Will Fondrie
@willfondrie
Head of Data Science & Engineering at Talus Bio. I post about data science, BioML, AI drug discovery, Python, mass spec, proteomics, and silly things my kids do. I also blog sometimes:
(11/13) I guess what I'm trying to say is: Welcome to the structure-free future of drug discovery.
(10/13) My favorite story is about the recent STAT6 inhibitors disclosed by Pfizer. Ptarmigan-1 is not only able to enrich for these in a screening scenario, but also localizes them to a shallow groove on the SH2 domain.
(9/13) Now the targets we actually built this for: 👉 intrinsically disordered: Boltz-2 goes near random, Ptarmigan-1 still enriches for active ligands 👉 cryptic pockets on CryptoBench: we beat P2Rank despite this being zero-shot, never trained to find pockets.
(8/13) Localization is where it gets fun. We score every residue and get a binding site map from sequence alone: 👉 top residue within 5 Å of the ligand for 92% of drug-like PoseBusters complexes 👉 engaged cysteine in the top 1% for 8 of 9 COValid targets, held-out ones included
(6/13) A screen with Ptarmigan-1 is just a nearest-neighbor lookup instead of a simulation. This clocks in at 10 ms per compound - ~5,000x faster than when we ran Boltz-2 on the same hardware. All 20,431 human proteins against a 3.4B compound library: 20 H100-hours.
(5/13) Protein-level labels from bioactivity can be used. Residue-level labels from chemoproteomics data can be used. We can even extract residue-level labels from the structures in the PDB. Key to it all: our massive swath of internal data on the disordered proteins of the regulome.
(3/13) So we skipped the structure. Ptarmigan-1 embeds every residue of a protein and every compound into one shared latent space, using fine-tuned foundation models for sequence and chemistry. The residue-level resolution is the real innovation here. Engagement is just cosine similarity. 🤖
#ASMS2026 coffee pilgrimage day 4: Torque Coffee. I had a nice cortado with their lighter espresso, Yellow Diamond (yes, they have two options!) This one is a bit further out in North Park, but a nice trek for some good coffee. Thanks for the rec @michaellazear.bsky.social!
#ASMS2026 coffee pilgrimage day 3: Goldchild coffee. It's a bit of a hike downtown, but 100% worth it. This one is the best I've tried so far. www.goldchildcoffee.com
Saw a post about GPU depreciation and I could hardly believe what second-hand H100s were selling for. The GPU market is wild.
It was nice to get away to Mt Rainier 🏔️ for a long weekend camping trip. We even got to see some marmots!
Definitely PhD level. No question 👏 But seriously, I thought other folks were joking 🙃
Even with everything going on, happy Indepenence Day to my friend in the US. Here is a version of “Battle Hymn of the Republic” that I’ve been working on. 🪕 I’ll eventually learn to sing, but today is not that day 🙃
Stop by poster ThP 289 to… calmly discuss… metadata collection and management with Anastasiya 👏 #ASMS2025
It was a lot of fun participating in the Biomarker workshop panel at #ASMS2025. Thank you to all of you who were there! Also, they got some pictures:
Fueling up with a cortado at Ceremony again this morning Come see my poster #ASMS2025 on Tuesday about our recommender system method for predicting the retention times od undetected peptides… called Cortado 😉 #ASMSBmoreEats
The sad news is that my hiking boots for the last 12 years finally deteriorated. So many adventures in them! The AT, Harpers Ferry, Chincoteague, 3 ft of snow in Baltimore, MN trails, Rainier, all kinds of trails in WA… You will be missed 🫡
First camping trip of the season done and dusted 🏕️ It was great to disconnect for a few days and I’m looking forward to the others we have planned this summer!
@jspaezp.bsky.social is talking about #AI in #Proteomics tomorrow as part of a webinar put on by @us-hupo.org and @abrf.bsky.social. Join in and hear about how ML and AI are integrating into all aspects of the proteomics workflow! 🤖 Register here 👇 us02web.zoom.us/webinar/regi...
Quote post with a picture you took in a National Park Mt Rainier National Park last summer 🏔️
I had fun writing a bit about #AI for #DrugDiscovery with Milad and Megan 🤖 Two thoughts: 👉 It's all about the data. Co's that have the best, proprietary data to power their models will win the day. 👉 A tightly coupled wet and dry lab shorten the active learning feedback loop. (1/3) 🧵
We got outside today and it was a refreshing break from the final grant writing push this week 😊
The kids wanted me to watercolor with them this weekend. They asked me to paint a robot 🤖 I think I nailed it 😉 …and ended up with a winning white elephant gift for someone this week? 😂
My first wetlab experiment in years was a success! I taught the my budding scientists about chromatography this morning. We separated colors in markers using coffee filters 🧑🔬 They’re well on their way to joining #TeamMassSpec 😉