Heck yes. Music pirates have done a whole lot to help preserve hip-hop history by ripping/archiving countless rare/underground 12" and cassettes that are definitely not available on streaming platforms; we're talking alternative versions of songs like clean edits, remixes, acapellas, instrumentals which supported hip-hop producers and DJs in the early days.
This has been a previous area of research for Google (https://ai.googleblog.com/2017/04/predicting-properties-of-m...). It remains routine to benchmark GNNs and other molecular machine learning models on predicting quantum mechanical properties including energies (which speed up geometry optimization)
I agree with the sentiment of this paper (AF can enable drug discovery), but in this specific instance, the authors had a real opportunity contribute a general finding to the scientific community but instead they put in the lowest amount of effort (to a point where they're almost saying nothing at all).
The target had dozens of related structures in the protein databank, including relatives with ~40% sequence identity. This target family has a very similar structure, and conserved active site residues. It's relevant that this target has approved cross-CDK family inhibitors (and thousands of data points of CDK family binders on ChEMBL). The conventional way to enable structure-based design is to build a homology model using a similar structure (see here: https://swissmodel.expasy.org/repository/uniprot/Q8IZL9?temp...), and in this case, there is very low deviation from the AF2 model and this "old fashioned" approach.
To recap, this target had a decent model that would have likely sufficed for drug discovery. The community already knows that "homology models" can be used for structure-based drug design, so any methodological hypotheses of this paper are not supported by evidence.
Although I agree the authors could have done homology modelling, in this case AlphaFold is already doing that. It knows all the related sequences (through the sequence database similarity graph that it embeds) and has a very sophisticated modelling system. In my guess (I'd have to check with my old friends to be sure) it does as well as if not better in producing atomic accuracy for structural predictions for homology modellers better than a typical modeller could produce.
This paper is mainly a flag planted so they can claim they landed on mars first and fastest.
There a quite a few things missing from this paper that would make it a good a good drug discovery paper.
First, they didn't discover a drug - they found a hit. 30 days from target to hit using conventional high-throughput biochemical screening would take 2-4 months. So, this is 3x faster, but that's not the rate limiting step. Validation and in vivo studies will take >4 mo and 1-12mo respectively.
Second, if we take this as a "we found a hit" paper, I want to know how specific your hit is. This would be one of the major advantages of using AF2 - screen against related proteins with some structural or functional similarity. This is the time intensive and oft overlooked part of good in vitro screening campaigns. Potency is nice (although 9 μM isn't impressive), but ultimately selectivity is paramount when targeting a class of proteins with well conserved binding sites, like kinases. If they found a promiscuous CDK inhibitor that happens to hit CDK20, then I bet there are tons of previously reported promiscuous CDK inhibitors that will hit CDK20 too.
Third, this paper is surprising because it exploits none of the cool new things AF2 could enable. In addition to what you mention above, the authors could have tried to counter screen (much faster in silico!), find an allosteric inhibitor, identify a PPI/complex inhibitor, or take a leap by generating a SAR series in silico and validating a few selected compounds in vitro.
Overall, this paper seems both incremental and misdirected. Saving 2 months in the discovery phase, pre-IP, is worth ~0. Not sure anyone there has much experience developing drugs. Hits are nice, but rarely the hard part. However, a hit on a protein from a structurally divergent class would be a major accomplishment.
another important point to notice is affinity. While 8 uM looks impressive, it is not that hard to develop such potency since compounds are likely to aim ATP binding pocket. It is big, deep and offers many hydrogen bond donors in hindge region. What important for such compounds is selectivity, since you want to inhibit only specific kinase, not all of them.
For me it looks like advertising of their platform, not actual scientific achievement.
And this, dear HN community, is the difference between an expert reading a paper pertaining to their field and the casual reader, or even scientists in unrelated fields reading this paper. I am just not equipped to judge the quality of research in the field.
it's not really a generalizable solution, though, because suppose you need an unnatural AA that needs a specialized ribosomal cavity and that cavity hasn't been developed yet. And you need the material by a year's time. You are not going to evolve a ribosome to make your peptide. And, there are post-translational modifications that are too big to fit into ANY cavity, though in theory you could use tricksy chemistry to figure out a downstream processing step, that chemistry may not exist yet, or it may require a pure enzyme that is a pain in the ass to obtain, etc. etc. Finally, charging the correct tRNA is not necessarily easy or efficient.
Unlike well-designed programming architectures, biology is usually not "trivially composable". Oddly, SPPS is. Yes, there are corner cases, but those are more the exception than the rule.
Our startup routinely orders the synthesis of hundreds of peptides for technology validation and drug discovery research (not suitable for human consumption). Costs for a small quantity through a contract research organizion can be $200-$2000 USD per peptide depending on desired purity, length, and chemical complexity. For some applications peptide arrays are suitable, and can drive the costs down to $10 USD per peptide or lower. In both cases, the turnaround time is 4-6 weeks, even though a peptide chemist could do the job in about half the time for a rush order.
One of the reasons we don't have them all is that individual genes can encode for multiple protein isoforms through alternative splicing. AlphaFold was only run on one. Otherwise, there's lots of important biochemical/biophysical processes that impact structure, as cells are only about 50% protein by weight.
Just in case you're not joking, it's worth noting that the majority of distributed molecular simulation (past and present) is spent studying "folded proteins" to discover structures of proteins that are often hidden from methods like AlphaFold (currently). For example, https://www.nature.com/articles/s41557-021-00707-0
Everything between the BRCT and RING domains of BRCA1 is an intrinsically unstructured region which DeepMind correctly predicts, https://pubmed.ncbi.nlm.nih.gov/15571721/
Another famous one would be R-___domain of CFTR, which was not resolved in experimental structure determination, and AlphaFold models correctly show disorder there. Nothing to be done in those cases except perform molecular simulation or other experiments to assess dynamic ensembles, https://alphafold.ebi.ac.uk/entry/P13569
Yet, there were still 136 human teams who competed in CASP14 (https://predictioncenter.org/casp14/docs.cgi?view=groupsbyna...), including DeepMind. Even if a significant fraction of these projects were done piggy-backing another grant, this work does receive research funding.
For example, check the posts in the early 2010's of sites like this, https://hiphop-thegoldenera.blogspot.com/, of course all the links are broken now.