Simulation of life using chemistry?

[Findthepin1]

So I had this idea a while ago. I wanted to simulate a bacterium or something in a computer program, but without having programmed it to act like a bacterium, but it's made of simulated atoms and particles, with all the energy levels and whatnot the same, and see what it does? Like, it is a collection of many, many virtual atoms arranged like the ones in a real bacterium, and see if it acts alive or just sits there no matter what is tried to get it to wake up (whatever would work in real life). Is this feasible, and what kind of computer processing power would be needed?

Edited September 8, 2016 by Findthepin1

[DerekL1963]

31 minutes ago, Findthepin1 said:

Is this feasible?

No.

[cubinator]

You are thinking of simulating the individual interactions of around a hundred billion atoms, even with something so small as a bacterium. The only computer powerful enough to simulate something like that that I know of is whatever runs the universe. If it were feasible, someone would already have done it.

However...

I do seem to recall that some scientists programmed a computer to mimic the brain of a simple creature with very few neurons, hooked it up to a robot, and the robot moved as if it were an animal. Here it is in action. The robot is not programmed to avoid walls, it is the simulated neuron interactions receiving sensory signals and reacting by moving the wheels.

[KSK]

@Findthepin1It's not nearly as finely grained as the model you were proposing but take a look at the OpenWorm project.

Something I've wondered about (although I have no idea how to set about it) is building a model of a cell but at the molecular level rather than the atomic. Or simpler(?) still, a model of the proteome (because modeling gene expression would be... challenging) Define each protein simply as an object that reacts with one or more other objects, at a given rate, to produce other objects.

It'd be interesting to see if you could put populations of simulated proteins in a box and have real life biochemical pathways to emerge.

Edited September 9, 2016 by KSK

[p1t1o]

9 hours ago, Findthepin1 said:

So I had this idea a while ago. I wanted to simulate a bacterium or something in a computer program, but without having programmed it to act like a bacterium, but it's made of simulated atoms and particles, with all the energy levels and whatnot the same, and see what it does? Like, it is a collection of many, many virtual atoms arranged like the ones in a real bacterium, and see if it acts alive or just sits there no matter what is tried to get it to wake up (whatever would work in real life). Is this feasible, and what kind of computer processing power would be needed?

Im not sure that we actually know enough about the fine-grained inner-workings of a living cell to even program it properly, let alone have the processing power to run it in real-time. Even the folding of a single protein molecule has been a frustrating challenge.

[Green Baron]

Spoiler

I remember there were simple algorithms from the 80s as a play with programming basic and pascal :-)I don't know of such algorithms today. Your statistical approach as a basis shouldn't be that difficult.

How about:

- a code

- a "mutation rat" for the code. You can play with it like certain areas of the code change faster than others etc. Can be arbitrarily complex :-)

- an "environment", a filter that sorts out certain code snippets. Can be arbitrarily complex.

- "competition" (?), other codes with different "mutation rates"

Next step: "environmental change" and "metabollism": resources and products change the "environment". Limits apply, competition gets a sense.

 

Processing power is very low, even for millions of "codes" if not programmed too badly ... the environment is calculated once each generation, the mutations are just bit-changes, so is the comparison. A laugh for a gigahertz-processor ...

 

Edit: nevermind, i misunderstood, you want to simulate an individual cell, yeah, that's more complex than a population ... :-)

 

Well, i'm not a biologist nor a chemist, what's your plan of modelling the steps from a herd of atoms to a cell ? Is it something like atoms - molecules - complex molecules - basic chemical processes - metabolism - organelles - cell and all the interfaces / interdependencies /reactions ? That would be very fine grained, heat, pressure, permeabiltiy, flow of things, ... i am not sure whether all these processes are understood irl. On the other hand, if you don't do it that fine-grained your atoms might well form something different than a cell ...

Am not sure whether it's possible, but if you have the knowledge of how a cell works at the level of atoms then let's ry to model it :-)

 

Edited September 9, 2016 by Green Baron

[todofwar]

I would say not currently possible. We can't simulate every atom in a single protein, much less the entire cell. Computational biology is a novel field, and I'm not terribly familiar with what they do but I assume they input the proteins and their known interactions, but still missing from that model is the very complicated set of small molecules which we know little about. So, people are trying for this but simulating a set of atoms just isn't feasible. Also, what do you mean by simulating? Depending on the level of theory, 50 atoms could take longer than the lifetime of the universe to calculate with today's supercomputers. Other levels are faster but miss out on some intricacies, and I don't think it's known how those little things might add up in a system like a cell.

