We could only understand well 2% of all our genetic materialleaving the rest of the information in a mess drawer that came to be called “junk DNA” because we did not understand what function it had in our body. But this has changed thanks to technology.
The solution. Google has wanted to collaborate with science to understand much better what 98% of our DNA does that it does not encode proteins and that we did not know its reason for being. But evolutionarily, if it has not been lost over the generations, it is because it must have had some relevant function.
In order to shed light on this ‘dark’ region we now have AlphaGenome, an AI model that is capable of reading massive fragments of our DNA and predicting with great precision how small mutations can alter cellular machinery and cause diseases. Like a cancer.
What we know. As we have said before, genetics knows that only 2% of the human genome contains instructions for making proteins. The rest of the DNA was a mystery for a long time until it was recently discovered that “switches”, known as enhancerswho decide when, where and how much a specific gene is expressed.
The problem arose when the genetic variations in these areas were difficult to interpret due to the great diversity of molecular consequences that it can cause. Until now, computational tools had to choose: either they analyzed very short sequences in great detail or they looked at long sequences in little detail. AlphaGenome has broken that barrier.
A million letters. In a very colloquial way, we can understand that human DNA is made up of letters (which are the different nitrogenous bases) and their combination literally generate a language. In this case, AlphaGenome’s great technical innovation is its ability to “zoom” and “pan” at the same time. The model takes as input 1 megabase (1 Mb) of DNA sequence, which is one million letters, and predicts thousands of functional genomic clues at single base pair resolution.
And this is a vital range, since genetic regulation occurs at a distance. And although in our minds we can imagine that the ‘switches’ of the different genes are right next to or above their targets, the reality is that they can be very far from it. In fact, 99% of validated enhancer-gene pairs fall within this 1 Mb range.
Its importance. It may sound like technical science fiction, but the impact of an AI understanding our “junk code” has very real consequences for patients and medicine in the near future. And AlphaGenome is not just a tool for biologists; It is a key to unblock personalized medicine.
One of the most important points is in what are known as ‘rare diseases’ where dedicating a research team is unfortunately not worth it due to the few people who have a disease. In this case, AI, by interpreting all the genetic material, can show many answers ‘hidden’ in that non-coding dark matter to offer new diagnostic avenues.
Computer drug. The tool promises to accelerate the design of advanced therapies. By predicting how a DNA sequence affects splicing (splicing) or expression, can be used to design drugs that act on the products generated by the genetic machinery of our cells. In this way, the door is opened to these precise and very non-invasive treatments by acting on a specific protein, which reduces its side effects.
Decrypt cancer. As has been demonstrated with leukemia, the model allows us to understand the most complex genetic mechanisms of this disease in which we are facing a constant struggle. It doesn’t just say “there is a mutation,” but explains how that mutation breaks cellular rules to activate an oncogene that is the precursor to future cancer. This is vital to develop drugs that attack the root of the problem and not just the symptoms.
Your availability. True to the philosophy of “open science” (with nuances), DeepMind has released the model code and weights for research, in addition to provide an online API for non-commercial use. Although logically AlphaGenome is not the end of the road, it does represent the most detailed map that we have ever woven of that “dark matter” that makes us human.
Images | digitale.de
GIPHY App Key not set. Please check settings