6 ms·

So do you think it actually understands something about the structure of the protein folding problem? It somehow detected something about the physics, topology,

by foxes 5y ago

So do you think it actually understands something about the structure of the protein folding problem? It somehow detected something about the physics, topology, the hard optimisation problem, and that it knows something about the geometry of that potential surface and can exploit that? Or is it just such a huge model it basically encodes an entire database after weeks and weeks of computation and has a more compressed form?

wiz21c 5y ago
My (basic) understanding is that 1/ there's some inductive bias (knowledge from researchers) 2/ data is definitely "compressed" in some ways 3/ since the model predicts better than the others, then it actually found some relationships in the data that were not found before. From what I understand, deep learning, although opaque and relying on tons of data, is a bit magical : although one would say "it's just probabilities", it actually does probabilities at a level where it actually figures some things. Plus, and that's very much a problem to me, Google does it at 100x the scale of a regular researcher. Since I just invested a year in studying data sciences, that worries me a lot : where am I suppose to work if, to produce meaningful results, you need way-too-expensive hardware...
- thomasahle 5y ago
  > where am I suppose to work if, to produce meaningful results, you need way-too-expensive hardware... I know how you feel, but I also think stories like this may be a wake-up call for some groups to invest more in hardware. The Baker and Tsinghua University groups are not small. They can afford more than 4 GPUs. Probably it's more about setting up a good pipeline. Once you get about 4 GPUs you need more than one machine to run them. Hopefully in the next years we'll see more open source tools to make it easy to "build your own GPU cloud".
  - dumb1224 5y ago
    And to use all that GPU power exclusively you need some good strategy too. Speaking from a medium size biology research centre.
    - thomasahle 5y ago
      True. It may be that a smaller research center will only rarely be able to saturate the cluster. Maybe it doesn't matter, like how other lab equipment is not in constant use. Another option may be for centers to team up and have shared machines? Or maybe compute as a service will eventually be cheap enough for this not to matter...
      - dumb1224 5y ago
        Currently only those team who are heavy-deep-learning got special exclusive queues on the GPU nodes on the cluster. If many users want to use the GPUs at the same time it might need some planning. I don't know if it is a solved problem in the HPC field though.
- sgt101 5y ago
  >if, to produce meaningful results, you need way-too-expensive hardware... If you are in a team that is looking at problems that justify massive hardware (in the sense that solving them will pay back the capital and environmental cost) then you will have access to said hardware. Most (almost all) AI and Data Science teams are not working on that kind of problem though, and it's often the case that we are working on cloud infrastructures where GPU's and TPU's can be accessed on demand and $100's can be used to train pretty good models. Obviously models need to be trained many times so the crossover point from a few $100 to a few $1000 can be painful - but actually many/most engagements really only need models that cost <$100 to train. Also many of the interesting problems that are out there can utilize transfer learning over shared large pretrained models such as Resnet or GPT-2 (I know that in the dizzying paced modern world these no longer count as large or modern but they are examples...) So for images and natural language problems we can get round the intractable demand for staggeringly expensive compute. Imagine that you had got a degree in Aeronautical Engineering, you are watching the Apollo program and wondering how you will get a job at NASA or something similar... but there are lots of jobs at Boeing and Lockheed and Cessna and so on.
- touisteur 5y ago