[KSK]

2 hours ago, p1t1o said:

Im not sure that we actually know enough about the fine-grained inner-workings of a living cell to even program it properly, let alone have the processing power to run it in real-time. Even the folding of a single protein molecule has been a frustrating challenge.

Agreed, which is why I was proposing to treat a protein essentially as a black box and not worry about its structure or how it actually works. All I'd need for the model is that protein W transforms substrate X into product Y with a rate constant of Z. Taking that one step further, I could include protein V, where protein V 'interacts' with protein W to modulate rate constant Z.

That sort of level of detail.

Even then, I suspect you're right and that we simply don't know enough (particularly about the kinetics of specific protein-protein interactions) to build a useful model of the entire proteome.

[p1t1o]

11 minutes ago, KSK said:

Agreed, which is why I was proposing to treat a protein essentially as a black box and not worry about its structure or how it actually works. All I'd need for the model is that protein W transforms substrate X into product Y with a rate constant of Z. Taking that one step further, I could include protein V, where protein V 'interacts' with protein W to modulate rate constant Z.

That sort of level of detail.

Even then, I suspect you're right and that we simply don't know enough (particularly about the kinetics of specific protein-protein interactions) to build a useful model of the entire proteome.

Honestly, I think that we do pretty well today, with statistical treatments of atomic-scale systems, no need to simulate them down to the atomic scale. Ineed, the anomalies that you would see constantly at such a small scale are ironed out at the statistical level, which does enhance certain qualities of the data.

I have no doubt though, that atomic-scale simulations of systems like cells are on the horizon, it is only a question of complexity. But I think interpretting the results would be the bigger challenge, the data output from such a simulation would be mind bogglingly large and as complex as life itself.

[todofwar]

27 minutes ago, KSK said:

Agreed, which is why I was proposing to treat a protein essentially as a black box and not worry about its structure or how it actually works. All I'd need for the model is that protein W transforms substrate X into product Y with a rate constant of Z. Taking that one step further, I could include protein V, where protein V 'interacts' with protein W to modulate rate constant Z.

That sort of level of detail.

Even then, I suspect you're right and that we simply don't know enough (particularly about the kinetics of specific protein-protein interactions) to build a useful model of the entire proteome.

That might be more feasible, but you need to take into account protein expression. Different proteins are produced at different amounts based on conditions of the cell. Not a useless exercise by any means, but very difficult. And you also have to add in small molecule signals, which are really a black box at this point. 

[lajoswinkler]

Even if it could be simulated using today's technology, it would not work. Stuffing a bunch of molecules and ions into a simulation area will just yield increasing entropy and the thing will decay.

Life is a specific behaviour of matter which had emerged way back in the geological past, but it had to be a gradual progress ending with self replicating molecules which were the basis of a protocell. That "spark" ignited a long time ago has been carried as a torch by that replication and then evolved into extremely complicated metabolism.

 

Stuff inside cells isn't enough. It also requires the stuff to be suitably energized (not all parts of each stuff can have the same energy, that means charge, conformations, ...), to be positioned properly and surrounded by proper stuff. There's a TON of data right there, much more than the number of all those particles. We're talking about numbers possibly breaching googol.

 

It's unreachable to us.

[kerbiloid]

As all individuals are unique and their bodies have absolutely different molecular patterns, the definition of "bacteria" on atomic level looks very indefinite.

Edited September 9, 2016 by kerbiloid

[KSK]

1 hour ago, todofwar said:

That might be more feasible, but you need to take into account protein expression. Different proteins are produced at different amounts based on conditions of the cell. Not a useless exercise by any means, but very difficult. And you also have to add in small molecule signals, which are really a black box at this point. 

Yeah, and gene expression is just wheels within wheels within wheels. Plus if we're trying to model a eukaryotic cell (which would be medically more significant) then we get into alternative splicing, intracellular trafficking, all sorts of good stuff.

A quick question then. How feasible do you think it would be to model a signaling pathway?  Start with a membrane receptor, assume (for the sake of the simulation) that the receptor has been activated by its cognate signaling molecule and end with 'something happens at the nucleus'. Anything after that is ignored, we're just trying to capture the signaling.

[KSK]

1 hour ago, lajoswinkler said:

Even if it could be simulated using today's technology, it would not work. Stuffing a bunch of molecules and ions into a simulation area will just yield increasing entropy and the thing will decay.

Life is a specific behaviour of matter which had emerged way back in the geological past, but it had to be a gradual progress ending with self replicating molecules which were the basis of a protocell. That "spark" ignited a long time ago has been carried as a torch by that replication and then evolved into extremely complicated metabolism.

