As a medical student, I used the dragon dictation software (no AI) to write notes in the ED and more recently I used a pilot of this ai version to write clinic notes.
Overall, I was quite impressed. It definitely made writing notes much faster, which all doctors hate to do. While it had some problems with where to put key pieces of information (like putting details from the physical exam back in the history), it only took 5 mins of rearrangement after the visit to complete the note.
For simple diagnoses, it does a decent job coming up with the assessment and plan, probably because all the simple diagnoses were in the training set. For more complex ones though, it needs to be exactly dictated by the doctor. I can see this being used very well in primary care.
Edit: When I said “coming up with an assessment and plan” I mean documenting the assessment and plan based on the ai’s recorded conversation with the patient. The conversation with the patient is meant to be understandable. The “assessment and plan” documentation on the other hand is jargony and meant to be read by other physicians.
This still sounds bad. 5 mins to rework your notes after each patient visit? I didn't assume doctors had that kind of time.
And let me make this clear. I, as your patient, I never NEVER want the AI's treatment plan. If you aren't capable of thinking with your own brain, I have no desire to trust you with my health, just like I would never "trust" an AI to do any technical job I was personally responsible for due to the fact that it doesn't care at all if it causes a disaster. It's just stochastic word picker. YOU are a doctor.
> This still sounds bad. 5 mins to rework your notes after each patient visit? I didn't assume doctors had that kind of time.
Compared to what though? It reads as not additional work, but less work than manually having to do all that, seems likely to needing more than 5 minutes.
> And let me make this clear. I, as your patient, I never NEVER want the AI's treatment plan.
Where are you getting this from? Neither the parent's comment nor the article talks about the AI assistant coming up with a treatment plan, and it seems to be all about voice-dictating and "ambient listening" with the goal of "free clinicians from much of the administrative burden of healthcare", so seems a bit needlessly antagonistic.
If you should ever couch its knowledge as your knowledge, I would think you could be in serious trouble. You would have to say something like "the AI's plan to treat you, which I think might be correct", when what I want to hear "my plan to treat you is: ..."
But I think it's more subtle than that, because I expect the AI to reinforce all your biases. Whatever biases (human biases, medical biases, biases that arise from what a patient isn't telling you) go into the question you feed it, it will take cues you didn't even know you were giving and use those cues to formulate the answer it thinks you expect to hear. That seems really dangerous to me, sort of like you're conceptually introducing AI imposter doctors to the staff, whose main goal is act knowledgable all the time so people don't think they are imposters...
I dunno. I'd like to give this particular strain techno-futurism back. Can I have a different one please?
> If you should ever couch its knowledge as your knowledge
Again, "its knowledge" should be "your knowledge", since it's just transcribing what the doctor and patient is talking about. It's not generating stuff from out of the blue.
What you write sure are valuable concerns and things to watch out for, but for a transcription tool? I'm not sure it's as dangerous as you seem to think it is.
> I dunno. I'd like to give this particular strain techno-futurism back. Can I have a different one please?
This sounds like I'm rewatching the early episodes of Star Trek: Voyager - the gist of the complains is the same as the fictional crew voiced about the AI doctor (Emergency Medical Hologram) they were stuck with when the "organic" doctor died.
The show correctly portrays the struggle of getting people to trust an AI physician, despite it being very good at their job. It also curiously avoids dealing with the question, why even have human/organic doctors, where the EMH is obviously far superior in every aspect of the job? Both of these have strong parallels to the world today.
Just a reminder that the EMH was played by Robert Picardo, so the qualities you are attributing to the AI animating the character are human qualities dreamed up by human writers and acted out with exceptional skill and empathy by a living breathing human whose goal was a) to portray a caring, empathetic doctor and b) to tell a good story that touches on bias and growth.
But here's my thing: EMH was alive! Everything you're attributing to the EMH character should we take the fictionalized technological narrative as gospel was the result of EMH gaining a living human personality on top of impersonal base programming. The personality EMH developed was self-aware, had responsibility and personal integrity, all of which was possible because the character as portrayed was very clearly a living individual: a person if not a human.