  Most big labs have large computing clusters and upgrade them from time to time. We've almost always needed huge computing power in the scientific domain, no? Anyway these days I see a lot of industrial investment in middle size computing datacenters full of GPUs. Sure Google scale is not within reach but I'm sure there's room for scrapier algorithms and training methods to demonstrate feasibility and then paying a lump sum to AWS (or some ml-expert-for-cloud-training SME) for the final step. Anyway, I thought the expensive part was data acquisition and labeling? I like the 'surrogate network' approach of learning a very-expensive-to-compute simulation or model, that doesn't need data, but the output of a costly simulation.
shawnz 5y ago
What's the difference between "knowing things" and finding a more compressed form of the solution space?
- hans1729 5y ago
  queue joscha bach: >Scientific logic is proving things by losslessly compressing statements to their axioms. Commonsense logic uses lossy compression, which makes it less accurate in edge cases, but also less brittle, more efficient, further reaching and more stable in most real-world situations.
- garmaine 5y ago
  Being able to extrapolate beyond mere variations of the training data. EDIT: A simpler example might be helpful. We could, for example, train a network to recognize and predict orbital trajectories. Feed it either raw images or processed position-and-magnitude readings, and it outputs predicted future observations. One could ask, "does it really understand orbital mechanics, or is it merely finding an efficient compression of the solution space?" But this question can be reduced in such a way as to made empirical by presenting the network with a challenge that requires real understanding to solve. For example, show it observations of an interstellar visitor on a hyperbolic trajectory. ALL of its training data consisted of observations of objects in elliptical orbits exhibiting periodic motion. If it is simply matching observation to its training data, it will be unable to conceive that the interstellar visitor is not also on a periodic trajectory. But on the other hand if it really understood what it was seeing then it would understand (like Kepler and Newton did) that elliptical motion requires velocities bounded by an upper limit, and if that speed is exceeded then the object will follow a hyperbolic path away from the system, never to return. It might not conceive these notions analytically the way a human would, but an equivalent generalized model of planetary motion must be encoded in the network if it is to give accurate answers to questions posed so far outside of its training data. How you translate this into AlphaFold I'm not so certain, as I lack the domain knowledge. But a practical ramification would be the application of AlphaFold to novel protein engineering. If AlphaFold lacks "real understanding", then its quality will deteriorate when it is presented with protein sequences further and further removed from its training data, which presumably consists only of naturally evolved biological proteins. Artificial design is not as constrained as Darwinian evolution, so de novo engineered proteins are more likely to diverge from AlphaFold's training data. But if AlphaFold has an actual, generalized understanding of the problem domain, then it should remain accurate for these use cases.
robbedpeter 5y ago
It's likely that many of the patterns it learned are encoded understanding of the form you mention, but not at a formal level of explication. The architecture of the system and design of the training methodology are laid out to specifically prevent direct database-esque "pattern in pattern out" failure mode. Similar to Google deep dream, there will be contextual features and feature clusters encoded into neurons that can be explored and extracted, and those could provide insights that can be sourcing translated into "hard" science, with explicit formulae and theory allowing a fully transparent model to be created. Like other transformer models, you can elicit the training data intact, but such scenarios are a statistically insignificant selection of the range of outputs the models are capable of producing. That doesn't mean anything with regards to accuracy of the novel output, though. With alphafold 2 going open source, it's possible that tools and methodologies to extract hard science from transformers will be formalized quickly and in the public eye. We have an amazingly powerful new tool, and the coming decades will be fascinating.
- joe_the_user 5y ago