Stuff inside cells isn't enough. It also requires the stuff to be suitably energized (not all parts of each stuff can have the same energy, that means charge, conformations, ...), to be positioned properly and surrounded by proper stuff. There's a TON of data right there, much more than the number of all those particles. We're talking about numbers possibly breaching googol.

It's unreachable to us.

Not sure I agree with that. Life is just(?) very complicated chemistry but other than that it's not special. At the level of individual enzymes catalyzing specific reactions, it can be understood and modeled in the same way that any other chemical reaction can.

How all the myriad components of a cell interact to produce a blob of living material is something that I suspect we'll only start getting to grips with using computer simulations. The question is, how far can we abstract those simulations and still have them tell us something useful.

I agree that simulating a cell to a level where you could zoom right in and watch say, the vibrational modes of alanine 24 on protein X changing in real time depending on the proximity of nearby water molecules, ions, charged residues on other amino acids etc.,  is likely to be impossible for the foreseeable future. But is all that fine detail necessary to get a useful simulation?

My gut instinct tells me, no it's not, in the same way that it's possible to simulate a neural network using extremely simplified 'neurons' that abstract away all the biological messiness of real neurons. However, my gut is frequently wrong and, as a rule, not to be relied upon.

Edited September 9, 2016 by KSK

[todofwar]

57 minutes ago, KSK said:

Yeah, and gene expression is just wheels within wheels within wheels. Plus if we're trying to model a eukaryotic cell (which would be medically more significant) then we get into alternative splicing, intracellular trafficking, all sorts of good stuff.

A quick question then. How feasible do you think it would be to model a signaling pathway?  Start with a membrane receptor, assume (for the sake of the simulation) that the receptor has been activated by its cognate signaling molecule and end with 'something happens at the nucleus'. Anything after that is ignored, we're just trying to capture the signaling.

That might be more feasible. Even a single signaling pathway would be more complicated than is normally assumed, you normally get signal A triggering effect B that amplifies effect C that leads to production of D that then actually signals the nucleus. 

I actually thought it would be interesting to try and build a physical analogue cell, because allot of what goes on has some analogous process in circuits.

[lajoswinkler]

38 minutes ago, KSK said:

Not sure I agree with that. Life is just(?) very complicated chemistry but other than that it's not special. At the level of individual enzymes catalyzing specific reactions, it can be understood and modeled in the same way that any other chemical reaction can.

How all the myriad components of a cell interact to produce a blob of living material is something that I suspect we'll only start getting to grips with using computer simulations. The question is, how far can we abstract those simulations and still have them tell us something useful.

I agree that simulating a cell to a level where you could zoom right in and watch say, the vibrational modes of alanine 24 on protein X changing in real time depending on the proximity of nearby water molecules, ions, charged residues on other amino acids etc.,  is likely to be impossible for the foreseeable future. But is all that fine detail necessary to get a useful simulation?

My gut instinct tells me, no it's not, in the same way that it's possible to simulate a neural network using extremely simplified 'neurons' that abstract away all the biological messiness of real neurons. However, my gut is frequently wrong and, as a rule, not to be relied upon.

I didn't say it was special in a supernatural sense. It's ordinary as any other natural process but it's easily distinguishable from burning sulfur or solidifying ammonia.

 

Is the fine detail needed for a useful simulation? I'd say it is. Life does have a margin where you can mess with it and it still resists (one of the things that makes it - life) changes, but it's not like a simple buffer solution in a beaker. It's a series of nested processes with a huge number of feedback loops, and some of those processes are emergent, therefore not directly caused by pure mechanics.

We can see how life easily collapses in the case of living organism being exposed to a neutron flux of sufficient energy. Lift the temperature of the whole organism by a fraction of a degree and it falls apart like a house of cards. Total collapse of metabolic pathways.

Applying chemicals, poking it with electrodes, that gives it an opportunity to spatially adjust. Now apply that to trying to simulate a whole system from the scratch at once... It's abominably difficult, quite possibly unreachable, given the number of bits needed for it.

 

What we might do (if we ever get the computing power, which is highly unlikely) is to simulate less complex systems and watch them get more complicated on their own. That might work. Might.

Trillions of atoms in a cell and the number of their states, even with a margin I mentioned applied, it still makes an ungodly thing to consider for a simulation.

[KSK]

1 hour ago, todofwar said:

That might be more feasible. Even a single signaling pathway would be more complicated than is normally assumed, you normally get signal A triggering effect B that amplifies effect C that leads to production of D that then actually signals the nucleus. 

I actually thought it would be interesting to try and build a physical analogue cell, because allot of what goes on has some analogous process in circuits.

