Lior Pachter said “Deep Mind AlphaFold (protein folding) results are just markedly different from what a lot of other methods are producing. This is not an incremental improvement.”
But protein folding is not solved. Not only is it not even a well-defined statement to say something like that (others have pointed out that there is a lot of subtlety in what one even means by “protein folding”) but it’s not even the winner for all the CASP14 proteins.
I don’t mind that Google hyped this. It’s impressive work they did and there are super exciting prospects for the field. I do mind that many (computational) biologists who ought to know better are going around screaming “protein folding is solved!” Have some self respect.
— Lior Pachter (@lpachter
Stephen Curry provides this analysis –
Firstly, there is no doubt that DeepMind have made a big step forward. Of all the teams competing against one another they are so far ahead of the pack that the other computational modellers may be thinking about giving up. But we are not yet at the point where we can say that protein folding is ‘solved’. For one thing, only two-thirds of DeepMind’s solutions were comparable to the experimentally determined structure of the protein. This is impressive but you have to bear in mind that they didn’t know which two-thirds of their predictions were correct until the comparison with experimental solutions was made.
Alphafold 2 will certainly help to advance biology. For example, as already reported, it can generate folded structure predictions that can then be used to solve experimental structures by crystallography (and probably other techniques). So this will help the science of structure determination go a bit faster in some cases.
However, despite some of the claims being made, we are not at the point where this AI tool can be used for drug discovery.
Alphafold 2 is predicting structures to an average accuracy of 0.16 nanometers but to get t0 reliable insights into protein chemistry or drug design we will need an average structure prediction accuracy of 0.03 nanometers.
A friend (who does not work in science) asked me today whether it is true that "protein folding has been solved". My short answer:
The AlphaFold method produced very impressive results on CASP14. Protein folding is not a solved problem. pic.twitter.com/ZMc4grC5iP
— Lior Pachter (@lpachter) December 1, 2020
The AlphaFold results are impressive not just because they are (on average) much better than other methods, but because the improvement is so great in just the last 2 years that it suggests much more is still possible.
— Lior Pachter (@lpachter) December 1, 2020
CASP is both the gold standard for assessing predictive techniques and a unique global community built on shared endeavour. Accuracy is measured on a range of 0-100 “GDT”. #AlphaFold has a median score of 92.4 GDT across all targets – its average error about the width of an atom. pic.twitter.com/cYCN12KxLZ
— DeepMind (@DeepMind) November 30, 2020
Brian Wang is a Futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
They are further than you think. There is no water in these models. Typically, functional agonist and antagonist molecules look very similar. The lack of hydration essentially renders these computations useless. This is a very old concern which the computational biologists just sweep under the rug because they don't know where to start. Rightly there should be research but what's needed are better basic understanding of water. Instead big money is pouring into computing structures in a fake waterless world.
Actually getting serious sick of these butthurt hot takes from people who don't like how progress like this reflects on their own career. That's what this is, pure spiteful resentment, antithetical to scientific progress, and only interested in ego and aggrandisement.
A useful critical comment that helps us understand the Alpha Fold advance in context.
Hasn't 2020 had enough sore losers? No, it is not 100%, but there is no reason you could not try a drug engineered to fit in the model it makes. Maybe it will fit, maybe it won't, but that beats waiting a year. And they will, of course, further improve the program.
We have seen this putdown stuff before with chess programs. The truth is that they don't know that the shapes of the proteins, when crystallized, look exactly the same as when they are in their natural environment. And they probably don't. The proximity of other identical molecules almost certainly affects the shape.
Until they start using this to design drugs, they will not know how it compares to crystallography. The machine could already be more accurate.
Run their protein folding program on a cerebras, and see what they can solve. The computer they use isn't even half as good as a Cerebras trillion transistor chip; then, they have a 2.6 trillion chip coming soon enough as well!
You always want different takes on something like this.
If Fermat really did have an easy proof for his last theory, it is just as well it was too big for the margin—otherwise Wiles approach would never have been invented.
Now for this to be done with better quantum computers.
I want to see such programs do fictional starship drawings