Sign in

John Parkhill

@johnparkhill.bsky.social
982 followers 384 following 173 posts

Senior director of machine learning @ Terray Therapeutics (www.terraytx.com). Father. NSF CAREER award giver-upper. Gibe and gambol enjoyer. Refugee from x.com/j0hnparkhill. bbgky.net

PostsRepliesMedia
John Parkhill @johnparkhill.bsky.social · 11/06/2026
Generative chemistry requires fast (~ms / molecule / GPU) and accurate retrosynthesis. Miles and Zach did a really good job with this: www.terraytx.ai/news-and-res...
terraytx.ai
TerraSynth - Terray Therapeutics
Built at the Intersection of AI and Experimentation
041
John Parkhill @johnparkhill.bsky.social · 24/03/2026
That's a follow folks.
010
John Parkhill @johnparkhill.bsky.social · 20/03/2026
Use colab for GPU if you like. Follow the instructions on this webpage: 164.92.115.7:5000. If you have trouble accessing the webpage you can DM, but otherwise no questions will be answered. Good luck. (Obviously you must be legally able to work for a US company to be hired)
164.92.115.7
000
John Parkhill @johnparkhill.bsky.social · 20/03/2026
-- Hiring ML Engineers time! -- Are you a smart person who wants to make therapeutics to help people using your ML chops? We will hire two engineers. We don't care about your pedigree, just your skills. Prove them by attempting the exercise linked below (<6 hours)
164.92.115.7
210
Reposted by John Parkhill
Derek Lowe @dereklowe.bsky.social · 29/01/2026
As illustrated every damn day in this wannabe administration.
1537
John Parkhill @johnparkhill.bsky.social · 28/01/2026
This is my new email signature.
011
John Parkhill @johnparkhill.bsky.social · 28/01/2026
Getting a 6mo lead in reading papers pre arxiv : priceless. Hilarious ppl are just taking the motivations for this naively. openai.com/prism/
openai.com
Prism | A free, LaTeX-native workspace for scientists
Write, edit, and collaborate on scientific documents in LaTeX with Prism—a free workspace integrating GPT-5.2 into research and writing.
010
John Parkhill @johnparkhill.bsky.social · 27/01/2026
These are the chipotle slop bowl of science. -- Me at lunchtime.
000
John Parkhill @johnparkhill.bsky.social · 26/01/2026
Carb free sweetened beef is the telelogical endpoint of American civilization, and that goal is in sight. arxiv.org/abs/2601.145...
arxiv.org
De novo design of protein binders targeting the human sweet taste receptor as potential sweet proteins
Excessive consumption of dietary sugars is a major contributor to metabolic disorders, driving global interest in finding alternative sweeteners with reduced caloric impact. Natural sweet proteins, su...
020
John Parkhill @johnparkhill.bsky.social · 25/01/2026
It's 16 years after citizens united. If you want to resist the government what you do is sell US stock, treasuries and buy gold, and foreign stock. (really)
000
John Parkhill @johnparkhill.bsky.social · 25/01/2026
000
John Parkhill @johnparkhill.bsky.social · 23/01/2026
Yeah... and if you open source the firmware, everyone wins :P
010
John Parkhill @johnparkhill.bsky.social · 23/01/2026
That's an M5stack, You can use claude to re-create and flash this firmware if you like it. It's a fun weekend project.
111
John Parkhill @johnparkhill.bsky.social · 18/01/2026
If you've actually managed to avoid holiday sickness by shifting hemispheres with your family this winter or sequestering. I wanna hear from you. HOW DID YOU DO IT.
000
John Parkhill @johnparkhill.bsky.social · 16/01/2026
It's basically impossible that Chat bot Ads will remain passive, because any competitor can do the agentic ads, and they will have a margin that is hundreds of times larger.
000
John Parkhill @johnparkhill.bsky.social · 16/01/2026
It's sade's birthday -- The real valentine's day for those who celebrate.