Yup, and as I understand it, all sorts of complicating factors such as redundancy, cross-talk between pathways and different pathways converging on a single point. As @lajoswinklersaid, nested processes and feedback loops. That's where I think these kind of simulations (assuming that they're at all workable) would be really interesting. Rather than dissecting out a single signaling 'pathway' at a time, treat the whole thing as a single interconnected network - and then start exploring what happens when you start modulating the activity of different enzymes in that network.

[KSK]

1 hour ago, lajoswinkler said:

I didn't say it was special in a supernatural sense. It's ordinary as any other natural process but it's easily distinguishable from burning sulfur or solidifying ammonia.

Is the fine detail needed for a useful simulation? I'd say it is. Life does have a margin where you can mess with it and it still resists (one of the things that makes it - life) changes, but it's not like a simple buffer solution in a beaker. It's a series of nested processes with a huge number of feedback loops, and some of those processes are emergent, therefore not directly caused by pure mechanics.

*snip*

What we might do (if we ever get the computing power, which is highly unlikely) is to simulate less complex systems and watch them get more complicated on their own. That might work. Might.

Trillions of atoms in a cell and the number of their states, even with a margin I mentioned applied, it still makes an ungodly thing to consider for a simulation.

Yup - the emergent processes were what I had in mind when I suggested that the only way to really get to grips with a cell is to simulate the whole thing.

And I completely agree that an atom-level simulation is an ungodly thing to consider, which is why I'm wondering if you need that level of detail. Could we just model an enzyme (for example) essentially as a chemical rate equation (which is trivial) with empirical corrections for additional factors, for example, the rate constant might depend on the concentration of other chemicals or enzymes in the simulation. No need to model every atom in every enzyme - just treat 'em as a collection of black boxes.

I'm optimistic that that kind of approach might work - I suspect you'll disagree but I'm enjoying the debate!  

 

[PB666]

16 hours ago, Findthepin1 said:

So I had this idea a while ago. I wanted to simulate a bacterium or something in a computer program, but without having programmed it to act like a bacterium, but it's made of simulated atoms and particles, with all the energy levels and whatnot the same, and see what it does? Like, it is a collection of many, many virtual atoms arranged like the ones in a real bacterium, and see if it acts alive or just sits there no matter what is tried to get it to wake up (whatever would work in real life). Is this feasible, and what kind of computer processing power would be needed?

How good are you at statistics?

Im going to say something here that is very theoretical but has very practical implications.

At the subatomic level, meaning how we understand electrons and protons, ions, etc, is determined much more by quantum mechanics that we biochemist like to account for. According to the copenhagen interpretation lack of space-time continuity at the quantum level resolves itself at the level of observation, which is suitable for a chemical interpretation. This sounds very theoretical, but on the scope of the cell, process can be practically inferred from the laws of mass action, this basically mean that quantum uncertainty ultimately results in a kind of distribution of outcomes over measurable time, at statistical outcome.

So lets say we took a bacteria and found it to be composed of 100,000 processes, and we created a stoichemetric equation, just as you did with chemistry. We would not get the same result, this is because in chemistry you frequently don't list the Nth possibilities which are random, quantum events would create outcomes from some chemistry that would be unexpected unless you ran each or the equation 100 times. 

So the next question is how much of a bacteria (in-terms of space-time) the scale of which is down in the 10E-40 range. You'de have to have enough to represent the continuoum of hundreds of atomic and molecular processes somewhere between 10 and 1000 times. You would not have to represent quantum events itself, but the statistical outcomes. Consider water all by itself.

Water is not water, or what we think is water is almost never in a pure state. You could not take water say at a given moment rotate the molecules in space and get exact replications in nearly quantum space-time. Aside from the bond wobbling that goes, stretching and shrinking, the electrons are moving further away and closer to water, sometimes electrons from adjacent molecules swap in a process where hydrogen is handed off or returned (and is the reason water has a pH of 7). Now we throw in ions such as sodium, potassium, etc, and this becomes a complex mess. You haven't even started with biochemistry. Biopolymers use water to force hydrophobic parts of its structure to interact, while at the same time exchange protons with water in very complicated ways (Histidine being an example). Ions in water flow into and out of pockets on biopolymers, they chelate, So this is the basic chemistry of life, that is pretty much to say that energy flow within biological systems is pretty much conducted by polymorphic aqueous states at some point in their pathways. There are exceptions such as fat mobilization and lipoprotein complexes. But this one super-system in and of itself given quantum mechanics is a nightmare to model.

 

[Green Baron]

I think the tenor is clear: the atomic level is too complicated and uncertain to model in a program and the step towards a chemical process has it's own uncertainties and complifications.

