Help
Cause, on 19 December 2020 - 07:23 PM, said:
Mezla PigDog, on 19 December 2020 - 11:39 AM, said:
Malankazooie, on 18 December 2020 - 06:09 PM, said:
DeepMind AI protein folding. What is it? Why is it important? What will its impact be? What will it mean for medical research? Will we see a surge in innovative therapies? Will it spin off and become impactful in other industries? What does the future hold? Reading the excitement about it and the possibilities, is the hype warranted?
If it is as accurate as they think it might be then it is massive. It will speed up research and drug design massively. Protein function is dictated by shape. Everything in your body is done by proteins. You can quite easily figure out the linear sequence of a protein from the DNA sequence but you can't guess how it will fold because it depends on loads of funky stuff in the cell. You have to make masses of protein to be able to see it enough to see how it folds but the process of making it puts it into an artificial environment so it folds differently to how it would in the cell. And it takes ages. So it's a pain in the ass, slow and inaccurate. If you can do it by AI then it is monumental.
My understanding is that it still requires significant experimental data to help build its model. The holy grail of biochemistry though is solving the protein folding problem. If we could predict how a sequence of protein would fold into its 3D shape we would basically be on the cusp of being able to design enzymes that do what we want. Understand any virus and how it works just by reading it's genome. Solve the enzymatic mechanisms of enzymes that are perhaps targets for drug therapies. Sadly the variables are astronomical and we don't even know what they are all
Protein folding is governed by (just to name a few)
Protein sequence (the sequence is both the code, the blueprint and the building blocks. Imagine a computer in which the hard drive is also the pc case, the monitor, the battery all at once)
Levinthals paradox (protein folding is highly controlled, if proteins sequences just sampled every possible 3D conformation that was possible it would take a million years, clearly their are mechanisms which limit its options and lock it down correct paths)
Hydrophobic effect (is the protein folded in water, lipids, combination?)
Translation speed and tRNA concentric a in a cell (if the mRNA code is translated to protein code too slow or too fast protein might fold incorrectly)
Molecular crowding (cells are insanely crowded compared to test tubes, having to much room to expand into might cause misfoldeing)
Chaperone proteins (some proteins need other proteins to act as scaffolds to guide correct folding)
Wuarternary structure (some proteins will only fold correctly if they are folded alongside and bond to another monomer unit of themselves or a different protein, in its absence it will misfold and not work)
Post translational modification ( a protein that folds correctly might still need another protein to come along and cut a part off, or add sugar molecules or change a functional group of an amino acid. Until that step is folded correctly to its instructions but remains worthless. Part of its code is kept somewhere else. These secondary, tertiary etc steps provide additional methods of control allowing protein concentrations and activities to be finely regilated)
This is also than an incomplete picture too. For a long time we thought proteins were all important. They are the machines that do all the work. However we are learning more and more how complex and how important regulation of genes are. After all a human and a chimpanzee are 99% identical. However by turning some genes on for longer, making more proteins that code for and others on for less you can get drastic different outcomes. Some dna we now know doesn't even code for protein, which was once thought to be junk dna that does nothing. We know now that it can code for segments of code that do nothing but bind to other segments of code to deactivate them for example.
So this is a big step forward but we are lifetimes away if it's even possible to even get close to what they did in the one Spider-Man movie where spider and the lizard are running computer models to determine if nw therapies would work on computer mice. In theory such a thing is possible. In practice....
With quantum computing it might be more like 10 years away.
'Most of physics, and all of chemistry are based on a single equation – the Schrödinger equation. The current state-of-the-art technology in computational chemistry is based on either truncating that equation or reformulating the problem to a simpler problem. In the first class of methods, arbitrarily high precision can be realized by increasing the allowed computation time. However, problems easily get out of hand, and only small systems can be simulated. With this method, the simulation of protein structure is often out of reach. The second class of methods is dominated by density functional theory (DFT). In DFT, the problem is simplified by considering the density of electrons, instead of the full electronic configuration. This works very well for a large number of systems but starts to break down when the interaction between electrons increases. Unfortunately, for many interesting properties such as protein folding, DFT doesn't provide enough accuracy. This is where quantum computing comes in.
Quantum computers are extremely good at calculating the Schrödinger equation. Instead of truncating the equation, or simplifying the problem, quantum computers can simulate systems with much higher accuracy. [...] The QPE algorithm provides an exponential speed-up over classical algorithms.
Quantum phase estimation is a game changer for simulations of systems with lots of interaction between electrons. By calculating the Schrödinger equation to arbitrary accuracy, we can now determine the energy surface of molecules. From the energy surface, we can learn about chemical properties such as reaction rates and binding energies. A better understanding of protein structure and dynamics can hence speed-up the fabrication of vaccines. Unfortunately, quantum computers do not have the required power to run these algorithms, and it is likely that we have to wait at least ten more years to see their full potential.'
https://www.capgemin...-crisis-part-1/
'Quantum computing: how conditions created by the COVID-19 shutdown are delivering "the best data we have ever seen".
Remotely controlled experiments are the way forward.
[...] the data have been excellent because the campus has been a ghost town. The lab's temperature hasn't wavered and there's little vibrational noise in the unoccupied building. It's one of very few university quantum experiments making real progress right now.'
https://www.nature.c...586-020-01937-x
'D-Wave Systems has offered free cloud computing time on its quantum computer to COVID-19 researchers.'
[...] The free quantum computing consulting services D-Wave is arranging include quantum programming expertise in scientific computing as well as in planning, management, and operations for front-line workers.'
https://spectrum.iee...the-coronavirus
This post has been edited by Azath Vitr (D'ivers: 19 December 2020 - 08:12 PM
