Full-Atom Molecular Dynamics embryo -> adult?

[JohnT89]

What would be a rough estimate of the computational complexity to simulate an embryo in a full-atom scale from the instant of conception to the death of the aging adult?

Its a purely theoretical question and I know that its something like daydreaming for the current technology, but I would like to have an idea in terms of order of magnitude as I am incredibly fascinated by this thing.

How many atoms are we roughly talking about? One thing I am sure is that it will be definitely possible someday, and I am very curious about what those colossal simulations might reveal

 

[Ryan_m_b]

Using this source we get a rough estimate of 7e27 atoms in an average human.

As for simulating that I have no real idea of how to work it out but I know that when simulating simple systems (i.e. a few atoms at a time) the computational power needed is huge because of all the variables. Adding X atoms to a system does not mean you need X more computational power, it means you need >X because you have to account for all the extra variables.

To simulate a cell from conception to adulthood would require a gargantuan amount of computational power, it's not something that we could even wait for Moore's law to give bearing in mind we've only got about a decade left before it levels off. In addition we would first have to know exactly how certain processes occur, currently we cannot even simulate how a protein folds, let alone how a body works in real time. We would also have to be able to simulate the environment of our simulated person to.

I doubt this will ever be practical, the ever increasing number of variables makes the issue almost impossible. You've also got to then cover the ethics of creating a life that is indistinguishable to the real thing just to experiment on it, just because it's a simulation doesn't mean it won't be conscious and have as much right to life as you or me.

 

[mishrashubham]

At the moment the fastest supercomputer on the planet is Tianhe 1A of China with a processing power of 2.6 petaFLOPS (FLoating Point Operations per Second - a measurement of processing power) or 2.6*10¹⁵ FLOPS. For comparision, an average calculator has a power of 10 FLOPS.

According to Wikipedia

Erik P. DeBenedictis of Sandia National Laboratories theorizes that a zettaflops (1021) (one sextillion FLOPS) computer is required to accomplish full weather modeling, which could cover a two week time span accurately. Such systems might be built around 2030.

Even a single cell is much much more complex than weather patterns, at the molecular and atomic scale. As noted above, the amount of processing power required for complete weather simulation differs from our current maximum capacity by many orders of magnitude and even that is only accurate to two weeks. Imagine what it would take to simulate an entire body for a span of many decades.

In addition we would first have to know exactly how certain processes occur, currently we cannot even simulate how a protein folds, let alone how a body works in real time.

Exactly. At the moment,we can find out the primary structure of a protein with relative ease from its gene sequence, but predicting its tertiary structure is extremely difficult. Knowing how an amino acid sequence eventually folds into a complex and functional 3-dimensional structure in extremely small time intervals (microseconds and milliseconds) is a major challenge. And the current state of computer simulation technique isn't helping either.

You've also got to then cover the ethics of creating a life that is indistinguishable to the real thing just to experiment on it, just because it's a simulation doesn't mean it won't be conscious and have as much right to life as you or me.

I wonder how well the Blue Brain Project turns out to be. I don't even know if I can trust those guys.

 

[Ryan_m_b]

It's an interesting coincidence but a few weeks ago I was talking to a colleague (a physicist who has a keen interest in computers) about the wiki article that mentions the zettaflops thing and he was puzzled why anyone would state such a precise number. He related to me a story about how he visited a weather simulation centre at one point and a key thing that came across was that the increasing variables meant that the increase in computer power required becomes semi-exponential.

To make matters worse thanks to chaos theory invariably something will happen that will make the whole simulation incorrect. I get the impression that even with better computing power you need better real-time sensory information to try and handle what's going on. Like you said though, a human body (not forgetting the environment to sustain them) is orders of magnitude more complex than that!

I'll start worrying about the BBP when they start simulating entire CNS's (or trying). Until then I'm content to observe how it works, though honestly I think the money could be better spent on designing better imaging technology to study the brain directly rather than trying to simulate small parts of it.

 

[mishrashubham]

He related to me a story about how he visited a weather simulation centre at one point and a key thing that came across was that the increasing variables meant that the increase in computer power required becomes semi-exponential.

I understand. Rendering a simple object collision-destruction simulation on Rayfire in 3ds Max took more and more time for every object or force I added.

 

[JohnT89]

Thanks for all answers, you pointed me in the right direction, but I still have doubts.

(Note that I study IT Engineering, not biology)

There is a public project, Folding@home, which uses Molecular Dynamics to simulate small proteins for short amounts of time and already does that since about 10 years. From what I know from their forums and documentation, they do a lot of simulations about the same protein in different initial conditions and then perform some kind of statistical analysis.

To get an idea of how large those simulations are and how long do they take, this is a list of their sub-projects (http://fah-web.stanford.edu/psummary.html).

Very roughly speaking, it takes about 6 hours on a modern quad core cpu with 8 threads to process around 77'000 atoms with explicit solvation (accurate model of the water) and do one single simulation. For comparison, one single simulation of ~1'000'000 atoms takes around 36-48 hours on a very similar hardware to be completed.
The problem is that they need a lot of those simulations in order to ensure that
- there is some 'recurrence' of results,
- there wasn't any lucky random error went past undetected
- the initial conditions are varied to check how and how much the simulation result changes.

This tells me that its not about simply simulating all the 7e27 atoms, but doing that many times so that some kind of statistical analysis would be possible.

I am not questioning the actual feasibility of this thing (I know, we all know that the needed computational power is crazy compared to the current technology),

but, from a theoretical standpoint, could simulating everything starting from a single embryo molecular model be actually 'useful'?

I mean, aren't the 'instructions' of human body all coded into DNA? Wouldn't simply simulating it decode most of the misteries of human body?

 

[Ygggdrasil]

One thing to remember about such simulations is that cells are not a closed system. Cells require a constant influx of materials and energy for growth. Furthermore, cells have complex interactions with their environment that can influence how the cell reads its DNA.

One particularly important example here is a phenomenon known as the maternal effect. There are several gene products, produced not by the embryo but by the mother's cells surrounding the embryo, that have important roles during the development of the embryo. This phenomenon is an example of how all of the "instructions" for the development of the embryo are not contained solely in its own DNA. Here, the mother's DNA, not the embryo's DNA, controls the traits of the embryo (i.e. whether the embryo displays a certain developmental defect depends only on whether the mother has the associated mutation. It does not matter whether the embryo itself has that mutation in its DNA).

Another thought about such simulations is the importance of initial conditions. You can't just plop DNA into a cell-like environment and expect it to perform normally. That DNA must be in the context of the proper regulatory factors and cellular machinery to be able to perform its proper function. Even if we had the technology to perform such simulations, we would still need to do a lot of fundamental research into the biology of the embryo before we could know the right parameters for the creation of such a simulation. Current research into cellular reprogramming (i.e. research with induced pleuripotent stem cells) and research at the J Craig Venter Institute on creating a synthetic microbe with a minimal genome could possibly give us some foothold into beginning to understand what's going on here.