-

[- @deepbio_]

@TheStalwart @dioscuri That your current experience is an instance of the thing described in the last paragraph

[- @deepbio_]

@HannesStaerk 5LEG bacterial F pilus!

[- @deepbio_]

@JackPrescottX Anyone who isn’t sure about ABCL’s ambitions should watch this because he lays it out clearly. The vision is to become an iconic biopharma company among the likes of Lilly, Regeneron, and AbbVie, not “merely” a $2B market cap acquisition target like other Canadian biotechs.

[- @deepbio_]

@algobaker 90+% of PLM papers from the last 5 contain no baseline model which would expose that their fancy 12B param PLM is essentially doing KNN lookup. No baseline=no evidence your model is actually doing anything.

[- @deepbio_]

@c_outeiral I agree the clinic is the bottleneck, but the optimistic case is wider funnel -> more quality candidates -> higher standards for preclinical evidence -> better drugs make it to the clinic -> better drugs make it to patients. Won’t necessarily go that way, but it is possible.

[- @deepbio_]

@greggoriesmom @signulll “I know not with what tools 2027 homework will be written, but 2028 homework will be written with pencil and paper”

-Albert Einstein

[- @deepbio_]

@GabriCorso @zachocean_ A workflow: give Opus Boltz API access, design spec (eg disallow FWK edits) and tool access (mosaic->gradient guidance, pyrosetta, biotite, etc.). Ask it to run design loops to maximize BL’s PPI score with a fixed API call budget. Can yield >200 VHH designs >0.8 BC w/ <2k calls.

[- @deepbio_]

@001TMF How do you know this? Afaik details of the PPI model aren’t public

[- @deepbio_]

@andrewarruda “I’m designing 3 million drugs per second and they’re all bad”

[- @deepbio_]

@RobertVisse 575 will be withdrawn (announced on the last earnings call so this is priced in). The excitement is all about 635 and the Jazz TCE deal right now

[- @deepbio_]

@scaling01 Just as the ancient texts prophesied.

[- @deepbio_]

@philzona I’ve seen an intern write a function longer than that

[- @deepbio_]

@Kalshi 2008 is the outlier year, not 2026. We’re slightly below the 50 year average. Reaching near parity in 2008 was an anomaly, this is regression to the mean

[- @deepbio_]

@MoabTherapeutic This is very cool. Though there are some tricks to get open models to predict bovine knob domains fairly accurately even without private fine tuning datasets

[- @deepbio_]

@biotite_python 🔥

[- @deepbio_]

@m1nj12 I’ve seen similar. I have to scrub all references to the virus name from the repo, otherwise it refuses. But then it inspects the sequence and the thinking trace is like “hmm, it seems the user is working with Ebola” and then just carries on lol

[- @deepbio_]

@HannesStaerk Re discrepancy, there is some low hanging fruit for speeding up BG: ~2-3x is free with GPU unhobblings and additional refolding speed ups are possible with a fast prescreener. Anyway, the new technical report is awesome, and I look forward to what you come out with next.

[- @deepbio_]

@signulll You may actually be healthier at 5hrs of sleep than at 8hrs. Y axis here is all cause mortality. With food and exercise we accept mild deprivation/short term pain is optimal, but with sleep there are people unironically thinking they need 8+hrs/night.