5 ms·

what's interesting is that the GoL inception is (as far as I understand) highly sensitive to initial conditions and timing in order for everything to sync up pr

what's interesting is that the GoL inception is (as far as I understand) highly sensitive to initial conditions and timing in order for everything to sync up properly, whereas the cell is really robust in terms of everything just jumping around randomly and yet still getting its tasks done

Pulcinella 6y ago
Yeah it feels as if the cell is a ridiculously parallel array of molecules where every molecule in the array is rapidly iterating through the array (colliding with every molecule in the cell as it diffuses) and type checking those items to see if it can operate on them (e.g. a protein breaking down a small molecule). It all works out because the operations are atomic (heh) (don’t need to worry about two proteins trying to break down the same molecule at the same time) and the type safety (proteins are highly specific on what kind of chemicals they target) means you don’t need to worry about trying to perform some invalid operation. Turns out collision detection is real fast in real life!
- xwdv 6y ago
  What would happen if a cell were somehow to be optimized such that every operation occurs in O(1) time rather than having to iterate?
  - shuntress 6y ago
    Cancer, probably.
  - nsxwolf 6y ago
    This is probably what happened to Tetsuo in Akira.
  - pas 6y ago
    Hm, well, it would require a "global" approach instead of what currently happens. Which is the opposite, local gradients, local diffusion, local parallel approaches. Many times things are being built, remodeled, moved, teared down, packed for moving, and unpacked for use. Everything is at the same time a tool for building bigger things and is just a piece of raw material if something else happens to disassemble it or use it for assembling something bigger. So opposed to that there would need to be some kind of ordering, signaling, queuing, synchronization, etc. Thanks to osmosis building block things are always pretty close when they are needed, and products are always getting spread out evenly too. (And there is a transport network in each cell that helps with those that need some help spreading.) Of course there are certain things that work based on a kind of synchronization, like the waiting for all actin fibers and the two copies of each chromosome during cell division (probably the metaphase) to be in place so they can start pulling apart the cell nucleus (anaphase), and then the daughter cells can separate. Of course this also depends on signals (cyclins). The G1 cyclin slowly accumulates, and unless it's inhibited (by eg. signals get produced when "lack of nutrients" state happens around the cell) it'll start a cascade to ultimately split the cell in two. Similarly there are certain genes that are environmental gradient dependent, usually these work when a stem cell differentiates into something specific. And the stem cell stuff is active because some parts of the DNA is conveniently exposed, certain signaling pathways are active - so the cell reacts to the gradient. And after differentiation those things get put away, locked down, coiled up, broken down, absorbed, etc. Okay, so how could a cell skip all those iterations? It needs to know what to do and when, that's environment dependent (locally signaled by other cells, and sometimes signals arrive from far via special stuff in the blood, eg. hormones). Something like that could be probably implemented in cells. Eg. "simply" enumerating all the functions of a cell and making every process inside it depend on outer signals, and maybe leave the simple accumulators. And when any process is about to start the cell would need to know if it has enough resources to go through with it. This would basically need yet another signaling and computational system. Or, sure, it might just start things and ... fail to finish it in time, but then it needs to disassemble whatever is half-finished. And basically that's already what happens. All the time. And if something is "really needed" it'll be finished fast. (Because signals inhibit things that slow things down, and also maybe slow down the cleanup crew too.)
- foobarian 6y ago
  So it's kind of like React hooks, except they are probabilistic and constrained by 3d geometry. And the things moving around fast inside the cell is the event loop. Fun!
- kneel 6y ago
  Another analogy: molecule interactions are like proof of work operations that unlock cell actions when the right fit (hash) is found. trillions of molecule interactions are required to find the right fit. the intricate network of transcription factors operate as a security system to unlock key processes in DNA regulation in eukaryotes. security must be maintain because the system is under constant attack from other information systems (viruses)
- thaumasiotes 6y ago
  > the type safety (proteins are highly specific on what kind of chemicals they target) means you don’t need to worry about trying to perform some invalid operation. You might want to look into the mechanics of heavy metal poisoning. There is no type safety, and there's no type checking. Everything attempts to interact with everything it encounters. Those interactions usually fail.
- candiodari 6y ago
  The thing that is ignored here is that while molecules move fast ... they go nowhere. It's a random walk. It takes 6 hours for your small protein molecule to move 1cm ... at 250kph. So while these molecules offering themselves up "automatically" works, it only works at very short ranges and it's a stochastic range. At 1nm they meet 100k times per second. At 1cm once per 6 hours. At 3cm once per month. So there is in fact a lot of bookkeeping, keeping things tethered where they need to be and dragging things around to be done inside the cell where everything magically happens. People in this thread aren't correct that it's not necessary to do that. You don't need to get positions exactly right, as in you don't have to get the amino acids in every ribosome, but you DO have to get them "more or less" where they need to be. If for example if the amino acid concentrations near ribosomes ever get low, that's the end of you in about 2h (there's a fungus that causes that).
  - gotostatement 6y ago
    interesting. so the idea that this all happens just because every molecule interacts with every molecule many times per second, as advanced in the blog post, is not correct?
elcritch 6y ago
One big difference is the dimensionality, of 2d vs 3d. Even within the 3d space, quantum chemistry effects produce an even higher order of "effective" dimensionality. Finding stable paths that don't interact in 2d is hard, but easier in 3d. Its even easier in hyper dimensional spaces where other chemical properties can be used to provide pathways for only specific reactions.

