What biological systems can be simulated with exaflops computers?

[jonjacson]

Summary:: Fugaku's supercomputer is 442 PetaFlops fast, what can we simulate with it?

So the most powerful supercomputer in these days is 0.4 exaflops fast. I assume we can simulate precisely the interaction between simple atoms, small molecules but... What is the limit?

Could we simulate a big protein?

Would it be possible to simulate the behavior of organelles? A mitochondrion?

Or even further, A whole human cell? An entire organ? A virus?

Probably someone working in molecular biology can stimate this roughly.

Obviously when I say simulate I mean ignoring what can be ignored (for example the fact that neutrons and protons are made of quarks) and taking into account what really matters (classical electrical interaction models and then quantum mechanics).

Any idea?

What I want to know is how far away we are of simulating the behavior of the full human body. We are made of 10 ** 27 atoms (again I mean roughly) so computing it with Fugaku still requires too much time. Is this correct?

Thanks

 

[Vanadium 50]

I think you need to be a lot clearer about exactly what you are "simulating". For example, in combustion, you are simulating heat and mass transfer, as well as the chemistry of combustion. A computer does a calculation: something is equal to some function of other things. You haven't described either side of the equation.

 

[jim mcnamara]

You are asking for opinions of a not well defined subject that is in its bare beginnings.

Foremost, what you are asking cannot happen soon:

As to simulating the whole human body, we need to gather information of all of the alleles (alternate genes like white and yellow corn). Then work out affects of :
various alleles,
introns (~8% of DNA is ancient viral DNA),
"jumping genes"
etc.
on all of the biochemistry associated with this mountain of genetic data. Think of solving a 22000 variable equation with variables each of which has the potential to affect other alleles in another biochemical cascade. And change randomly. Or get turned off.

Then, comes the secondary effects of retrovirus infections which add genes to existing DNA, and all of the other kinds of epigenetic effects. Like Skinner's Heredity-Environment interaction. We need to know fully what the envrionment does to gene expression. For each gene and all of its alleles.

The above is an absurdly gigantic moving target. We need to get the ammunation first. For each human. There are 7+ billion of them. So after we get a reasonably complete picture we can start an approximate simulation of human life.

Check back in 20 years.

This topic as it stands today, is more speculative than scientific. Moved to General Discussion.

 

[jonjacson]

I think you need to be a lot clearer about exactly what you are "simulating". For example, in combustion, you are simulating heat and mass transfer, as well as the chemistry of combustion. A computer does a calculation: something is equal to some function of other things. You haven't described either side of the equation.

Ok I mean simulating the dynamics, how the system evolves with time.

 

[Vanadium 50]

And where is the equals sign in that?
I think that the reason you are being so vague is you don't understand what you want.

 

[BWV]

On top of our DNA, also wouldn't you also need to model the gut biome?

 

[atyy]

Here is a report simulating the whole human body

https://pubmed.ncbi.nlm.nih.gov/27347791/
Finite element modeling for predicting the contact pressure between a foam mattress and the human body in a supine position

 

[phinds]

Check back in 20 years.

Feeling a bit optimistic today, are we?

 

[jim mcnamara]

Well, given that we can now do full DNA sequences on many species including primates, and use the results regularly in Bio research, 20 years seems like a reasonable checkpoint. Covid really put the pedal to the metal on many research areas. IMO.

Check out 'Nutrient Metabolism Structures, Functions, and Genes' by Martin Kohlmeier. It is about 820 pages of what gene sets do what steps in the known nutrient processing cascades.

There is also a companion book on nutrigenetics for medical practitioners - i.e., applying the pathology of aberrant alleles to everyday medicine. Celiac Disease is an example, so is the reason why Vibrio bacteria commonly found in raw oysters are harmless to thousands of people, and make a very few people very ill every year. Has to do with serum Fe levels. And one allele.

 

[noequilibrium]

For what it's worth, OP, I don't think you're being vague at all.

The simulation method that you're thinking about is called molecular dynamics. This is a heavily used technique in computational biochemistry that pretty much involves solving Newton's equations (with various modifications) for a system of particles and their associated potentials. If your system is large enough and you run your simulation for a long enough time, you can derive mechanical and thermodynamic properties from the data you generate.

I have simulated the interaction between a Coronavirus spike protein and ACE2. I did it using my laptop, software called GROMACS and some freely available structure files. I don't think anybody has simulated a complete virus yet, but I wouldn't say that that this is unfeasible with the technology we have.

The question is: how big can you go? Quite big, actually... but perhaps not big enough for most of what you're asking for. I recall from last year that the record for the number of atoms simulated was around a hundred million. The problem with molecular dynamics isn't so much how many atoms you can simulate, but how long you can simulate for: you're thinking about problems in nanoseconds here. I've heard of a group simulating a protein for just over a millisecond, which is very long in molecular dynamics terms.