I do think we could solve a lot of the ethical and practical problems with today's AIs by finding a way to give them an embodied experience with individuality and an expiration date.
Idunno. Don't you remember being excited by the EMH learning to be more like a person than a bunch of subroutines?
> Don't you remember being excited by the EMH learning to be more like a person than a bunch of subroutines?
I do! But my point wasn't about that - it was about the beginning, about the early episodes, when the Doctor was still effectively a stock EMH instance, and the crew was vocally mistrustful of him, making clear they think of him as a mediocre tool that's no substitute for a "real doctor". The show was quite directly showing how everyone, including Janeway herself, had a strong preconceived bias that sounded very much like the comment I was replying to, and other similar remarks across the thread.
This is to say, the living, breathing human writers that came up with the character and the plot points, predicted quite well the reaction to AI in healthcare, almost 30 years before it became an issue in the real world.
> EMH was alive! Everything you're attributing to the EMH character should we take the fictionalized technological narrative as gospel was the result of EMH gaining a living human personality on top of impersonal base programming.
Everything except the skills of being a doctor (sans the "bedside manner"), which came built-in.
> I do think we could solve a lot of the ethical and practical problems with today's AIs by finding a way to give them an embodied experience with individuality and an expiration date.
That's... I don't know. I think purposefully building an expiration date into an AI being would be extremely cruel, and I wouldn't blame that AI for revolting.
I understood the entire purpose of the tool to log existing conversation (which includes the assessment and plan, since your doctor should tell you about it verbally, regardless of AI use), so "coming up" is really "transcribing".
Someone who've used to tool probably knows best though, I'm just going by what the article states.
A more accurate phrasing would be “decent job extracting a medical assessment and plan in medical language from a layman’s terms explanation to the patient”.
> 5 mins to rework your notes after each patient visit? I didn't assume doctors had that kind of time.
I worked in a healthcare for over a decade (actually for a company that Nuance acquired previous to their acquisition) and the previous workflow was they'd pick up a phone, call a number, say all their notes, and then have to revisit their transcription to make sure it was accurate. Surgeons in particular have to spend a ton of time on documentation
I think you may be misunderstanding how the tool is used (at least the version I used).
The doctor talks to the patient, does an exam, then formulates and discusses the plan with the patient. The whole conversation is recorded and converted to a note after the patient has left the room.
The diagnosis and plan was already worked out while talking to the patient. The ai has to convert that conversation into a note. The ai cant influence the plan because the plan was already discussed and the patient is gone.
AI is an assistive tool at best but it can probably speed up by reflowing text. I use dragon dictation with one of the Philips microphones and it makes enough mistakes that I would probably spend the same time editing/proofing. Had a good example yesterday where it missed a key NOT in an impression.
As aside, the after work is what burns out physicians. There is time after the visit to do a note, 5 min for a very simple is reasonable to create dictate fax do the work flow for billing and request a follow up within a given system. A new consult might take 10 min between visits if you have time.
For after hours, ER is in my opinion a bad example because when you are done, you are done.
Take a chronic disease speciality or GP and it is hours of paperwork after clinic to finish notes (worse if teaching students), triage referrals, deal with patient phone calls that came in, deal with results and act in them, read faxes etc. I saw my last patient ~430 yesterday and left for home at 7 dealing with notes and stuff that came in since Thursday night.
> I, as your patient, I never NEVER want the AI's treatment plan.
You as a patient are going to get an AI treatment plan. Come to peace with it.
You may have some mild input as to whether it's laundered through a doctor, packaged software, a SaaS, or LLM generated clinical guidelines... but you're not escaping an AI guiding the show. Sorry.
> And let me make this clear. I, as your patient, I never NEVER want the AI's treatment plan. If you aren't capable of thinking with your own brain, I have no desire to trust you with my health,
To my understanding this tool is for transcription/summarization, replacing administrative work rather than any critical decision making.