010
John Parkhill @johnparkhill.bsky.social · 01/12/2025
My dread of holiday flights...
020
John Parkhill @johnparkhill.bsky.social · 06/06/2025
This is the way arxiv.org/abs/2506.044...
arxiv.org
Multiscale guidance of AlphaFold3 with heterogeneous cryo-EM data
Protein structure prediction models are now capable of generating accurate 3D structural hypotheses from sequence alone. However, they routinely fail to capture the conformational diversity of dynamic...
150
John Parkhill @johnparkhill.bsky.social · 21/05/2025
Beautiful. arxiv.org/abs/2505.13447
arxiv.org
Mean Flows for One-step Generative Modeling
We propose a principled and effective framework for one-step generative modeling. We introduce the notion of average velocity to characterize flow fields, in contrast to instantaneous velocity modeled...
061
John Parkhill @johnparkhill.bsky.social · 08/05/2025
The problem with healthcare in the US is the lack of a market. Consumers are captive, and there is no price discovery process. Despite this, improved technology continues to arrive because the humans pursuing treatments genuinely care.
050
John Parkhill @johnparkhill.bsky.social · 08/05/2025
... And yet we live in a time where fairly complex quantitative reasoning is free. Some would say a reckoning is due. Anyone reading resumes and using AI these days is probably genuinely worried... I am.
000
John Parkhill @johnparkhill.bsky.social · 07/05/2025
I can't decide how I feel about someone vibecoding what is asked for in the job description. It shows initiative? But probably should have added "oh, don't put vibecode emoji's everywhere" to the system prompt.
010
John Parkhill @johnparkhill.bsky.social · 07/05/2025
When your whole exec team is dumping the stock for the last year, this is implied.
030
John Parkhill @johnparkhill.bsky.social · 05/05/2025
The persistence of docking in drug design has been explained. :P www.nature.com/articles/s41...
nature.com
Using an algorithmic approach to shape human decision-making through attraction to patterns - Nature Communications
Here, using data from a competition between academic teams to influence decision-making, the authors show that people, intending to maximize their profit, prefer options delivering rewards in a discov...
000
John Parkhill @johnparkhill.bsky.social · 02/05/2025
abcnews.go.com/Politics/rfk...
abcnews.go.com
RFK Jr. reports up to $1.2M in credit card debt, $30M net worth
From law firm payouts to endorsements and book deals, nominee for HHS secretary Robert F. Kennedy Jr. raked in at least $12 million in the past two years.
000
John Parkhill @johnparkhill.bsky.social · 02/05/2025
I will fund a study examining either: author preference for outcomes of an LLM based review process over human cartel, or publication impact in an LLM review process vs. human cartel.
010
John Parkhill @johnparkhill.bsky.social · 01/05/2025
gemini is pretty good at making manim animations with a little help, but boy without UV, manim would be impossible to install.
000
John Parkhill @johnparkhill.bsky.social · 28/04/2025
Yes. this type of thing happens constantly in drug discovery, the cereblon binding etc. but that's precisely why models cannot _replace_ experiments. They are based on observations which are themselves often contradictory, or impossible to explain without entire new domains of knowledge.
000
John Parkhill @johnparkhill.bsky.social · 28/04/2025
If you think we can avoid Thalidomide like outcomes without animal models today, you're an idiot. If you think you're merely marketing, you should appreciate that idiots will listen to you, and the consequences will be real. I will bet anyone 1000$ if this happens it ends poorly.
3303
John Parkhill @johnparkhill.bsky.social · 25/04/2025
The transition between "wins nobel prize" and "says vitamin C cures mortality" or "HIV doesn't cause AIDS" or "lobotomies for everyone" or "eugenics is cool" - or "disease is over because I interpolated crystal structures" is crazy short. (all utterances of nobel prize winners)
010
John Parkhill @johnparkhill.bsky.social · 20/04/2025
A virtual cell is like trying to emulate a computer with only the rom, no instruction set, and 20 types of registers you don't even know exist and have no hope to simulate because their physics is chaotic and nonlinear.