So, you mentioned you want to simulate at the level of quantum mechanics. This is possible: there are a number of attempts at doing molecular dynamics at an ab-initio level. One that comes to mind is Car-Parinello molecular dynamics, but I think that this method isn't as theoretically watertight as most people would like.

There is an enormous amount of money and brainpower being pumped into this kind of computational research. There's a literal billionaire investor-turned-scientist working on this exact thing (D. E. Shaw). I'm not sure what the next twenty years will hold, but I'm hopeful it'll be cool nevertheless.

 

[atyy]

Obviously when I say simulate I mean ignoring what can be ignored (for example the fact that neutrons and protons are made of quarks) and taking into account what really matters (classical electrical interaction models and then quantum mechanics).

This is the part that is vague. When allowing things to be ignored, one has to say what one is using the calculation for, or what properties one is interested in. The answer of what may be ignored changes depending on what phenomena one is interested in studying.

 

[f95toli]

Making a computer x10 faster doesn't help much at all for many problems.
It really depends on how the problems scale; for problems that scale exponentially (which includes many interesting problems in computational chemistry) a linear scaling of your computer resources will barely help at all; adding just one more atom might increase the resources needed by many orders of magnitude.

This is the why people are interested in quantum computers; since they work in very different way we hope that we will be able to use them to simulate problems that would be completely intractable on a classical computer irrespective of how big it is,
Many of the most important problems are indeed in biomedicine.

 

[phinds]

Well, given that we can now do full DNA sequences on many species including primates, and use the results regularly in Bio research, 20 years seems like a reasonable checkpoint.

Well, perhaps I'm more pessimistic than is warranted but you yourself pointed out the complexity of the issue.

I do remember being surprised fairly recently that a computer beat a GO master, but I think that, relatively speaking, that's a much less complex problem and is amenable to a more procedural kind of solution.

 

[BWV]

With GO, Chess and now protein folding, self-training ANNs vastly outperform brute force, any hope of this in more complex molecular simulations?

 

[jonjacson]

Here is a report simulating the whole human body

https://pubmed.ncbi.nlm.nih.gov/27347791/
Finite element modeling for predicting the contact pressure between a foam mattress and the human body in a supine position

Well, given that we can now do full DNA sequences on many species including primates, and use the results regularly in Bio research, 20 years seems like a reasonable checkpoint. Covid really put the pedal to the metal on many research areas. IMO.

Check out 'Nutrient Metabolism Structures, Functions, and Genes' by Martin Kohlmeier. It is about 820 pages of what gene sets do what steps in the known nutrient processing cascades.

There is also a companion book on nutrigenetics for medical practitioners - i.e., applying the pathology of aberrant alleles to everyday medicine. Celiac Disease is an example, so is the reason why Vibrio bacteria commonly found in raw oysters are harmless to thousands of people, and make a very few people very ill every year. Has to do with serum Fe levels. And one allele.

THanks, the book looks interesting.

For what it's worth, OP, I don't think you're being vague at all.

The simulation method that you're thinking about is called molecular dynamics. This is a heavily used technique in computational biochemistry that pretty much involves solving Newton's equations (with various modifications) for a system of particles and their associated potentials. If your system is large enough and you run your simulation for a long enough time, you can derive mechanical and thermodynamic properties from the data you generate.

I have simulated the interaction between a Coronavirus spike protein and ACE2. I did it using my laptop, software called GROMACS and some freely available structure files. I don't think anybody has simulated a complete virus yet, but I wouldn't say that that this is unfeasible with the technology we have.

The question is: how big can you go? Quite big, actually... but perhaps not big enough for most of what you're asking for. I recall from last year that the record for the number of atoms simulated was around a hundred million. The problem with molecular dynamics isn't so much how many atoms you can simulate, but how long you can simulate for: you're thinking about problems in nanoseconds here. I've heard of a group simulating a protein for just over a millisecond, which is very long in molecular dynamics terms.

So, you mentioned you want to simulate at the level of quantum mechanics. This is possible: there are a number of attempts at doing molecular dynamics at an ab-initio level. One that comes to mind is Car-Parinello molecular dynamics, but I think that this method isn't as theoretically watertight as most people would like.

There is an enormous amount of money and brainpower being pumped into this kind of computational research. There's a literal billionaire investor-turned-scientist working on this exact thing (D. E. Shaw). I'm not sure what the next twenty years will hold, but I'm hopeful it'll be cool nevertheless.

Yes I was thinking in those kind of methods, thanks for those details, that is what I wanted to learn about.

Making a computer x10 faster doesn't help much at all for many problems.
It really depends on how the problems scale; for problems that scale exponentially (which includes many interesting problems in computational chemistry) a linear scaling of your computer resources will barely help at all; adding just one more atom might increase the resources needed by many orders of magnitude.

This is the why people are interested in quantum computers; since they work in very different way we hope that we will be able to use them to simulate problems that would be completely intractable on a classical computer irrespective of how big it is,
Many of the most important problems are indeed in biomedicine.

I didn't know that problem of scaling, thank you so much!