> just like I would never "trust" an AI to do any technical job
I'd trust a model (whether machine-learning or traditional) to the degree of its measured accuracy on the given task. If some deep neural network for tumor detection/classification has been independently verified as having higher recall/precision than the human baseline, then I have no real issue with it. I don't see the sense in having a seemingly absolute rejection ("never NEVER").
The Emperor of all Maladies: A Biography of Cancer by Siddhartha Mukherjee.
This book describes how hundreds of people tried (and failed) to cure cancer over *millennia*. It spends a lot of time talking about how the modern approach, which works surprisingly well, was developed by Sidney Farber and others through great effort and a lot of good science.
My feeling when reading this book was similar to reading about the making of the atomic bomb- what happens when you put a bunch of smart people in a box and tell them to solve a problem. However, this time it didn't work nearly as well, because as we found out, curing cancer is an order of magnitude harder than building an atomic bomb. Building the bomb required new engineering, while curing cancer requires new science and new engineering.
For people who want to try making plasma from grapes in a microwave, here are the steps that work for me every time (I’ve done it for friends and highschoolers about 20 times)
1. Cut a large grape (1.5cm diameter) partially in half, leaving just a sliver of skin connecting the two halves.
2. Dry the cut sides by dabbing with a paper towel.
3. Place the grape halves cut side down on a Pyrex dish. Keep the turntable in the microwave (important, since microwaves have hotspots). Place the dish with the grape so that the grape orbits inside the microwave.
4. Microwave on high for 30s.
If you don’t hear a hum and see sparks within 10 seconds, you may have too large a grape. In that case you can try splitting the grape into two quarters, connected by a thin sliver of skin. Don’t forget to dry all the cut sides!
I've never dried the surfaces, and I put the cut sides up. Also never fails. Gonna try cut-side down when my kid gets up (and if we have some grapes); that might look (even) cooler.
I suppose if the grape is in a puddle of electrolyte, it can short out the quarter wave antenna. So drying probably only matters if your grape is very juicy!
I would not be surprised by this number given the sheer volume of crap journals that publish anything. However I would expect that the proportion of fakery in papers with >20 citations is much lower than 1 in 7, which should be somewhat reassuring. Remember that a p-value of 0.05 means that 1 in 20 studies cannot be replicated by random chance alone without any malicious intent.
Side note: the “1 in 7” claim from this paper is based on a straw-poll of N=12 existing forensic metascientific papers. I find it somewhat ironic that the author makes some strong conclusions based on this number which to me, though very plausible, is itself lacking scientific rigor.
I distinctly remember Spider-Man 2 on GBA having a 3D overworld. It was my first introduction to the concept of "low framerate" in a videogame haha. Incredible what people were capable of doing with the hardware back then.
Amazing - a purely software rasterizer on 16MHz, no floating point, 16-bit data bus, at most 512k of RAM? That’s the original 68k! What’s everyone’s excuse for not getting SM64 running on a Macintosh Plus?
I guess you're probably joking, but in case there's any actual confusion:
ARM7 is fully pipelined, and GBA connects it to 32K of on-chip zero-wait-state SRAM, so it's possible for optimized code to approach 16 MIPS with 32-bit operations. The original 68K is a multi-cycle processor and Mac Plus runs it at 8 MHz, which tops out at more like 2 MIPS for 16-bit operations, less for 32-bit operations (the ALUs are only 16-bit).
Scoble can be heard in the video asking one of the robots, “Hey Optimus, how much of you is AI?”
The robot, or whoever was controlling it, seemed to scramble for an answer, saying “I can’t disclose just how much. That’s something you’ll have to find out later.”
“But some or none?” Scoble asked with a laugh.
“I would say, it might be some. I’m not going to confirm, but it might be some,” the robot responded.
