Pascal Notin
@NotinPascal
Followers
1K
Following
2K
Media
43
Statuses
443
Research in AI for Protein Design @Harvard | Prev. CS PhD @UniofOxford, Maths & Physics @Polytechnique
Boston
Joined September 2020
🧬 😴 TIRED: Scaling protein models to billions of parameters hoping they'll memorize all of evolution and generalize beyond.🔥 WIRED: Smart retrieval-augmented models that dynamically access what they need from sequence databases.
🚨ICML Paper Alert🚨.What if finding the right protein homologs wasn't a slow search, but a learned part of the model itself?.We introduce 𝐏𝐫𝐨𝐭𝐫𝐢𝐞𝐯𝐞𝐫, an end-to-end framework that learns to retrieve the most useful homologs for self-supervised reconstruction! (1/12)
1
6
92
RT @AlexanderTong7: Thrilled to announce I'm starting as a Principal Investigator at #Aithyra in Vienna! We'll be developing generative mod….
0
37
0
RT @AlissaHummer: Our paper on generalizable antibody-antigen binding affinity prediction has been featured on the cover of the August Issu….
0
7
0
RT @ferruz_noelia: I’ve thoroughly enjoyed reading two (VERY!) recent papers that model protein sequences by retrieving evolutionary inform….
0
28
0
RT @Nature: A generative artificial-intelligence tool has designed a synthetic CRISPR system that successfully edits human DNA. https://t.c….
nature.com
Nature - A generative artificial-intelligence tool has designed a synthetic CRISPR system that successfully edits human DNA and sharply reduces off-target effects.
0
141
0
RT @AvivSpinner: 1/5 Biological data is noisy, redundant, and ever-growing. 🗣️. In our new paper (first paper of my post doc!! ⚡️), we trac….
0
24
0
0
0
1
RT @LabWaggoner: AI expands the repertoire of CRISPR-associated proteins for genome editing.@NatureNV preview by @NotinPascal . https://t.co….
0
2
0
My @Nature News & Views on this breakthrough: Special thanks to @AvivSpinner for their valuable feedback, and to @nature and @NatureNV for the support in writing this piece! 🙏.
nature.com
Nature - A generative artificial-intelligence tool has designed a synthetic CRISPR system that successfully edits human DNA and sharply reduces off-target effects.
1
0
12
Congratulations to the entire @ProfluentAI team on this incredible milestone! OpenCRISPR-1 represents a paradigm shift - the first AI-designed CRISPR protein to successfully edit human DNA with fewer off-target effects. We're moving from discovery-based to engineered biology. 🧬.
Excited to have our AI research published in @Nature today. Proud of the @ProfluentBio team and the extensive final version available under open-access. OpenCRISPR is a milestone. It's the first successful demonstration of editing the human genome with a molecule fully designed.
1
4
74
RT @Align_Bio: 1/4.🚀 Announcing the 2025 Protein Engineering Tournament. This year’s challenge: design PETase enzymes, which degrade the ty….
0
72
0
RT @schwabpa: Save the date! Machine Learning for Drug Discovery (MLDD) is happening soon on Monday 30 June, 2025. MLDD aims to bring toge….
0
30
0
RT @AvivSpinner: I wonder if the gym membership for ProteinGym covers RNAGym as well?? oh wait, it's all free and 100% open source!.
0
4
0
Links:.🔗 Paper: 💻 Code: 9/9.
github.com
Contribute to MarksLab-DasLab/RNAGym development by creating an account on GitHub.
0
1
5
Congratulations to the entire RNAGym team! @rohitarorayyc @mvrfalo @christian_choe_ @c_sheare @aaronkollasch @fiona__qu @ruben_weitzman @artemg2718 @sarahgurev Erik Xie @deboramarks .8/9.
1
0
6
The moderate performance across all tasks reveals exciting opportunities! Key directions: RNA-specific training data, integrating structure-function relationships, and improving non-canonical base pair prediction. RNAGym provides a standardized foundation for progress. 7/9.
1
0
1
🌀 Tertiary structure: 215 diverse 3D structures from recent PDB entries. NuFold leads monomers (0.393 TM-score), AlphaFold3 dominates complexes (0.381 TM-score). Non-Watson-Crick interactions remain a major challenge for all methods.6/9
2
3
7
🔗 Secondary structure: 901k chemical mapping profiles using DMS & 2A3 reactivity. EternaFold achieves top performance (0.656 F1-score), closely followed by CONTRAfold & Vienna. Traditional thermodynamic methods are still competitive with newer deep learning approaches.5/9
1
0
1
🔬 Fitness prediction: 70 assays across tRNA, ribozymes, aptamers & mRNAs (1M+ mutations). Evo 2 performs best overall, but performance varies dramatically by RNA type: RNA-FM excels at tRNA/aptamers while Evo 2 leads mRNA tasks. Lots of room for improvement across the board!.4/9
1
0
2
RNAGym tackles three essential RNA prediction tasks: 🔬 Fitness prediction: How mutations affect RNA function 🔗 Secondary structure: Base-pairing patterns 🌀 Tertiary structure: 3D molecular architecture.All evaluated zero-shot to test true generalization!.3/9.
1
0
2
Why do we need this? RNA modeling faces major challenges: limited experimental data (<1% of PDB entries), inherently less stable structures than proteins, and evaluation has been scattered across different studies with varying approaches. 2/9.
1
0
4