entrepreneur, founder Om, founder of 1859, inc. — love intersection of AI and techbio — prefer to build.
San Diego, CA
Joined June 2009
- Tweets3.7K
- Following2.2K
- Followers655
- Likes6.7K
Devon Cayer retweeted
Enough of the cure talk. Where are the cures for these diseases?
x Alzheimer's
x Parkinson's
x Heart and vascular diseases
x Diabetes
x Hypertension
x Most cancers
x Multiple sclerosis
x Most autoimmune diseases
X Kidney diseases
x Liver diseases
x Lung diseases
x Depression
x ALS
and a long list of others
Devon Cayer retweeted
> mythos is so good at cyber it can't be released
also
> mythos can't detect 20k fraudulent chinese accounts attacking it
Devon Cayer retweeted
Introducing our first model, Un-0!
We trained an image generator powered by a backbone of coupled oscillators in place of a more traditional conventional neural network.
Devon Cayer retweeted
Studying biology has long meant working with fragmented tools. With Atera, for the first time ever, biology can be measured as a complete system in its native context, at scale, removing the constraints that have held spatial biology back: bit.ly/48bYvYF
ALT 10x Genomics introduces Atera, a new platform to redefine how biology is measured and understood.
ALT Atera removes the trade-offs that have constrained research, unlocking a new era of insight that will transform our understanding of science and human health.
ALT Atera delivers whole-transcriptome spatial data with single-cell sensitivity at scale.
Devon Cayer retweeted
Today, we’re launching AI Connectors, new MCP-based capabilities that connect scientific data across AI tools in R&D.
Teams can now use data from common enterprise applications like Notion, SharePoint, and Snowflake directly in #BenchlingAI.
Pre-built connectors for life science applications like @elicitorg, @gxl_ai, @quiltdata, @SeqeraLabs also bring relevant scientific context together with your experimental records.
AI in science only works when you can connect your data to the right tools.
Learn more: benchling.com/news/benchling…
Devon Cayer retweeted
What's the real moat in AI drug discovery? Not the models.
Jacob Berlin, cofounder and CEO of @Terray_Tx, and his team built a microarray chip the size of a fingernail that can measure billions of small molecule interactions. Their chemistry data set is now 40x larger than what’s publicly available. For small molecules, where the search space is essentially infinite, that scale makes all the difference.
His take: "You can't learn fast enough if you're only distributing your models to others."
On this episode of Transcribed, Jacob and I talk about why chemistry is harder than biologics for AI, what "AI abundance" means when your models give you 10,000 molecules and your chemists can only make 50, and why whatever you pick to start on is your actual business.
Devon Cayer retweeted
Unpopular opinion: iPTM > 0.9 doesn't mean your protein will bind.
Aalpha Bio measured 7 million binding affinities. Correlation with iPTM/IPSAE? Almost zero.
These scores are great negative filters. High score = maybe. Low score = definitely not.
If you're picking your top 50 candidates by iPTM alone, you're leaving good designs on the table and sending bad ones to synthesis.
Multi-metric composite scoring.
Every time.
Devon Cayer retweeted
We just released a big update to Rowan's Python API, with the goal of making it easier for (1) regular programmers (2) people using Claude Code/Codex, and (3) "AI scientists" to build atop Rowan.
Every workflow now returns a typed output result that mirrors the web interface:
Devon Cayer retweeted
RNA-seq analysis in seconds using GPUs | bioRxiv biorxiv.org/content/10.64898…
Devon Cayer retweeted
Our biggest feature launch yet—we've been working on this for almost as long as it took us to get the MVP of Rowan out!
Excited to finally share Rowan FEP with the world:
Devon Cayer retweeted
I tested models from @AnthropicAI @OpenAI @Google @Zai_org @MiniMax_AI and @Kimi_Moonshot on whether they can create a publication-level view of a protein-ligand binding site and the results were surprising.
TL;DR: Anthropic models did the best and Gemini models did the worst.
Task was simple:
"Load 5DEL and create a publication ready view of the ORO and FMN binding site"
Here are the results:
1. Sonnet 4.6 from @AnthropicAI:
- Fast
- Used the tools pretty well
- Created the best view with clear labelling and color choices
- always felt in control of what it was going for
- very impressed
I am Open-Sourcing PyMolAI!
Meet PyMolAI, an AI agent that can talk to your protein structures.
Built on top of PyMOL, PyMolAI lets you interact with your structures in plain language. Whether you're:
- Analyzing protein structures
- Aligning complexes
- Creating publication-ready figures
- Or running design workflows
PyMolAI interprets your request, executes the necessary PyMOL commands, and manages the workflow for you.
It integrates with @OpenBioAI APIs, giving you access to tools like Boltz, ProteinMPNN, and BoltzGen — directly from your PyMOL session.
It has local chat history with session syncing, so you can pick up exactly where you left off.
Devon Cayer retweeted
I decided to experiment with my own (small) protein design competition!
It's specifically to test how well VHH pipelines work for binder design, without the usual careful tuning.
You submit a @modal script, I run it and test on benchbb.bio
blog.booleanbiotech.com/vhh-…
Devon Cayer retweeted
I used Claude Opus 4.5/4.6 (and a bit of Codex GPT-5.3) to port edgeR to Python. See edgePython github.com/pachterlab/edgePy…
This allowed me to develop a single-cell DE method that extends NEBULA with edgeR Empirical Bayes. All in one week. Details in doi.org/10.64898/2026.02.16.…