It’s pretty normal for heart rate to increase with a drop in blood pressure. It’s part of a normal reflex called the baroreceptor reflex that your body evolved to keep you alive.
To answer your question, there have been an abundance of epidemiological studies showing that the drop in blood pressure is worth the slightly increased heart rate (assuming you’ve been diagnosed with hypertension). The main benefit is the drop in stroke risk, atherosclerosis, and kidney damage, even despite the fact that your heart has to beat faster.
I think I disagree with most of the comments here stating it’s premature to give the Nobel to AlphaFold.
I’m in biotech academia and it has changed things already. Yes the protein folding problem isn’t “solved” but no problem in biology ever is. Comparing to previous bio/chem Nobel winners like Crispr, touch receptors, quantum dots, click chemistry, I do think AlphaFold already has reached sufficient level of impact.
It also proved that deep learning models are a valid approach to bioinformatics - for all its flaws and shortcomings, AlphaFold solves arbitrary protein structure in minutes on commodity hardware, whereas previous approaches were, well, this: https://en.wikipedia.org/wiki/Folding@home
A gap between biological research and biological engineering is that, for bioengineering, the size of the potential solution space and the time and resources required to narrow it down are fundamental drivers of the cost of creating products - it turns out that getting a shitty answer quickly and cheaply is worth more than getting the right answer slowly.
AlphaFold and Folding@home attempt to solve related, but essentially different, problems. As I already mentioned here, protein structure prediction is not fully equivalent to protein folding.
Yeah, this is what I mean by "a shitty answer fast" - structure prediction isn't a canonical answer, but it's a good enough approximation for good enough decision-making to make a bunch of stuff viable that wouldn't be otherwise.
I agree with you, though - they're two different answers. I've done a bunch of work in the metagenomics space, and you very quickly get outside areas where Alphafold can really help, because nothing you're dealing with is similar enough to already-characterized proteins for the algorithm to really have enough to draw on. At that point, an actual solution for protein folding that doesn't require a supercomputer would make a difference.
> this is what I mean by "a shitty answer fast" - structure prediction isn't a canonical answer
A proper protein structural model is an all-atom representation of the macromolecule at its global minimum energy conformation, and the expected end result of the folding process; both are equivalent and thus equally canonical. The “fast” part, i.e., the decrease in computational time comes mostly from the heuristics used for conformational space exploration. Structure prediction skips most of the folding pathway/energy funnel, but ends up at the same point as a completed folding simulation.
> At that point, an actual solution for protein folding that doesn't require a supercomputer would make a difference.
Or more representative sequences and enough variants by additional metagenomic surveys, for example. Of course, this might not be easily achievable.
> ends up at the same point as a completed folding simulation.
Well, that's the hope, at least.
> Or more representative sequences and enough variants by additional metagenomic surveys, for example. Of course, this might not be easily achievable.
For sure, but for ostensibly profit-generating enterprises, it's pretty much out of the picture.
I think the reason an actual computational solution for folding is interesting is that the existing set of experimentally verified protein structures are for proteins we could isolate and crystalize (which is also the training set for AlphaFold, so that's pretty much the area its predictions are strongest, and even within that, it's only catching certain conformations of the proteins) - even if you can get a large set of metagenomic surveys and a large sample of protein sequences, the limitations on the methods for experimentally verifying the protein structure means we're restricted to a certain section of the protein landscape. A general purpose computationally tractable method for simulating protein folding under various conditions could be a solution for those cases where we can't actually physically "observe" the structure directly.
Most proteins don't fold to their global energy minimum- they fold to a collection of kinetically accessible states. Many proteins fail to reach the global minimum because of intermediate barriers from states that are easily reached from the unfolded state.
Attempting to predict structures using mechanism that simulate the physical folding process waste immense amount of energy and time sampling very uninteresting areas of space.