  It's likely that many of the patterns it learned are encoded understanding of the form you mention, but not at a formal level of explication. - The thing I'd be curious about is whether or not "not formalized" would imply "not consistently generalizing", whether it would have to be train all over if given a problem similar to but identical with, the problem it solves.
  - robbedpeter 5y ago
    No, there wouldn't be an implication toward performance by "not formalized." I mean the function is obfuscated by its encoding in the model. To formalize, you'd need to pick apart those encodings and derive explicit formula and conjectures. The models in situ are essentially black boxes, but that can be fixed. The tools to do that with transformers in particular and neural networks in general are still pretty new and specific to particular models. https://ai.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html https://ai.googleblog.com/2015/06/inceptionism-going-deeper-... Someone with extensive knowledge of protein folding and chemistry and any relevant domain would need to manually categorize different neurons, their contexts, and effects, then experiment with the model in a static/procedural inference mode, then parse out the overall functions. By reproducing well known simple instances and tracing the propagation of information through the network, you can validate what a model knows by mapping to current knowledge. Using the model in this way can also lead to interesting real world experiment choices, by processing targets suggested by real life phenomena and then coordinating from known functions of the model and real life experiment. Different proteins in coronavirus, for example, would likely have a different cascade of activations in the model than the flu virus. Any shared features between the two might be captured in the activation patterns, possibly leading to new information about the diseases that would opaque to science without these models.
l33tman 5y ago
Nature has re-used existing folds all over the place (partially because of genetic mutation but also because it's improbable to come up with stable folds from scratch by evolution I would guess), this was encapsulated by earlier award-winning systems like Rosetta. There is probably a finite number of folds in nature, with most of the difference being in the outward-facing amino acids "citation needed" :) So an extremely large DL network would have a good chance to find and integrate ("compress") all the existing folds and sub-folds that human researchers or Rosetta missed or just hadn't the time to investigate and characterize yet (I'm not an expert on Rosetta by far btw so please expand if you are :). I would venture to say it's a good problem fit for DL methods (as was impressively demonstrated). Regarding your question, "does it understand something about the structure of the protein folding problem" - expanding on the above, I would say it understands enough, but it probably doesn't understand the generics of chemistry as protein and their folding is a biased subset. The output is (as far as I remember) an atom distance matrix and not atom trajectories etc. so folding dynamics is not part of the model (this is btw an important part of protein science as well).
twanvl 5y ago
An important input to this (and similar) algorithms is multiple sequence alignment, which tells the algorithm which parts of proteins are preserved between species and variants, and which amino-acids mutate together. So already it is relying on natural selection to do some of the work. And the algorithm will probably not work very well if you input a random sequence not found in nature and ask it to find the folding.
- haihaibye 5y ago
  I hope not, as knowing how a novel mutation in a patient alters a protein would be extremely useful when trying to find disease causing variants.
dekhn 5y ago
it's pretty clear what it does. It uses the evolutionary information expressed in multiple sequence alignments to make reasonable judgements about interatomic distances, which are used as constraints for a force field. We've been doing variations on this for decades. The evolutionary information encoded in multiple sequence alignments is pretty much all you need to fold homologous proteins (apparently). No, this technique doesn't do anything about the harder problems of actually understanding the raw physics of protein folding (nor, does it seem, that we need that to solve downstream problems).

