Calculating 3-D protein structures

Thanks Ninshub,

I'll just add a couple of extra thoughts.

My description of RNA was deliberately simplified a bit - for example there are things called INTRONS inside most genes, and these trigger some splicing and editing of the RNA chain before it is used as a template to create a protein.

The other point I'd like to emphasis is the need to explain where the code in each gene comes from. This can't possibly be the result of some sort of chance process. Nobody on Earth knows how to do it, because it is super hard. Remember that that protein chain has to fold up into a 3D structure, and it is that which does whatever it is meant to do. The parts of an enzyme that actually do the work usually involve amino acids at several points in the chain that only come together as a result of the folding process. Every change in the code results in a different amino acid on the chain, which can change the functioning of the protein, or change the way it folds - which simply creates a random mess.

If you don't explain how this information is created, you haven't explained much.

David

Yes, it is super difficult. I just might point out that recently researchers have developed and trained a special AI system to do this (come up with the 3-D structure from the sequence). Indirectly it is Man accomplishing the feat. But I don't think even the Third Way people would propose that logic built-in to single cells could even collectively solve the problem in some way analogous to the man-made AI, or in any other way.

Of course, the system would also need to have predicted the 3-D structure necessary to solve the biological problem, a calculation probably also out of reach of a cellular logic system.

Yes, it is super difficult. I just might point out that recently researchers have developed and trained a special AI system to do this (come up with the 3-D structure from the sequence). Indirectly it is Man accomplishing the feat. But I don't think even the Third Way people would propose that logic built-in to single cells could even collectively solve the problem in some way analogous to the man-made AI, or in any other way.

Do you have a link to that - does it get it right all the time, and what range of proteins does it work with?

David

Do you have a link to that - does it get it right all the time, and what range of proteins does it work with?

David

From https://www.cnet.com/science/biology/goo...o-science/ :

"....Google's sister company, DeepMind, announced it has successfully used artificial intelligence to predict the 3D structures of nearly every catalogued protein known to science. That's over 200 million proteins found in plants, bacteria, animals, humans — almost anything you can imagine.
..........................
AlphaFold was trained by DeepMind engineers to predict protein structures without requiring laboratory presence. No crystals, no electron firing, no $100,000 experiments.

To get AlphaFold to where it is today, first, according to the company's website, the system was exposed to 100,000 known protein folding structures. Then, as time passed, it started to learn how to decode the rest.
..........................
...usually — actually in many cases — the error is really tiny. So we call that a near-atomic precision."

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

"In a recent special issue of Science, several groups described how AlphaFold helped them piece together the nuclear pore complex, one of the most fiendish puzzles in biology. The giant structure consists of hundreds of protein parts and controls everything that goes in and comes out of the cell nucleus. Its delicate structure was finally revealed by using existing experimental methods to reveal its outline and AlphaFold predictions to complete and interpret any areas that were unclear. This powerful combination is now becoming routine in labs, unlocking new science and showing how experimental and computational techniques can work together."

Comment:
 
The article doesn't say, but it looks like the human developers of AlphaFold don't really know exactly how the AI program actually solves the problem in any particular case. Like with many other contemporary large scale AIs like ChatGPT. In the most difficult cases the researchers using AlphaFold apparently have to use traditional experimental methods to work out the rough shape. Then once AlphaFold starts it is sort of an impenetrable "black box". This "black box" apparently boils down to a vast amount of number crunching in finding common patterns in large amounts of existing data as to what DNA sequences correlate with what 3D protein structures.

No deterministic causal analysis, but it apparently regularly achieves accuracy competitive with the very expensive experimental approach. 

It seems to me that AlphaFold's success doesn't make the apparently unsolvable protein evolution enigma any more solvable by natural undirected processes like RM + NS, than it was before. It just makes the level of intelligence that must underlie macroevolution even more unfathomable.

It seems to me that AlphaFold's success doesn't make the apparently unsolvable protein evolution enigma any more solvable by natural undirected processes like RM + NS, than it was before.

Wow - I was totally unaware of that development, but even so I agree it doesn't solve the protein evolution enigma.

It will be interesting to see if it ultimately helps with finding novel protein structures, because if you have to perform some vast computation to make one tentative step in sequence space, it may be hard to get very far.

David

Wow - I was totally unaware of that development, but even so I agree it doesn't solve the protein evolution enigma.

It will be interesting to see if it ultimately helps with finding novel protein structures, because if you have to perform some vast computation to make one tentative step in sequence space, it may be hard to get very far.

David

I think you have a misperception on this issue. The reason why no one can come up with an analytical formula to predict the 3d structure of a protein is due to the vast amount of variables included in this hypothetical ‘folding function’. For this type of problems machine learning is a great tool as it can approximate unknown multivariate functions by “learning” from provided input-output pairs. In other words, there’s nothing to “solve” but continously refining the models by providing incremental more data.

The reason why no one can come up with an analytical formula to predict the 3d structure of a protein is due to the vast amount of variables included in this hypothetical ‘folding function’.

Er, where exactly did I talk about "an analytical formula to predict the 3d structure of a protein"?

David

Er, where exactly did I talk about "an analytical formula to predict the 3d structure of a protein"?

David

It implicit follows from “solve the protein evolution enigma.”

There’s nothing ‘magic’ going on here, it’s just a function having many many variables.

It implicit follows from “solve the protein evolution enigma.”

I guess your understanding of an analytic formula isn't the same as mine. Yours seems to include anything that could be calculated on a computer.

There’s nothing ‘magic’ going on here, it’s just a function having many many variables.

To be honest, we don't know exactly what is going on here. AI has a long history of hype.

David

I guess your understanding of an analytic formula isn't the same as mine. Yours seems to include anything that could be calculated on a computer.

To be honest, we don't know exactly what is going on here. AI has a long history of hype.

David

I fully understand (and sympathize) with you seeking supernatural explanations in everyday phenomena, but I'm afraid that protein folding can be explained entirely through physical laws.

By the way, definition of analytical formula (or analytical expression) can be read here https://en.m.wikipedia.org/wiki/Closed-form_expression

I fully understand (and sympathize) with you seeking supernatural explanations in everyday phenomena, but I'm afraid that protein folding can be explained entirely through physical laws.

Somehow I think you are completely missing the point of this discussion.

1)      I'm sure that if you gave an AI all the data on near death experiences, it could discover some useful information, but that wouldn't mean that NDE's can be explained by physical laws!

2)      We are trying to consider the chance of a new protein emerging by random mutations of the DNA string with natural selection. Even if proteins didn't fold, there would be an impossible combinatorial barrier stopping this from happening. The problem is that as it would be created step by step, most of the partially formed proteins have no value in themselves, they only have value as part of the final solution. Natural selection can't take future values into account.

AIs don't spit out analytic expressions (anyway, that issue doesn't seem relevant here) an AI simply prints out what it 'thinks' is a most probable answer based on information - including information on the internet.

3)      Unless you start with a set of novel proteins for which 3-D structure have not been determined, and which is then subject to 3-D structure determination after the AI has printed out its results, there is every chance the AI obtains information off the internet - which basically means that it cheats. Remember that AI's were supposed to be super-competent drivers, until they were tested on the roads and a number of pedestrians and others had paid a very heavy price.

David