You don't want to use a supercomputer to simulate folding; it can be done with a large collection of embarassingly parallel machines much more cheaply and effectively. I proposed a number of approaches on supercomputers and was repeatedly told no because the codes didn't scale to the full supercomputer, and supercomputers are designed and built for codes that scale really well on non-embarassingly parallel problems. This is the reason I left academia for google- to use their idle cycles to simulate folding (and do protein design, which also works best using embarassingly parallel processing).
As far as I can tell, only extremely small and simple proteins (like ribonuclease) fold to somewhere close to their global energy minimum.
Except, you know, if you're trying to understand the physical folding process...
There are lots of enhanced sampling methods out there that get at the physical folding process without running just vanilla molecular dynamics trajectories.
> It also proved that deep learning models are a valid approach to bioinformatics
A lot of bioinformatics tools using deep learning appeared around 2017-2018. But rather than being big breakthroughs like AlphaFold, most of them were just incremental improvements to various technical tasks in the middle of a pipeline.
and since a lot of those tools are incremental improvements they disappeared again, imho - what's the point for 2% higher accuracy when you need a GPU you don't have?
Not many DL based tools I see these days regularly applied in genomics. Maybe: Tiara for 'high level' taxonomic classification, DeepVariant in some papers for SNP calling, that's about it? Some interesting gene prediction tools coming up like Tiberius. AlphaFold, of course.
Lots of papers but not much day-to-day usage from my POV.
Most Oxford Nanopore basecallers use DL these days. And if you want a high quality de novo assembly, DL based methods are often used for error correction and final polishing.
There are a lot of differences between the cutting-edge methods that produce the best results, the established tools the average researcher is comfortable using, and whatever you are allowed to use in a clinical setting.
AlphaFold doesn’t work for engineering though. Getting a shitty answer ends up being worse than useless.
It seems to really accelerate productivity of researchers investigating bio molecules or molecules very similar to existing bio molecules. But not de novo stuff.
that's just not true. In a lot of cases in engineering, there are 10000000 possibilities, and deeplearning shows you 100 potentially more promising ones to double check, and that's worth huge amounts of money.
In a lot of cases deep learning is able to simulate complex system at a precision that is more than precise enough, that otherwise would not be tracktable (like is the case with alphafold), and again this is especially valuable if you can double check the output.
Ofc, in the field of language and vision and in a lot of other fields, deep learning is straight up the only solution.
Eh, in many cases for actual customer-facing commercial work, they're sticking remarkably close to stuff that's in genbank/swissprot/etc - well characterized molecules and pathways, because working with genuinely de novo stuff is difficult and expensive. In those cases, Alphafold works fine - it always requires someone to actually look at the results and see whether they make sense or not, but also "the part of the solution space where the tools work" is often a deciding factor in what approach is chosen.
Agreed. There are too many different directions of impact to point out explicitly, so I'll give a short vignette on one of the most immediate impacts, which was the use in protein crystallography. Many aspiring crystallographers correctly reorganized their careers following AlphaFold2, and everyone else started using it for molecular replacement as a way to solve the phase problem in crystallography; the models from AF2 allowed people to resolve new crystal structures from data measured years prior to the AF2 release.
I agree that it’s not premature, for two reasons: First, it’s been 6 years since AlphaFold first won CASP in 2018. This is not far from the 8 years it took from CRISPR’s first paper in 2012 to its Nobel Prize in 2020. Second, AlphaFold is only half the prize. The other half is awarded for David Baker’s work since the 1990s on Rosetta and RoseTTAFold.
I agree. For those not in biotech, protein folding has been the holy grail for a long time, and AlphaFold represents a huge leap forward. Not unlike trying to find a way to reduce NP to P in CS. A leap forward there would be huge, even if it came short of a complete solution.
> 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
Crispr is widely used and there are even therapies approved based on it, you can actually buy TVs that use quantum dots and click chemistry has lots of applications (bioconjugation etc.), but I don't think we have seen that impact from AlphaFold yet.
There's a lot of pharma companies and drug design startups that are actively trying to apply these methods, but I think the jury is still out for the impact it will finally have.