by foxes 5y ago

wiz21c 5y ago

My (basic) understanding is that 1/ there's some inductive bias (knowledge from researchers) 2/ data is definitely "compressed" in some ways 3/ since the model predicts better than the others, then it actually found some relationships in the data that were not found before. From what I understand, deep learning, although opaque and relying on tons of data, is a bit magical : although one would say "it's just probabilities", it actually does probabilities at a level where it actually figures some things. Plus, and that's very much a problem to me, Google does it at 100x the scale of a regular researcher. Since I just invested a year in studying data sciences, that worries me a lot : where am I suppose to work if, to produce meaningful results, you need way-too-expensive hardware...

thomasahle 5y ago
> where am I suppose to work if, to produce meaningful results, you need way-too-expensive hardware... I know how you feel, but I also think stories like this may be a wake-up call for some groups to invest more in hardware. The Baker and Tsinghua University groups are not small. They can afford more than 4 GPUs. Probably it's more about setting up a good pipeline. Once you get about 4 GPUs you need more than one machine to run them. Hopefully in the next years we'll see more open source tools to make it easy to "build your own GPU cloud".
- dumb1224 5y ago
  And to use all that GPU power exclusively you need some good strategy too. Speaking from a medium size biology research centre.
  - thomasahle 5y ago
    True. It may be that a smaller research center will only rarely be able to saturate the cluster. Maybe it doesn't matter, like how other lab equipment is not in constant use. Another option may be for centers to team up and have shared machines? Or maybe compute as a service will eventually be cheap enough for this not to matter...
    - dumb1224 5y ago
      Currently only those team who are heavy-deep-learning got special exclusive queues on the GPU nodes on the cluster. If many users want to use the GPUs at the same time it might need some planning. I don't know if it is a solved problem in the HPC field though.
sgt101 5y ago
>if, to produce meaningful results, you need way-too-expensive hardware... If you are in a team that is looking at problems that justify massive hardware (in the sense that solving them will pay back the capital and environmental cost) then you will have access to said hardware. Most (almost all) AI and Data Science teams are not working on that kind of problem though, and it's often the case that we are working on cloud infrastructures where GPU's and TPU's can be accessed on demand and $100's can be used to train pretty good models. Obviously models need to be trained many times so the crossover point from a few $100 to a few $1000 can be painful - but actually many/most engagements really only need models that cost <$100 to train. Also many of the interesting problems that are out there can utilize transfer learning over shared large pretrained models such as Resnet or GPT-2 (I know that in the dizzying paced modern world these no longer count as large or modern but they are examples...) So for images and natural language problems we can get round the intractable demand for staggeringly expensive compute. Imagine that you had got a degree in Aeronautical Engineering, you are watching the Apollo program and wondering how you will get a job at NASA or something similar... but there are lots of jobs at Boeing and Lockheed and Cessna and so on.
touisteur 5y ago
Most big labs have large computing clusters and upgrade them from time to time. We've almost always needed huge computing power in the scientific domain, no? Anyway these days I see a lot of industrial investment in middle size computing datacenters full of GPUs. Sure Google scale is not within reach but I'm sure there's room for scrapier algorithms and training methods to demonstrate feasibility and then paying a lump sum to AWS (or some ml-expert-for-cloud-training SME) for the final step. Anyway, I thought the expensive part was data acquisition and labeling? I like the 'surrogate network' approach of learning a very-expensive-to-compute simulation or model, that doesn't need data, but the output of a costly simulation.

shawnz 5y ago

What's the difference between "knowing things" and finding a more compressed form of the solution space?

hans1729 5y ago
queue joscha bach: >Scientific logic is proving things by losslessly compressing statements to their axioms. Commonsense logic uses lossy compression, which makes it less accurate in edge cases, but also less brittle, more efficient, further reaching and more stable in most real-world situations.
garmaine 5y ago
Being able to extrapolate beyond mere variations of the training data. EDIT: A simpler example might be helpful. We could, for example, train a network to recognize and predict orbital trajectories. Feed it either raw images or processed position-and-magnitude readings, and it outputs predicted future observations. One could ask, "does it really understand orbital mechanics, or is it merely finding an efficient compression of the solution space?" But this question can be reduced in such a way as to made empirical by presenting the network with a challenge that requires real understanding to solve. For example, show it observations of an interstellar visitor on a hyperbolic trajectory. ALL of its training data consisted of observations of objects in elliptical orbits exhibiting periodic motion. If it is simply matching observation to its training data, it will be unable to conceive that the interstellar visitor is not also on a periodic trajectory. But on the other hand if it really understood what it was seeing then it would understand (like Kepler and Newton did) that elliptical motion requires velocities bounded by an upper limit, and if that speed is exceeded then the object will follow a hyperbolic path away from the system, never to return. It might not conceive these notions analytically the way a human would, but an equivalent generalized model of planetary motion must be encoded in the network if it is to give accurate answers to questions posed so far outside of its training data. How you translate this into AlphaFold I'm not so certain, as I lack the domain knowledge. But a practical ramification would be the application of AlphaFold to novel protein engineering. If AlphaFold lacks "real understanding", then its quality will deteriorate when it is presented with protein sequences further and further removed from its training data, which presumably consists only of naturally evolved biological proteins. Artificial design is not as constrained as Darwinian evolution, so de novo engineered proteins are more likely to diverge from AlphaFold's training data. But if AlphaFold has an actual, generalized understanding of the problem domain, then it should remain accurate for these use cases.

