Howard Salis

NEW PAPER led by @grace_vezeau! We rationally engineered synthetic riboswitches to carry out sense & respond for human biomarker proteins [mCRP and IL32-gamma] that indicate disease or infection. 16-fold activation without trial-and-error! nature.com/articles/s4146…

@SynBio1 and 3) developing countermeasures (vaccines, treatments) so fast that any threat is mitigated before it becomes a national security issue. That removes state-sponsored attacks as existential threats. Bioterror is no longer terror-inducing.

Tough question. My perspective is that locking down information isn't going to work The biosecurity police don't want to be the enemies of information because that will put them into an adversarial position with scientists and erode the trust they need to do their jobs Biosecurity should be 10% worrying about infohazards. 90% of the focus should be on 1) The physical layer (keeping bad actors away from specialized equipment and facilities) 2) The trust layer (building a community of scientists who care and watch out for each other)

Dank memes for biosecurity red teams

@The__Taybor As a long time nanopore user, is it possible to extract the DNA that has passed through pores, using adaptive sampling to selectively pass or flick based on sequence correctness? Would be a *great way* of combining DNA assembly & sequencing in one integrated workflow...

Good to see Adaptive Sampling working out. I originally called it ‘Read until’ which is not very good, Ewan Birney coined the term Adaptive Sampling. It’s an old idea, took a long time to come to fruition.

@SynBio1 Specifically: salislab.net/software/build… I wrestle with this question.

@SynBio1 What are your thoughts on a lab open-sourcing another oligopool-based DNA assembly method? Option A: open-source it because the genie is out of that bottle already Option B: don't open-source it because improved algorithms solving this problem will add more biosecurity risk

I want my super intelligent AGI to be like Vic Fontaine (Deep Space 9). An emotive superintelligent hologram, but no need to lord it over the corporeals.

At @plasmidsaurus, we discovered our $50, 3 day TAT RNA-seq service can also detect mycoplasma contamination from several prominent species, and we've already alerted over a hundred labs to likely contamination. We're working on adding this to the service for free very soon.

@ProfTomEllis @SynBio1 A research org that only expects to exist for <10 years will absolutely cut all sorts of corners because why not?

@SynBio1 Plus universities can offer PhDs, and right or wrong this allows them to do world class research without having to pay world class salaries. So it’s another reason why universities *should* be the best value place to do most research.

We're seeing the "unbundling" of the research and teaching functions of the modern university More federal money moving to institutes that don't teach (and thus have lower overheads) I'm not opposed to experiments like this per se. Would probably be better to have them run by an admin that doesn't hate science and scientists

@ATinyGreenCell What's your typical LLM workflow to accelerate writing tests & harnesses?

I am become auditor. 2400+ tests of every function of this 57,000 line codebase and finally pinging back no bugs found. Now to add just a few more tweaks to the UI and make some parts modular for later expansion by user request and we good to ship v1.0.0 next week!

@SynBio1 This is what happens when GPT plays the Reverse UNO card on your image request.

@SynBio1 My cat *loves* whipped cream from the can, but not if you whip it yourself w/o the sugar. It could be the texture, or perhaps he’s a mutant.

My new startup is going to genetically engineer cats to restore the sweet taste receptor because then you could give your cat a blueberry sometimes and they'd be like yum pmc.ncbi.nlm.nih.gov/articles/PMC20…

@rlacombe @MoAlQuraishi @gembioworkshop @iclrconf You're thinking that more states = more complexity, but it's actually not that difficult to compute. We're developing whole-cell models & solving grand canonical ensembles involving 100s of DNA binding factors to calc binding probabilities. Much easier @ isotherm + isobaric!

@hsalis @MoAlQuraishi @gembioworkshop @iclrconf But orders of magnitude less complexity. And that’s not even counting PPI and binding affinity interactions that don’t break bonds.

“In the long run, interpretability is hopeless. Physics being reducible to equations was an exception, there is no reason for the world to be comprehensible to our puny brains.” – @MoAlQuraishi #ICLR2024

@rlacombe @MoAlQuraishi @gembioworkshop @iclrconf There are more chemical reactions taking place inside a refinery than in the human body.

@hsalis @MoAlQuraishi @gembioworkshop @iclrconf Simply that a refinery is not a good example of irreducible complexity.

@rlacombe @MoAlQuraishi @gembioworkshop @iclrconf Then by induction, biological systems have state functions. What's your point?

@hsalis @MoAlQuraishi @gembioworkshop @iclrconf Chemical systems have state functions.

Today we're publishing our index of unmet needs in human disease: 2443 indications scored and ranked on burden of disease, prevalence, pipeline activity, and treatment burden. We hope this will help drug developers identify overlooked medical problems convoke.bio/blog/introduci…

@Avi_Marcus @ATinyGreenCell @_inc0_ Nitrogenase uses ATP and electrons to catalyze its reaction. And they are high potential electrons that require special carriers (ferredoxins).

@ATinyGreenCell @_inc0_ I have to admit that I am more intrigued by an enzyme being able to absorb 9000 photons a second productively. How doesn’t it denature???

"That which is asserted without evidence can be dismissed without evidence." -Hitchens' Razor

@beaverd the funny thing is that nitrogenase isn't just a single enzyme and the rate-limiting step is not N2 binding so ....

I invented a nitrogen fixation enzyme that works on sunlight and makes Haber-Bosch obsolete k_cat at 1 atm: ~9,000 N2/site/sec I have no idea how to sell it so I guess i'll just open source it and forgo the nobel prize?

@anshulkundaje IMO it's herd mentality to train a transformer or two with genome sequence info. It's not a novel concept anymore, it just requires compute. I'd love to see AI people take a real big swing and develop better algorithms for training physics-informed NNs. Weights could have meaning

@hsalis My honest opinion is that it is great that we have unconventional institutes like Arc that can allow scientists to take risks & do things that the majority might consider crazy. That is exactly what they shud do. My primary issue is with the way results are communicated.

"The Bitter Lesson has fully arrived in sequence biology and protein structure. Evo 2, AlphaFold 2 and 3, ProGen3, RFdiffusion". This sentence has some issues IMO. 1/

@anshulkundaje This is all a product of some harsh facts. Arc is not subject to the same scientific standards as the rest of us. If they wrote proposals, they would be read by AI-agnostic scientists who would judge them harshly. What do we learn? What is it for? How can it be used? Why this?

@anshulkundaje and [2] Evo2 doesn't carry out common/desired tasks. You can't ask it to generate the DNA sequence of a promoter that binds to TF1, TF2, ... with a desired regulatory function. Or a mRNA sequence that binds to the ribosome in a desired way. Just generates "nature-like" sequences.