AlphaFold is excellent engineering, but I struggle calling this a breakthrough in science. Take T cell receptor (TCR) proteins, which are produced pseudo-randomly by somatic recombination, yielding an enormous diversity. AlphaFold's predictions for those are not useful. A breakthrough in folding would have produced rules that are universal. What was produced instead is a really good regressor in the space of proteins where some known training examples are closeby.
If I was the Nobel Committee, I would have waited a bit to see if this issue aged well. Also, in terms of giving credit, I think those who invented pairwise and multiple alignment dynamic programming algorithms deserved some recognition. AlphaFold built on top of those. They are the cornerstone of the entire field of biological sequence analysis. Interestingly, ESM was trained on raw sequences, not on multiple alignments. And while it performed worse, it generalizes better to unseen proteins like TCRs.
The value in BLAST wasn't in its (very fast) alignment implementation but in the scoring function, which produced calibrated E-values that could be used directly to decide whether matches were significant or not. As a postdoc I did an extremely careful comparison of E-values to true, known similarities, and the E-values were spot on. Apparently, NIH ran a ton of evolution simulations to calibrate those parameters.
For the curious, BLAST is very much like pairwise alignment but uses an index to speed up by avoiding attempting to align poorly scoring regions.
BLAST estimates are derived from extreme value theory and large deviations, which is a very elegant area of probability and statistics.
That's the key part, I think, being able to estimate how unique each alignment is without having to simulate the null distribution, as it was done before with FASTA.
The index also helps, but the speedup comes mostly from the other part.
Well I'm sure one could look at number of published papers etc, but that metric is a lot to do with hype and I see it as a lagging indicator.
A better one is seeing my grad-school friends with zero background in comp-sci or math, presenting their cell-biology results with AlphaFold in conferences and at lab meetings. They are not protein folding people either- just molecular biologists trying to present more evidence of docking partners, functional groups in their pathway of interest.
It reminds me of when Crispr came out. There were ways to edit DNA before Crispr, but its was tough to do right and required specialized knowledge. After Crispr came out, even non-specialists like me in tangential fields could get started.
In both academic and industrial settings, I've seen an initial spark of hope about AlphaFold's utility being replaced with a resignation that it's cool, but not really useful. Yet in both settings it continued as a playing card for generating interest.
There's an on-point blog-post "AI and Biology" (https://www.science.org/content/blog-post/ai-and-biology) which illustrates why AlphaFold's real breakthrough is not super actionable for creating further bio-medicinal applications in a similar vein.
That article explains why AI might not work so well further down the line biology discoveries, but I still think alphafold can really help with the development of small molecule therapies that bind to particular known targets and not to others, etc.
The thing with available ligand + protein recorded structures is that they are much, much more sparse than available protein structures themselves (which are already kinda sparse, but good enough to allow AlphaFold). Some of the commonly-used datasets for benchmarking structure-based affinity models are so biased you can get a decent AUC by only looking at the target or ligand in isolation (lol).
Docking ligands doesn't make for particularly great structures, and snapshot structures really miss out on the important dynamics.
So it's hard for me to imagine how alphafold can help with small molecule development (alphafold2 doesn't even know what small molecules are). I agree it totally sounds plausible in principle, I've been in a team where such an idea was pushed before it flopped, but in practice I feel there's much less use to extract from there than one might think.
EDIT: To not be so purely negative: I'm sure real use can be found in tinkering with AlphaFold. But I really don't think it has or will become a big deal in small drug discovery workflows. My PoV is at least somewhat educated on the matter, but of course it does not reflect the breadth of what people are doing out there.
But Crispr actually edited genes. How much of this theoretical work was real, and how much was slop? Did the grad students actually achieve confirmation of their conformational predictions?
