Mohammed AlQuraishi

v2 looks amazing. that jump is even more incredible than the first. More context from v1 in 2018: https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

The Glass Bead Game is one of the best books I've read, and I think palpably changed how I think about some things even though I read it as an adult. I even wrote a blog post about it: https://moalquraishi.wordpress.com/2013/05/05/the-glass-bead-game-by-h… Another book I'd recommend (in a very different category) is Godel, Escher, and Bach, by Douglas Hofstadter.

For background, read this fantastic blog post by the same author, Mohammed AlQuraishi at Harvard Medical School, from a year ago: https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

The best take on AlphaFold: https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

It's the ultimate prediction problem. And short term, computational bio methods are driving R&D costs up, not down. But arguably, techniques like DeepFold constitute current best hope https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

Mohammed AlQuraishi elaborated on this in his 2018 "What Just Happened?" blog post that was discussed on HN when the news was announced. [1] It's a fairly long read, but he goes into a lot more detail as to why the problem isn't yet solved. Interestingly, he also notes that pharma and academia should feel some embarrassment from DeepMind's achievement: "What is worse than academic groups getting scooped by DeepMind? The fact that the collective powers of Novartis, Pfizer, etc, with their hundreds of thousands (~million?) of employees, let an industrial lab that is a complete outsider to the field, with virtually no prior molecular sciences experience, come in and thoroughly beat them on a problem that is, quite frankly, of far greater importance to pharmaceuticals than it is to Alphabet. It is an indictment of the laughable “basic research” groups of these companies, which pay lip service to fundamental science but focus myopically on target-driven research that they managed to so badly embarrass themselves in this episode." [1] https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju… (Edited to clarify that the blog post was not recent.)

"There are dozens of academic groups, with researchers likely numbering in the (low) hundreds, working on protein structure prediction. We have been working on this problem for decades, with vast expertise built up on both sides of the Atlantic and Pacific, and not insignificant computational resources when measured collectively. For DeepMind’s group of ~10 researchers, with primarily (but certainly not exclusively) ML expertise, to so thoroughly route everyone surely demonstrates the structural inefficiency of academic science." "What is worse than academic groups getting scooped by DeepMind? The fact that the collective powers of Novartis, Pfizer, etc, with their hundreds of thousands (~million?) of employees, let an industrial lab that is a complete outsider to the field, with virtually no prior molecular sciences experience, come in and thoroughly beat them on a problem that is, quite frankly, of far greater importance to pharmaceuticals than it is to Alphabet. It is an indictment of the laughable “basic research” groups of these companies, which pay lip service to fundamental science but focus myopically on target-driven research that they managed to so badly embarrass themselves in this episode." From: https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

> Let me get the most important question out of the way: is AlphaFold’s advance really significant, or is it more of the same? I would characterize their advance as roughly two CASPs in one ―https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

I would pay skepticism to entrepreneurial commentaries on AlphaFold (CEO of a company based on competing technology has hot take, who would have guessed?), but I would mind the caveats and areas of improvement pointed out by domain experts i.e. Baker, AlQuraishi, etc. Keeping in mind that the latter usually spend more time praising AlphaFold than not -- and yes, they have more domain knowledge than pretty much anyone on HN: https://moalquraishi.wordpress.com/2021/07/25/the-alphafold2-method-pa… It's almost entirely positive. There was a slice in time where I was effectively paid to find as many gaps in AlphaFold as I could, and, even then, I couldn't help but be impressed.

An academic tour de force of little practical benefit, because of insufficient openness from the AF2 team. For nuance, see https://moalquraishi.wordpress.com/2020/12/08/alphafold2-casp14-it-fee…

I think the story that rammed that home to me is this one, which is, ironically, about a stagnant (but important!) scientific field being totally wiped out by Google in the space of just 4 years by applying AI to the problem: https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju… https://moalquraishi.wordpress.com/2020/12/08/alphafold2-casp14-it-fee… Yes, progress was (not exactly) coming through tons of tiny incremental advances, right up until an industrial lab got involved - as a side project no less - at which point the academics were blown out of the water.

I have wondered about folding@home as well, another distributed project begun in another age that seems not, from the outside, to have lived up to its early hype. In the meantime, it seems that Alphafold has surpassed conventional efforts at understanding protein folding, including folding@home's. https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

According to at least one expert in the field, it's a bit of both. (https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

Do you know if he turned to deep learning after AlphaFold's stunning performance at CASP13 in 2018? I haven't heard anything from that particular niche since the "AlphaFold @ CASP13: “What just happened?”" blog post: https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

This blog post seems to anticipate this happening: https://moalquraishi.wordpress.com/2019/04/01/the-future-of-protein-sc…

Interesting post by the author that puts this into perspective of the current academic landscape: https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

https://deepmind.com/blog/alphafold/ https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

> AlphaFold is an important advance, and it is laughable that the author is dismissing it citing NP-hardness Well, the NP-hardness argument might not be the relevant one, but how useful the structures determined by AF2 are has yet to be demonstrated. Protein folding is very complex; what AF2 has in the training set and CASP in the test set are proteins which were accessible to structure determination up to now at all; most proteins were measured in crystallized (i.e. not their natural) form, so the resulting static structure is likely not representative; and not to forget that many proteins get another conformation than the one to be expected by thermodynamics etc. e.g. because they're integrated in a complex with other proteins and/or "modified" by chaperones; so it would be quite naive to assume that from now on you can just throw a sequence into the black box and the right structure comes out. This article two clicks away from the referenced one is much more relevant: https://moalquraishi.wordpress.com/2020/12/08/alphafold2-casp14-it-fee…

[Levinthal's paradox](https://en.wikipedia.org/wiki/Levinthal's_paradox) was the thought experiment that "proved" that protein folding is non-random. This is currently the dominant attitude in the field. As others have stated, misfoldings do occur, and the protein is trying to achieve its lowest energy conformation, but the simplest answer is that we don't have a complete answer, hence the desire to come up with a reliable computational solution. We know how certain residues affect tertiary structure, and we know the measurements of secondary structures with a surprising degree of accuracy, but for a protein with potentially hundreds of residues, there are simply too many degrees of freedom to simply tack on amino acids and come up with a structure (although this is essentially the approach that markov chain, monte carlo solutions use). [This article gives a pretty good synopsis of where computational approaches to the protein folding problem stand.](https://moalquraishi.wordpress.com/2018/12/09/alphafold-casp13-what-ju…

> Therefore we should design programming languages that on the surface look very unoptimized, but specify enough information that compilers can do a really good job. It sounds like no one knows what such a programming language would look like. This entire post would be painfully familiar to anyone working with probabilistic programming. An intriguing overview of the state of the art was posted here a little while ago: https://moalquraishi.wordpress.com/2015/03/29/the-state-of-probabilist… The task here (I'm adapting an older comment of mine) is that the human should write a relatively simple procedural program that says "if we were to fix a bunch of properties/states of the natural world and provide them as input, here's the output/reading you'd hypothetically get," and then from that, the "compiler" and runtime create a function that, given a large number of outputs/readings, provides a posterior probability distribution over those hidden properties/states of the world. The implementation choices for this "function" can make a difference of days or weeks of runtime on huge computer clusters. The problem is that machine learning researchers use all sorts of insights to make manual optimizations when hand-rolling machine learning code - for instance, if there are real-world reasons why a certain type of search over your space may be well-suited or ill-suited to the problem domain. And not only can the languages not capture these insights yet, but even if they…