robbedpeter 5y ago

It's likely that many of the patterns it learned are encoded understanding of the form you mention, but not at a formal level of explication. The architecture of the system and design of the training methodology are laid out to specifically prevent direct database-esque "pattern in pattern out" failure mode. Similar to Google deep dream, there will be contextual features and feature clusters encoded into neurons that can be explored and extracted, and those could provide insights that can be sourcing translated into "hard" science, with explicit formulae and theory allowing a fully transparent model to be created. Like other transformer models, you can elicit the training data intact, but such scenarios are a statistically insignificant selection of the range of outputs the models are capable of producing. That doesn't mean anything with regards to accuracy of the novel output, though. With alphafold 2 going open source, it's possible that tools and methodologies to extract hard science from transformers will be formalized quickly and in the public eye. We have an amazingly powerful new tool, and the coming decades will be fascinating.

joe_the_user 5y ago
It's likely that many of the patterns it learned are encoded understanding of the form you mention, but not at a formal level of explication. - The thing I'd be curious about is whether or not "not formalized" would imply "not consistently generalizing", whether it would have to be train all over if given a problem similar to but identical with, the problem it solves.
- robbedpeter 5y ago
  No, there wouldn't be an implication toward performance by "not formalized." I mean the function is obfuscated by its encoding in the model. To formalize, you'd need to pick apart those encodings and derive explicit formula and conjectures. The models in situ are essentially black boxes, but that can be fixed. The tools to do that with transformers in particular and neural networks in general are still pretty new and specific to particular models. https://ai.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html https://ai.googleblog.com/2015/06/inceptionism-going-deeper-... Someone with extensive knowledge of protein folding and chemistry and any relevant domain would need to manually categorize different neurons, their contexts, and effects, then experiment with the model in a static/procedural inference mode, then parse out the overall functions. By reproducing well known simple instances and tracing the propagation of information through the network, you can validate what a model knows by mapping to current knowledge. Using the model in this way can also lead to interesting real world experiment choices, by processing targets suggested by real life phenomena and then coordinating from known functions of the model and real life experiment. Different proteins in coronavirus, for example, would likely have a different cascade of activations in the model than the flu virus. Any shared features between the two might be captured in the activation patterns, possibly leading to new information about the diseases that would opaque to science without these models.

l33tman 5y ago

Nature has re-used existing folds all over the place (partially because of genetic mutation but also because it's improbable to come up with stable folds from scratch by evolution I would guess), this was encapsulated by earlier award-winning systems like Rosetta. There is probably a finite number of folds in nature, with most of the difference being in the outward-facing amino acids "citation needed" :) So an extremely large DL network would have a good chance to find and integrate ("compress") all the existing folds and sub-folds that human researchers or Rosetta missed or just hadn't the time to investigate and characterize yet (I'm not an expert on Rosetta by far btw so please expand if you are :). I would venture to say it's a good problem fit for DL methods (as was impressively demonstrated). Regarding your question, "does it understand something about the structure of the protein folding problem" - expanding on the above, I would say it understands enough, but it probably doesn't understand the generics of chemistry as protein and their folding is a biased subset. The output is (as far as I remember) an atom distance matrix and not atom trajectories etc. so folding dynamics is not part of the model (this is btw an important part of protein science as well).

twanvl 5y ago

An important input to this (and similar) algorithms is multiple sequence alignment, which tells the algorithm which parts of proteins are preserved between species and variants, and which amino-acids mutate together. So already it is relying on natural selection to do some of the work. And the algorithm will probably not work very well if you input a random sequence not found in nature and ask it to find the folding.

haihaibye 5y ago
I hope not, as knowing how a novel mutation in a patient alters a protein would be extremely useful when trying to find disease causing variants.

dekhn 5y ago

it's pretty clear what it does. It uses the evolutionary information expressed in multiple sequence alignments to make reasonable judgements about interatomic distances, which are used as constraints for a force field. We've been doing variations on this for decades. The evolutionary information encoded in multiple sequence alignments is pretty much all you need to fold homologous proteins (apparently). No, this technique doesn't do anything about the harder problems of actually understanding the raw physics of protein folding (nor, does it seem, that we need that to solve downstream problems).