[ivan.moony]

DeepMind's protein-folding AI has solved a 50-year-old grand challenge of biology

Protein folding is a problem that is linked to traveling salesman problem, that is in turn linked to unsolved P vs. NP problem. Currently, solving such problems is doable, but it takes a tremendous amount of time to solve them (ranges over centuries of computing for 200+ parameters). Allegedly, AplhaFold (DeepMind's protein-folding solving AI) can find the accurate protein folding in a mater of days. This is that black-box effect of ANNs where a solution is found without knowing how it is really solved. We train the network for a few days, and it magically solves all similar problems afterwards without explaining how the problems are really solved.

To anchor DeepMind's achievement in the real world, protein folding prediction could help finding new drugs and vaccines in less trial-error cycles and with more precision in fighting diseases. In a case of Covid-19, an anti-protein (whose folding fits structurally tight onto Covid-19 critical zone) could be fastly computed, which leads to more effective vaccines in less time spent in laboratory.

Read the full story here.

[MagnusWootton]

But where is the result?!?!?!

If they can do that more NP problems should be all solved as well,  seems a bit fishy to me,  but it still could be true that they did because things are a little cloak and dagger about this technology stuff in general.

[LOCKSUIT]

But where is the result?!?!?!

If they can do that more NP problems should be all solved as well,  seems a bit fishy to me,  but it still could be true that they did because things are a little cloak and dagger about this technology stuff in general.

"AlphaFold builds on the work of hundreds of researchers around the world. DeepMind also drew on a wide range of expertise, putting together a team of biologists, physicists and computer scientists. Details of how it works will be released this week at the CASP conference and in a peer-reviewed article in a special issue of the journal Proteins next year. But we do know that it uses a form of attention network, a deep-learning technique that allows an AI to train by focusing on parts of a larger problem. Jumper compares the approach to assembling a jigsaw: it pieces together local chunks first before fitting these into a whole.

DeepMind trained AlphaFold on around 170,000 proteins taken from the protein data bank, a public repository of sequences and structures. It compared multiple sequences in the data bank and looked for pairs of amino acids that often end up close together in folded structures. It then uses this data to guess the distance between pairs of amino acids in structures that are not yet known. It is also able to assess how accurate these guesses are. Training took “a few weeks,” using computing power equivalent to between 100 and 200 GPUs."

[MagnusWootton]

Believe me or not, neural networks are actually an NP problem already!,  because the amount of states is the exponent of how many input neurons they have,   so we shouldnt even be able to train feedforward NN's because there is too many states to train,  but they do actually work still,   so some NP seeming things arent actually NP.    P suffices to get the job done even tho the problem looks exponential.