Folding Forum

Nert wrote:Thanks for this link. As a layman I found it mostly understandable. I think that I now understand the difference between Alpha Fold and FAH: Alpha Fold predicts what a protein will look like after it's folded. FAH determines the individual steps a protein goes through to get to that final shape.

Exactly .

For the specific example of COVID19, there's a vulnerable "mouth" under the head of the "spike" which "opens up" periodically that can be attacked by specific protein fragments, but it's closed most of the time. There are pictures of that process in some of FAH's associated supporting documentation that works for a layman like myself.

Finding the final shape is still an important goal for many scientific studies -- but from my perspective, that's the easy part. I really appreciate the finger-shadow-puppet description of X-ray chrystolography. It is a great way to describe the traditional methodology to the layman.

Can we expect in futur a fahclient using an AI engine ?

I say this because recently :

— i read some documentation about Languagetool (see languagetool.org ).
An Advanced user can set up a server which is easy but also a feature named "languagetool-neural-network" (see https://github.com/gulp21/languagetool-neural-network ) which is not easy. So today we can use AI software in our PC.

— there are announces about AI chipset for our mother board or included in processor as gpu.

— last but not least Google announces a breakthrough about protein folding with AI.


Do you explore this technology ?

There are several discussions of the AI breakthrough reported by Google. Basically they're solving a different problem than FAH is solving. Both sets of scientific goals are worthy of research.

So far, AI doesn't seem to help FAH achieve its scientific goals.

A lot of what we've been doing with Folding@home recently has been focused on understanding mechanisms underlying disease by studying conformational transitions in the presence and absence of cofactors or disease-relevant mutations, or predicting how new molecules might bind and inhibit disease targets. I'm optimistic that the advances from AlphaFold2 will eventually make it into our system preparation pipelines to assist us in more rapidly setting up calculations that address these problems, but it's still a long way off even from being a practical tool in the toolchain for that purpose.

Still, I'm excited to see more details once they choose to actually report what they did in any useful detail!

~ John Chodera // MSKCC

An update on AlphaFold.

https://www.deepmind.com/blog/alphafold ... n-universe

I'm curious if anyone knowledgeable read through the AlphaFold 2 July update & has any comments on it.
https://www.deepmind.com/blog/alphafold ... n-universe

If this method truly allows revealing every structure of the protein universe, how relevant is FAH & how does it affect its future?

DeepMind is concentrated on finding final structures, what the protein will look like when fully folded. It does not work on the unfolded state, or the transitions the protein goes through while changing from the unfolded state to the folded one. From the description of the results it also is not clear if DeepMind also identifies alternate folded states.

The results it does provide though are useful for creating starting conditions for simulations done by F@h. F@h does find various stable folded states that have various periods of persistence. It also can be used to identify probable transition paths between the various stable folded states.