Nick Semenkovich, MD PhD

Dad⁴. Founding member of @MedicalCollege Data Science Institute, focused on liquid biopsies / AI x Bio. Via @MIT ⨉ @WashU ⨉ @MassGenBrigham

[Nick Semenkovich, MD PhD @semenko]

Great to see a ESMFold2 - and with a permissive MIT license!

[Roshan Rao @proteinrosh]

Announcing ESMFold2, our new state-of-the-art structure prediction model capable of predicting structure from single sequences or MSAs. ESMFold2 improves on benchmarks of protein-protein interaction and is particularly strong on predictions of antibody-antigen complexes.

[Diego del Alamo @DdelAlamo]

Two labs independently and simultaneously found that one specific head in the first layer of ESM3 does most of the work unpacking structural input tokens (link in the QT)

[Nick Semenkovich, MD PhD @semenko]

@aaronmring @cremieuxrecueil 100% agree. Outside of HeFH, this has always felt like an odd target population to de-risk ABEs. Little opportunity to expand to primary prevention and I'd bet little MACE benefit over inclisiran (or oral enlicitide).

[Nick Semenkovich, MD PhD @semenko]

@MariosGeorgakis @CasualVariant 100% agree. Very hard to justify undefined base editing risk when so many alternative therapies with established safety profiles exist. (Especially hard with PCSK9 orals so close!)

[Marios Georgakis @MariosGeorgakis]

The two most interesting findings of this paper:

1⃣In observational analyses, there were 52,887 significant protein-disease associations!

It's of course implausible that all those represent links that in any way point to causal biology.

Only 0.06% (!) of them (n = 33) had directionally concordant, high-confidence support from genetic analysis (cis-Mendelian randomization+colocalization)

It just highlights how ridiculous it was to rely on this kind of analyses for drug target discovery/validation in the past.

[Nick Semenkovich, MD PhD @semenko]

Pre-auth remote heap buffer overflow in dnsmasq is wild: https://t.co/Or9XawXLfD

[Nick Semenkovich, MD PhD @semenko]

Very nice evaluation of key structure prediction tasks!

[Eva Smorodina @EvaSmorodina]

We examined how well AI-based antibody-antigen structure prediction pipelines perform at the practical discovery-critical task of differentiating true binders from plausible but incorrect ones. Short answer: not very well.

The paper: https://t.co/ZKYhAFykOl

[Leo Wan @LeoWanPhD]

Antibody design is stuck on a data problem, not a model problem. The PDB holds 250,000 structures. Only 811 are unique VHH-antigen complexes. This is the training set the entire field is competing over, and it grows slowly.

[Nick Semenkovich, MD PhD @semenko]

@tallphil Love this!! Would love to get involved. Played with something similar first in python with RSeQC: https://t.co/0uACSKyx25 High test coverage also exposes some cool edge case bugs like a longstanding error in soft clipping: https://t.co/b3Iu5GLNij

[Nick Semenkovich, MD PhD @semenko]

And just like that, an army of "health ai" startups that were just LLM wrappers died.

[Epic @HeyEpic]

What if you could talk directly to your patient's chart?

Meet Ask Art.

At @TGHCares, clinicians are asking questions about their patients' records and getting cited answers, anchored in the patient's data.

[Charlie Petty @incredutility]

The Gates Foundation could liquidate and fund ~18 months of the NIH budget

[Nick Semenkovich, MD PhD @semenko]

@neoavatara No, times have changed. Step 2 scores are now a huge signal in resident selection.

[Nick Semenkovich, MD PhD @semenko]

@francisdeng @mahesh_shenai @sfgreen16521 Those industries are all very different. Uber and Lyft exploit unskilled workers. NFL has a players union with actual advocacy for salary and benefits.

[Nick Semenkovich, MD PhD @semenko]

@ciirarref Residents should make more, too.

[Nick Semenkovich, MD PhD @semenko]

This makes perfect mathematical sense and is an entire field of study: queueing theory.

[Dilan Esper @dilanesper]

this makes no mathematical sense. it's the same number of people

[Nick Semenkovich, MD PhD @semenko]

@JohnDCook John no not you.

[Nick Semenkovich, MD PhD @semenko]

@iskander Ok but is the giraffe saliva personalized by AI?

[Nick Semenkovich, MD PhD @semenko]

@PradyuPrasad There isn't. People get compassionate use / early experimental / phase 1 drugs all the time.

[Nick Semenkovich, MD PhD @semenko]

I'm very sympathetic to that and deal with it sometimes. I encourage those to join a trial when possible (it frequently is!) / compassionate use of early stage drugs. Many folks are also tragically exploited and spend their last days (and dollars) on snake oil. (Even Steve Jobs did!)

[Nick Semenkovich, MD PhD @semenko]

This n=1 anecdote is almost certainly placebo / adjuvant effect (would love the actual data!). The article's screenshot shows c-kit's sequence, which is exceptionally well studied and profiled for neoantigens (e.g. in GIST). Presumably the compassionate use denial was for a standard TKI?

[vittorio @IterIntellectus]

this is actually insane

> be tech guy in australia
> adopt cancer riddled rescue dog, months to live
> not_going_to_give_you_up.mp4
> pay $3,000 to sequence her tumor DNA
> feed it to ChatGPT and AlphaFold
> zero background in biology
> identify mutated proteins, match them to drug targets
> design a custom mRNA cancer vaccine from scratch
> genomics professor is “gobsmacked” that some puppy lover did this on his own
> need ethics approval to administer it
> red tape takes longer than designing the vaccine
> 3 months, finally approved
> drive 10 hours to get rosie her first injection
> tumor halves
> coat gets glossy again
> dog is alive and happy
> professor: “if we can do this for a dog, why aren’t we rolling this out to humans?”

one man with a chatbot, and $3,000 just outperformed the entire pharmaceutical discovery pipeline.

we are going to cure so many diseases.
I dont think people realize how good things are going to get