The higher the abstraction, the more possibilities to model. Population genetics have cute algorithms, but that's not what was asked :-)

 

[lajoswinkler]

2 hours ago, KSK said:

Yup - the emergent processes were what I had in mind when I suggested that the only way to really get to grips with a cell is to simulate the whole thing.

And I completely agree that an atom-level simulation is an ungodly thing to consider, which is why I'm wondering if you need that level of detail. Could we just model an enzyme (for example) essentially as a chemical rate equation (which is trivial) with empirical corrections for additional factors, for example, the rate constant might depend on the concentration of other chemicals or enzymes in the simulation. No need to model every atom in every enzyme - just treat 'em as a collection of black boxes.

I'm optimistic that that kind of approach might work - I suspect you'll disagree but I'm enjoying the debate!  

 

It might work, but then it might not. I suspect the resistance (on several higher levels) to change is embedded in the enormous number of molecules and atoms bumping around. But then again, I would like to see the black box approach for enzymes, even though that would also be incredibly difficult to model. What would the degrees of spatial freedom be, for all those black boxes? See, this opens a whole new set of problems.

Edited September 9, 2016 by lajoswinkler

[todofwar]

47 minutes ago, lajoswinkler said:

It might work, but then it might not. I suspect the resistance (on several higher levels) to change is embedded in the enormous number of molecules and atoms bumping around. But then again, I would like to see the black box approach for enzymes, even though that would also be incredibly difficult to model. What would the degrees of spatial freedom be, for all those black boxes? See, this opens a whole new set of problems.

One step simpler would be to not model the actual three dimensional space. Just a set of conditions run through your sim, with proteins assumed to be able to interact with their appropriate partners. The diffusion terms could be included in th overall rates. 

 

[KSK]

20 minutes ago, todofwar said:

One step simpler would be to not model the actual three dimensional space. Just a set of conditions run through your sim, with proteins assumed to be able to interact with their appropriate partners. The diffusion terms could be included in th overall rates. 

 

Sorry @lajoswinkler - I think 'black box' was a bad choice of words on my part. @todofwar's answer is closer to what I had in mind - and in fact better than what I had in mind because I hadn't got as far as including diffusion terms.

Edited September 9, 2016 by KSK

[lajoswinkler]

14 hours ago, todofwar said:

One step simpler would be to not model the actual three dimensional space. Just a set of conditions run through your sim, with proteins assumed to be able to interact with their appropriate partners. The diffusion terms could be included in th overall rates. 

 

That is the problem - internal and external spatial distribution of enzymes and their coenzymes is exactly their key feature. Their shape, conformation and position in space is what makes them enzymes. If you take that away, you can model them as black boxes with a certain value of probability of reacting with a substrate but then it would be a poor glimpse of a simulation. It would not be a simulation of biological life, but a simulation of something I'd call a different name. It would be a form of life, but not biological one.

 

Biological life has spatial structures as building blocks. Example:

cytochrome oxidase systems, self assembly after 20 μs. Each of those wormy things is a -CH₂- chain.

 

Or this.

 

Not to mention replication, transcription, protein folding...

 

And there's also the extracellular environment which is 100% needed for a cell to function and is therefore inseparable. Insanity.

[todofwar]

@lajoswinkler I'm well aware of the importance that structure plays, and that even bacterial cells seem to have more internal organization than wax previously thought. But it's impossible to model that many atoms. It takes me 24 hours with a pretty damn powerful computer to optimize a simple complex with 100 atoms. And that's low level theory. For this kind of project you need to simplify to some extent. So you can go with what I call the circuit model, which only looks at proteins in and out products and their effects. That gets you a good bit of the way there, but not all the way of course. Maybe you can move on and model each enzyme as a particle, with their diffusion modeled explicitly, which is harder but might give more insights. You can then have enzymes as shapes corresponding to their physical envelope and try to have the rules for association thrown in, and get more out, but I'm willing to bet that's the limit. And missing from all of this of course is the translation machinery, and just as important the small molecule metabolome. But it all depends on what you want to get out of it. Maybe looking at how a protein behaves in the context of the cell explains some quirks of its known in vitro behavior, but to get at that which level is really necessary? Similar to arguments about quantum mechanics. Which level of theory is necessary for getting the answers you want? And they're all approximations at the end of the day.

 

Not related to above but the replies will merge anyway:

I woder if someone could make a video game out of this. Kind of like the KSP of computational bio. You start with a cell and a gene to express some basics, and you have to evolve parts to make your cell more efficient. Each time you split off a new cell you get points. I'm thinking it would work like minecraft, building systems to bring resources to the right place.