020
John Parkhill @johnparkhill.bsky.social · 20/04/2025
Only someone who can't think at 2nd order would ever take virtual cell models seriously without many unrelated breakthroughs. There are obvious, indisputable reasons they are stupid, and no one needs them more than me.
130
John Parkhill @johnparkhill.bsky.social · 18/04/2025
If your field is getting more "perspectives" papers or "position" papers, you're going to get feudalism and dreck. I'd hop :)
000
John Parkhill @johnparkhill.bsky.social · 18/04/2025
"give it to god"-- god of course being the transformer.
000
John Parkhill @johnparkhill.bsky.social · 18/04/2025
William Gibson owes it to the world to do another neuromancer, but where it's just descriptions of how lame the real person or prompt is behind every high follower X shitpost account. Instead of the real sexual content, dms with e-girls. Coin swindles making 5k instead of heists
000
John Parkhill @johnparkhill.bsky.social · 16/04/2025
Honestly these mostly solidified mafias. Certainly not the worst loss within the trashfire of cuts.
000
John Parkhill @johnparkhill.bsky.social · 15/04/2025
030
John Parkhill @johnparkhill.bsky.social · 15/04/2025
000
John Parkhill @johnparkhill.bsky.social · 14/04/2025
If universities don't stand up to the "dark enlightenment" they will lose all their relevance.
000
John Parkhill @johnparkhill.bsky.social · 10/04/2025
We're Hiring. ats.rippling.com/terraytx/job...
ats.rippling.com
Machine Learning Engineer | Current Openings at Terray
Position Summary: &nbsp;Terray Therapeutics is seeking a Machine Learning Engineer to lead the development of cutting-edge models and tools...
000
John Parkhill @johnparkhill.bsky.social · 04/04/2025
MSAs are precisely the reason this field is stuck...
010
John Parkhill @johnparkhill.bsky.social · 03/04/2025
The amount of disrespect one has for other people's thoughts, is often crazily proportional to how desperately one craves approval and validation. It's a joke embedded in humanity.
010
John Parkhill @johnparkhill.bsky.social · 03/04/2025
I wanted to say something comforting, best I got is: at least you don't need to have rage guilt/doubt?
030
John Parkhill @johnparkhill.bsky.social · 01/04/2025
Imagine getting ratio'd by a guy with half your rizz on your own crappy propaganda platform 😂
020
John Parkhill @johnparkhill.bsky.social · 28/03/2025
Beautiful work. So beautiful I think it must be viewed as a suggestion. Still pretty. biorxiv.org/content/10.1...
biorxiv.org
Enthalpy-entropy trade-offs in the evolution of ligand specificity in a superfamily of transcription factors
Proteins operate through ligand and solvent interactions governed by thermodynamics, yet the enthalpy-entropy trade-offs that guide their functional evolution remain poorly understood. The LacI/GalR f...
000
Reposted by John Parkhill
Derek Lowe @dereklowe.bsky.social · 27/03/2025
Harvard Medical School, are you OK with this outcome? Who will speak up for her?
621953
John Parkhill @johnparkhill.bsky.social · 27/03/2025
You can also optimize a basket of decisions such that it's good whether or not you have signal - the epistemic uncertainty that always dominates our world. If you look for people/entities optimizing baskets of decisions rather than greedily suddenly you see it in all success.
000
John Parkhill @johnparkhill.bsky.social · 27/03/2025
One of the most powerful ideas unused by the average person is to optimize baskets of decisions rather than individual decisions. The reason for this is: even if you have no MEAN signal on the individual decisions, you can optimize the basket.
100
John Parkhill @johnparkhill.bsky.social · 26/03/2025
Is medicinal chemistry semi-martingale? 😀
010
John Parkhill @johnparkhill.bsky.social · 25/03/2025
For giving each other ego handjibbers ofc.
000