Computational Biology PhD @Yale MBiochem @UniofOxford
- Tweets 13
- Following 162
- Followers 76
- Likes 39
Excited to share KnitNet at tomorrow's Eterna seminar!
Join us Wednesday, Oct. 14, at 3:30 pm ET (12:30 pm PT).
I'll discuss how we combine generative modeling with verifier-guided optimization to accurately design RNA sequences for complex structures, including pseudoknots.
Join the Eterna community tomorrow, Oct. 14 at 12:30 pm PT to hear @AF_Panescu discuss her RNA inverse design method and HJ Lee introduce his 3D prediction puzzle for modeling RNA aptamers docking a ligand.
Details at: eternagame.org/news/16535855
1
3
8
428
Antonia Panescu retweeted
Thrilled to announce Zatom-2, an atomistic generative model spanning molecular, material, and protein science! Special thanks to @MirunaCretu2 for leading this effort. (1/11)
🌐 Project page: zatom-ai.github.io/nucleus
ALT Zatom-2's multitask pretraining architecture for atomistic chemical and biological data.
12
17
96
9,397
Antonia Panescu retweeted
Major props @DidiKieran @jgkoob @lemerleau @IIMCB_Poland @FrankRHutter @AF_Panescu @chaitjo @rishabh16_ for generating 600,000 sequences to fold into 1000 synthetic pseudoknots! The wet lab results will help these researchers improve their computational methods for RNA design.
OpenKnot Round 9 features secondary structures designed by players to create compact, pseudoknotted RNA using 240 bases. GenAI teams also will test whether their methods perform as well as the expert players at a longer length.
Check out the puzzle at: eternagame.org/labs/14343053
ALT Screenshot of Eterna puzzle interface
3
8
4,374
Presented this work at #ICML2026 @genbio_workshop !
We introduce an order-agnostic approach to RNA inverse design. The result is more diverse, higher-confidence designs with higher solve rates while requiring fewer samples than standard best-of-N methods.
1
5
51
4,835
📄 Paper: openreview.net/pdf?id=3tWscv…
💻 Code: github.com/antonia-panescu/S…
4
388
Had a great time attending ChemAI NYC 2026, presenting #CatalystBusters
If you’re looking for a physics-based, deterministic benchmark for your ML-generated enzyme sequences, consider using this!
1
1
10
1,187
Had a great time contributing to this collaborative effort across biomedical domains!
A key question we asked was: can AI agents tell when the available data is sufficient to support a reliable conclusion? An important step toward trustworthy AI-driven scientific discovery 🤖🧬
1/ 🤖Can agents solve problems in science? Most agent benchmarks test math/coding or reduce research to Q&A. Neither captures real research: multi-step workflows, messy data, tool use, knowing when a task is even feasible. So we built SciAgentArena: 200 tasks, 6 domains.
2
583
A Lab with a View
Excited to spend my summer at the @BerkeleyLab in California, extending our previous work on Zatom-1!
Check out the paper here: arxiv.org/abs/2602.22251
1
2
9
2,146
This summer, I’ll also be at ChemAI NYC 2026 presenting CatalystBusters, a physics-based filter for triaging enzyme variants before synthesis, and at @icmlconf in Seoul, South Korea, where I will share some recent methodological work for RNA design.
1
282
Antonia Panescu retweeted
Introducing Zatom-1, the first end-to-end, fully open-source foundation model for 3D chemistry! This was a great collaborative effort with many brilliant scientists. I'm grateful to have played a small part.
Paper: arxiv.org/abs/2602.22251
Code: github.com/Zatom-AI/zatom
ALT Architecture of the Zatom-1 foundation model for 3D molecules and materials. A unified Transformer trunk processes (noisy) atomic, structural, and conditional inputs. The model features two training stages: (1) multimodal flow pretraining using the final trunk layer L’s representations to create new molecules or materials; and (2) multi-task finetuning of additional downstream task backbones and heads that use a specified trunk layer K’s representations to predict properties, energies, and forces.
3
28
137
9,134