AlphaFold 2- The Great Hype
You might have read the recent ‘AI breakthrough’ or ‘It will change everything’ or ’50-year-old grand challenge solved’ kind of headline which seems cool and it is but most of the hype is created by media. As the paper has not been published yet so we do not have the full details. Here I will give a brief overview of everything around this problem. This blog is divided in three subcategories:
- What is Protein-folding?
- What did Alphafold-1 do?
- Why is everyone talking about it?
What is Protein-Folding?
The wiggly thing you see above is the fundamental building block of life. The one that created us and all of the living organisms around us. Each protein is a series of amino acids connected and oriented in such a way it achieves maximum stability. Nearly every function our body performs can be traced back to one or more proteins and how they move and change. Which are controlled by genes encoded in DNA.
Protein functionality highly depends on its 3-D structure. The main reason why a protein folds is because of molecular forces like Van der waals, H-bond and many other play a major role in this world. Each molecular attraction/repulsion plays a significant role in its structure and thus in chemical properties.
What did Alphafold-1 do?
DeepMinds’s AlphaFold-1 has outperformed in progress and accuracy to predict complex protein structures as compared to other methods of protein structure prediction. Which was also submitted to CASP — ranked at top in ranking, focused on specifically on the hard problem of modeling target shapes from scratch, without using previously solved proteins as templates — achieving a high degree of accuracy when predicting the physical properties of a protein structure, and then used two distinct methods to construct predictions of full protein structures.
Both of these methods relied on deep neural networks that are trained to predict properties of the protein from its genetic sequence. It uses a CNN architecture. The accuracy for these model measured in GDT(Global Distance Test) was around ~58%
Why is everyone talking about it?
The last version of alphafold did not match the GDT of the conventional structure prediction technique like X-ray crystallography holding accuracy of above 90%, and it was considered that it would take a few decade to solve the problem but here come Alphafold with a GDT(acccuracy) of 92.4% for CASP14. These exciting results open up the potential for biologists to use computational structure prediction as a core tool in scientific research.
For the latest version of AlphaFold, used at CASP14, they created an attention-based neural network system, trained end-to-end, that attempts to interpret the structure of this graph, while reasoning over the implicit graph that it’s building. The major difference between the two is that this uses a transformer based approach then conventional CNN. We will have more detail soon as they publish their paper.
