As if sequencing cost were the bottleneck.. looking at you UPS/DHL/FedEx delivery drivers who keep losing samples at least once a month

Emails that don't pass my fool proof (enough) sanity test don't get an answer. It's nearly always exactly the AI written ones

bsky.app/profile/maxf...

I set myself a reminder to check this 1 year later & turns out I was mostly wrong🥲
Here search results from on "DNA", last year (orange) vs now (blue) and the indexing issue is only noticeable for the current year ('24), barely an increase for '23
New hypothesis: all (bio?) topics peaked around '21?

wouldn't it make much more sense to plot this relative to normal distribution found across the proteome?
Too lazy to extract the numbers and make a scatter plot, but I think Ala and Glu are at least as off the axis as the aromatics
www.uniprot.org/uniprotkb/st...

Hi Maxim, thanks. I see that only changes were newly predicted and these were indeed made with AF2 again, not AF3.
However, not clear to me yet if the provided MSAs (for unchanged entries) are the originals from 2021ish or if you recomputed

Pretty vague, given the bombast. Since the only numbers I see there aren't apples to apples, can you contextualise for us whats now the actual change in $/genome (or some other quantitative metric)?

Not sure the backwards reasoning here is apt? IMO BindCraft does stand out *because* of large number of tests (12 protein targets by my count). They may still have been unusually lucky / biased in target selection, but maybe it isn't the most obvious example to illustrate ur very valid overall point

also, this editorial's answer to the question is yes (in violation of Betteridge's law), recommending to just don't produce AI slob that doesn't add anything

can't find the article now, but I saw a very useful graph once w/ # of articles vs # of reviews in a field:
-many articles, few reviews: write review
-few articles, many reviews: write review of reviews
-many articles, many reviews: write specialized review
-few articles, few reviews: write proposal

Maxim could you kindly clarify: did you actually recompute all the structures, or are the models identical to the previous AFDB version? And are the MSAs the ones from back in day, or do these reflect more recent db searches? If new predictions were made - with which AF version?

Great post! In agree that complexity is underestimated, lack of knowledge the bottleneck to design, & intuition key to success. Likely, AI can work work around some, but not all of the issues without explicit data.
Do you mind sharing what pockets you think work better than others with bindcraft?