From what I read, it seems the current situation is like this: latest cutting-edge lab techniques get to an accuracy of around 90%. For example, they put a protein under an electron microscope or something, and from the images obtained they achieve a precision of 90%. Now Deepmind features an AI that has demonstrated in contests that it can also virtually fold proteins to 90% accuracy. Therefore, the problem is kind of solved.

Please correct me if my interpretation is wrong. But if it's correct, then a big question remains: how the hell can they know if a configuration is 100% true? Even the best technology can only reach 90%, so where do they get that 100% for comparison?

  • $\begingroup$ Just a guess - they could have created a synthetic protein where they know the exact sequence and used it as a reference? $\endgroup$
    – user438383
    Dec 21, 2020 at 16:04
  • $\begingroup$ Can you tell a bit more on the source which claims 90% precision of the laboratory methods of structural determination? $\endgroup$ Dec 22, 2020 at 17:37
  • 1
    $\begingroup$ It seems I've made a mistake there. Lab results were actually set out as the 100 threshold, and Alphafold was compared against it. Because even lab techniques can have errors, they consider 90+ scores are about as good as 100. $\endgroup$
    – longtry
    Dec 23, 2020 at 2:12


You must log in to answer this question.

Browse other questions tagged .