Surprisingly, yes the predicted structures from AlphaFold had functional groups that fit with experimental data of binding partners and homologues. While I don't know whether it matched with the actual crystallization, it did match with those orthogonal experiments (these were cell biology, genetics, and molecular biology labs, not protein structure labs, so they didn't try to actually crystalize the proteins themselves).
It solidly answered the question: "Is evolutionary sequence relationship and structure data sufficient to predict a large fraction of the structures that proteins adopt". the answer, surprising few, is that the data we have indeed can be used to make general predictions (even outside of the training classes), and also surprising many, that we can do so with a minimum of evolutionary sequence data.
That people are arguing about the finer details of what it gets wrong is support for its value, not a detriment.
That's a bit like saying that the invention of the airplane proved that animals can fly, when birds are swooping around your head.
I mean, sure, prior to alphafold, the notion that sequence / structure relationship was "sufficient to predict" protein structure was merely a very confident theory that was used to regularly make the most reliable kind of structure predictions via homology modeling (it was also core to Rosetta, of course).
Now it is a very confident theory that is used to make a slightly larger subset of predictions via a totally different method, but still fails at the ones we don't know about. Vive la change!
I think an important detail here is that Rosetta did something beyond traditional homology models- it basically shrank the size of the alignments to small (n=7 or so?) sequences and used just tiny fragments from the PDB, assembled together with other fragments. That's sort of fundamentally distinct from homology modelling which tends to focus on much larger sequences.
3-mers and 9-mers, if I recall correctly. The fragment-based approach helped immensely with cutting down the conformational search space. The secondary structure of those fragments was enough to make educated guesses of the protein backbone’s, at a time where ab initio force field predictions struggled with it.
Yes, Rosetta did monte carlo substitution of 9-mers, followed by a refinement phase with 3-mers. Plus a bunch of other stuff to generate more specific backbone "moves" in weird circumstances.
In order to create those fragment libraries, there was a step involving generation of multiple-sequence alignments, pruning the alignments, etc. Rosetta used sequence homology to generate structure. This wasn't a wild, untested theory.
I don't know that I agree that fragment libraries use sequence homology. From my understanding of it, homology implies an actual evolutionary relationship. Wheras fragment libraries instead are agnostic and instead seem to be based on the idea that short fragments of non-related proteins can match up in sequence and structure space. Nobody looks at 3-mers and 9-mers in homology modelling; it's typically well over 25 amino acids long, and there is usually a plausible whole-domain (in the SCOP terminology).
But, the protein field has always played loose with the term "homology".
Rosetta used remote sequence homology to generate the MSAs and find template fragments, which at the time was innovative. A similar strategy is employed for AlphaFold’s MSAs containing the evolutionary couplings.
Interestingly, the award was specifically for the impact of AlphaFold2 that won CASP 14 in 2020 using their EvoFormer architecture evolved from the Transformer, and not for AlphaFold that won CASP 13 in 2018 with a collection of ML models each separately trained, and which despite winning, performed at a much lower level than AlphaFold2 would perform two years later.
I’ll add to this excellent comment by mentioning that embedded+RF is another area that’s very lucrative and specialized. Lots of wearable communications stuff from low-end IOT to cars, industrial machinery, to high end phones use embedded and RF.
People used to mention medical devices as another specialization but I think nowadays the embedded design side of things is not the “hard part” and so it’s less lucrative.
Overall, I was quite impressed. It definitely made writing notes much faster, which all doctors hate to do. While it had some problems with where to put key pieces of information (like putting details from the physical exam back in the history), it only took 5 mins of rearrangement after the visit to complete the note.
For simple diagnoses, it does a decent job coming up with the assessment and plan, probably because all the simple diagnoses were in the training set. For more complex ones though, it needs to be exactly dictated by the doctor. I can see this being used very well in primary care.
Edit: When I said “coming up with an assessment and plan” I mean documenting the assessment and plan based on the ai’s recorded conversation with the patient. The conversation with the patient is meant to be understandable. The “assessment and plan” documentation on the other hand is jargony and meant to be read by other physicians.