Science Breakthrough of the Year: Artificial intelligence algorithms reveal 3D shapes of proteins

[SkyKnight2]

Summary

The 2021 Science Breakthrough of the Year was announced as the artificial intelligence-driven algorithms that predict the 3D shapes of proteins with only the input of each protein's sequence of amino acids. Proteins are dynamic structures made up of sequences of individual building blocks called amino acids. The complex interactions between these amino acids determine the protein's final exquisitely complex 3D shape (see image below of how AI predicted how two proteins form a complex involved in DNA repair in yeast). Given the sheer number of possible interactions between each individual amino acid, even modest-size proteins could assume an astronomical number of possible shapes, making it almost impossible to predict. However, with recent advancements of AI-driven algorithms, the structures of proteins are being solved at lightning speed. This year, the AI program DeepMind reported solving the structures for 350,000 proteins found in the human body (44% of all known human proteins).

 

 

Quotes

Quote

Protein structures could once be determined only through painstaking lab analyses. But they can now be calculated, quickly, for tens of thousands of proteins, and for complexes of interacting proteins.

My thoughts

This is a huge accomplishment that will advance many different fields. One such application is the design of new drug candidates. Currently, scientists are even using it to model the effect of mutations in the Omicron variant's spike protein (to predict whether or not antibodies will be able to neutralize it).

 

Computer programmers do not get the credit they deserve!

 

Sources

 https://www.science.org/content/article/breakthrough-2021#section_breakthrough

[Sauron]

Finally some AI development that is not just "I can draw a human looking face semi-reliably"

 

I wonder if this makes F@H obsolete...?

[CarlBar]

9 minutes ago, Sauron said:

Finally some AI development that is not just "I can draw a human looking face semi-reliably"

 

I wonder if this makes F@H obsolete...?

 

No, this process gets computationally wepensive really fast. it might be that F@H switches over to it mind. Also useful other article from when the results where first announced.

 

https://www.science.org/content/blog-post/protein-complex-structure-predictions-already

[Quackers101]

Also for deepmind into the google metaverse, in alphabetical order.

 

Well they have been doing some fun things, here they talk about their free public database of protein structures.

https://www.youtube.com/watch?v=Gk-PyJSNlcI

 

https://alphafold.ebi.ac.uk/

[Doobeedoo]

So when will I be able to rapidly self repair like lizard?

[SkyKnight2]

Quote

DeepMind reported using several GPUs for days to make individual predictions, whereas our predictions are made in a single pass through the network in the same manner that would be used for a server; after sequence and template search (~1.5 hours), the end-to-end version of RoseTTAFold requires ~10 min on an RTX2080 GPU to generate backbone coordinates for proteins with fewer than 400 residues, and the pyRosetta version requires 5 min for network calculations on a single RTX2080 GPU and an hour for all-atom structure generation with 15 CPU cores.

Computational costs are always to be kept in mind. The above is a quote from the original August 2021 paper from Science magazine where researchers first published RoseTTAFold. It apparently requires much less computational power than DeepMind's AlphaFold. Although with the current chip shortage, many may have to rely on F@H still. I know my 970 couldn't handle it (lol).

 

https://www.science.org/doi/10.1126/science.abj8754

[Beskamir]

I'm pretty sure protein folding is NP hard so I doubt this will actually be particularly accurate on proteins it hadn't seen before without having to take an insane amount of time computing them. My reasoning is that if this works flawlessly then all NP hard problems become solvable by this AI and that would probably cause a lot of issues since last I checked AI is not actually magic.

[Brooksie359]

13 hours ago, Beskamir said:

I'm pretty sure protein folding is NP hard so I doubt this will actually be particularly accurate on proteins it hadn't seen before without having to take an insane amount of time computing them. My reasoning is that if this works flawlessly then all NP hard problems become solvable by this AI and that would probably cause a lot of issues since last I checked AI is not actually magic.

While AI isn't magic what it can do is pretty magical in nature. 

[StDragon]

Now do the reverse. Analyze a theoretical protein into an mRNA sequence that a cell can manufacture.