DeepMind AlphaFold: using AI for scientific discovery
After several years of work, in January 2020 DeepMind has released AlphaFold — which predicts the 3D structure of a protein based solely on its genetic sequence. The 3D models of proteins that AlphaFold generates are far more accurate than any that have come before.
Protein Folding problem
Proteins are complex molecules, which we encounter everywhere and which are crucial to life.
Genes are recipes for proteins and are encoded in DNA.
An error in the genetic recipe may result in disease or death, that’s why understanding proteins is crucial to our understanding of life.
But knowing genetic recipe for a protein doesn’t mean you know its shape right away. Proteins are comprised of chains of amino acids.
But DNA only contains information about the sequence of amino acids–not how they fold into shape. That’s why we talk about protein folding problem, which is a problem on which scientists have worked for decades.
We want to know how proteins fold. And the bigger the protein, the more difficult it is to model, because there are more possible interactions inside. .
In their recent paper published in Nature, DeepMind has managed to train a neural network to make accurate predictions of the distances between pairs of residues, which convey more information about the structure than contact predictions.
DeepMind has trained deep neural networks to predict properties of the protein from its genetic sequence. The properties our networks predict are:
- the distances between pairs of amino acids and
- the angles between chemical bonds that connect those amino acids.
The first development is an advance on commonly used techniques that estimate whether pairs of amino acids are near each other.