Pulcinella 6y ago

Yeah it feels as if the cell is a ridiculously parallel array of molecules where every molecule in the array is rapidly iterating through the array (colliding with every molecule in the cell as it diffuses) and type checking those items to see if it can operate on them (e.g. a protein breaking down a small molecule). It all works out because the operations are atomic (heh) (don’t need to worry about two proteins trying to break down the same molecule at the same time) and the type safety (proteins are highly specific on what kind of chemicals they target) means you don’t need to worry about trying to perform some invalid operation. Turns out collision detection is real fast in real life!

xwdv 6y ago
What would happen if a cell were somehow to be optimized such that every operation occurs in O(1) time rather than having to iterate?
- shuntress 6y ago
  Cancer, probably.
- nsxwolf 6y ago
  This is probably what happened to Tetsuo in Akira.
- pas 6y ago
  Hm, well, it would require a "global" approach instead of what currently happens. Which is the opposite, local gradients, local diffusion, local parallel approaches. Many times things are being built, remodeled, moved, teared down, packed for moving, and unpacked for use. Everything is at the same time a tool for building bigger things and is just a piece of raw material if something else happens to disassemble it or use it for assembling something bigger. So opposed to that there would need to be some kind of ordering, signaling, queuing, synchronization, etc. Thanks to osmosis building block things are always pretty close when they are needed, and products are always getting spread out evenly too. (And there is a transport network in each cell that helps with those that need some help spreading.) Of course there are certain things that work based on a kind of synchronization, like the waiting for all actin fibers and the two copies of each chromosome during cell division (probably the metaphase) to be in place so they can start pulling apart the cell nucleus (anaphase), and then the daughter cells can separate. Of course this also depends on signals (cyclins). The G1 cyclin slowly accumulates, and unless it's inhibited (by eg. signals get produced when "lack of nutrients" state happens around the cell) it'll start a cascade to ultimately split the cell in two. Similarly there are certain genes that are environmental gradient dependent, usually these work when a stem cell differentiates into something specific. And the stem cell stuff is active because some parts of the DNA is conveniently exposed, certain signaling pathways are active - so the cell reacts to the gradient. And after differentiation those things get put away, locked down, coiled up, broken down, absorbed, etc. Okay, so how could a cell skip all those iterations? It needs to know what to do and when, that's environment dependent (locally signaled by other cells, and sometimes signals arrive from far via special stuff in the blood, eg. hormones). Something like that could be probably implemented in cells. Eg. "simply" enumerating all the functions of a cell and making every process inside it depend on outer signals, and maybe leave the simple accumulators. And when any process is about to start the cell would need to know if it has enough resources to go through with it. This would basically need yet another signaling and computational system. Or, sure, it might just start things and ... fail to finish it in time, but then it needs to disassemble whatever is half-finished. And basically that's already what happens. All the time. And if something is "really needed" it'll be finished fast. (Because signals inhibit things that slow things down, and also maybe slow down the cleanup crew too.)
foobarian 6y ago
So it's kind of like React hooks, except they are probabilistic and constrained by 3d geometry. And the things moving around fast inside the cell is the event loop. Fun!
kneel 6y ago
Another analogy: molecule interactions are like proof of work operations that unlock cell actions when the right fit (hash) is found. trillions of molecule interactions are required to find the right fit. the intricate network of transcription factors operate as a security system to unlock key processes in DNA regulation in eukaryotes. security must be maintain because the system is under constant attack from other information systems (viruses)
thaumasiotes 6y ago
> the type safety (proteins are highly specific on what kind of chemicals they target) means you don’t need to worry about trying to perform some invalid operation. You might want to look into the mechanics of heavy metal poisoning. There is no type safety, and there's no type checking. Everything attempts to interact with everything it encounters. Those interactions usually fail.
candiodari 6y ago
The thing that is ignored here is that while molecules move fast ... they go nowhere. It's a random walk. It takes 6 hours for your small protein molecule to move 1cm ... at 250kph. So while these molecules offering themselves up "automatically" works, it only works at very short ranges and it's a stochastic range. At 1nm they meet 100k times per second. At 1cm once per 6 hours. At 3cm once per month. So there is in fact a lot of bookkeeping, keeping things tethered where they need to be and dragging things around to be done inside the cell where everything magically happens. People in this thread aren't correct that it's not necessary to do that. You don't need to get positions exactly right, as in you don't have to get the amino acids in every ribosome, but you DO have to get them "more or less" where they need to be. If for example if the amino acid concentrations near ribosomes ever get low, that's the end of you in about 2h (there's a fungus that causes that).
- gotostatement 6y ago
  interesting. so the idea that this all happens just because every molecule interacts with every molecule many times per second, as advanced in the blog post, is not correct?

elcritch 6y ago

One big difference is the dimensionality, of 2d vs 3d. Even within the 3d space, quantum chemistry effects produce an even higher order of "effective" dimensionality. Finding stable paths that don't interact in 2d is hard, but easier in 3d. Its even easier in hyper dimensional spaces where other chemical properties can be used to provide pathways for only specific reactions.