DeepMind finally cracks the protein-folding problem at the CASP14 competition (Or rather, is almost there; see here and here for a more sober take). As noted here a while back, it's astonishing to observe how outsiders have come into a field and solved the core problem the field had been working on for decades. Good news in that it means there may be free lunches waiting for us when better equipped outsiders (With money, AI, or anything else) enter new fields.
Does this constitute a solution of the static protein structure prediction problem? I think so but there are all these wrinkles. Honest, thoughtful people can disagree here and it comes down to one’s definition of what the word “solution” really means. Let me explain why I consider this a solution.
While the current list of caveats is a long one, making it seem that AF2 has a way to go before tackling all corner cases and elaborations, it is my expectation that this is not the case because the core intellectual problem has been solved. I believe that everything on the preceding bullet list, excepting the very last item (and possibly the second-to-last) is now an engineering rather than a scientific problem.
I promised to write one piece of armchair sociology and so here it is. Why was it DeepMind, rather than an academic group, that built AF2?
(From the second link)
Looking at various TFP trends. Weirdly, the author says that "The received wisdom is that we are living through a period of slow TFP growth, and that this slowdown in TFP growth kicked in around the early 2000s.", while in my online circles no one ever makes that claim; standard wisdom is what the author ends up showing, that the stagnation in TFP started around 1970.
Can breakthrough innovations be made systematically? A podcast with Noubar Afeyan (co-founder of Moderna and founder and CEO of Flagship Pioneering)
The Hallmarks of Health (In addition to the Hallmarks of Cancer, and the Hallmarks of Aging)
Predicting scientific breakthroughs (in the paper, Nobel Prize-winning papers) based on knowledge structure variations (How the paper is cited early on). Interesting, and the control they are using is not "all papers" but the harder "papers that were published in the same journals and got a similar number of citations four years after publication". AUC is ~0.6 so not great, but they just used citations, not anything in the text itself.
Is it possible to forecast scientific breakthroughs in the early stage, based on the radical changes they proceed to bring about following breakthroughs’ birth? In Kuhn’s time, this seemed difficult, as scientific revolutions were not easily validated in empirical data. However, with the development of digitalization technologies, large-scale scholarly data has become widely available, bringing the possibility to quantitatively study scientific breakthroughs—even scientific revolutions—from the data. In this study, we aim to answer the previous question by exploring the unique features that breakthroughs have left in the space of science, from the perspective of their early citing structures
For the first time ever, an artificial thymus has been produced (From stem cells and a decellullarised rat thymus)
A workforce training program (Year Up) for low-income young adults shows impressive effects in a multi-city RCT, even 5 years after the program finished.
The thymus involutes, so does the pineal gland. How does this matter for aging?
Matt Clancy on the publish-or-perish system and how that affects the quality of science. Also some interesting peculiarities of how the protein folding field works as a whole.
Stripe: Platform of platforms
Reducing costs by a factor of 1900x in the US Navy.
No great battery stagnation, cost reductions continue
Watch Adam Marblestone and I present FROs and some of our aging-related work.
The number of cells sequenced in a single study has increased exponentially over the last decade.
Huge increases in the accuracy of Cryo-EM in the last few years (Used to determine protein structure)
A truly glorious year for semiconductors; after AMD's Zen3 and Apple's M1 now RISC-V cores achieve a new milestone in performance per watt, being 550x more efficient than the M1 itself.
It's hard to get drugs to, for example, the brain. But some parasites do get there. It then takes an extra step to engineer parasites for drug delivery.
Milestones in cancer research. And my hot take on what the future of cancer therapy should be. Twitter tells me I'm wrong. Also, I noted earlier this year that there had been a potential breakthrough towards a universal cancer therapy at Cardiff University which went somewhat unnoticed. That line of research has continued and so far it still seems promising.