Jian Tang
@tangjianpku
Followers
5K
Following
867
Media
36
Statuses
312
Associate Professor at Mila, AI for drug discovery. Generative AI and Geometric DL for protein design. Founder of @BioGeometryAI
Montréal, Québec
Joined July 2010
Check our latest work at #ICLR2025 on using discrete diffusion language models for understanding protein dynamics!.
ICLR'25 | Structure Language Models for Protein Conformation Generation.Keywords: language modeling | discrete diffusion | PLM fine-tuning.🔗arXiv: 🐱🐙Github: 📍Fri | 25 Apr. 10am | Hall 3 + Hall 2B.Happy to chat if you're in 🇸🇬!!
0
7
45
(5/5) Discerning binders from non-binders is one of the hardest tasks in biomolecular modeling. GeoFlow-V2 has a robust scoring module with highly competitive AUROC scores when benchmarked on 10 targets with ground-truth Ab:Ag affinity data.
0
0
2
(4/5) GeoFlow-V2 can design Ab/Nb with high epitope compliance and rich CDR-mediated interactions. Within 1000 samples, it recovers ground-truth interaction patterns (FR RMSD < 8Å) in 5/7 cases, while RFAntibody recovers for 2/7 cases.
1
0
1
(3/5) GeoFlow-V2-ab is a lightning-fast smaller model specifically for Ab/Nb structure prediction. It is faster and more accurate than current SOTA (~200x faster than AF2!). The model lays a solid foundation for our Ab virtual screening and optimization pipeline.
1
0
1
(2/5) GeoFlow-V2 achieves superior performance on low-homology Ab:Ag complex structure prediction. The score gets even higher with constraints specified. Supported constraints include hotspot (token), epitope (token2chain), contact (token2token) & structural templates.
1
1
4
🔥(1/5) Introducing GeoFlow V2: A unified atomic diffusion model for protein design.-Unifies structure prediction & de novo design w/ versatile constraint support.-SOTA results in Ab:Ag folding & epitope-specific Ab design. Try it and read our report at
2
66
296
We are giving a tutorial on AI for protein design at this year's #AAAI, covering topics including:.(1) Protein Representation Learning;.(2) Protein Structure and Dynamics Prediction;.(3) Protein Design. Welcome to join us!.
🤔Excited about Protein + AI? Join our "AI for Protein Design" tutorial at #AAAI2025, with gorgeous @Oxer22 @divnori @WengongJin @tangjianpku and Jiwoong Park. 🧬✨. 📅 8:30–12:30 (EST), Feb 26 .📍Room 117, Philadelphia Convention Center, Philadelphia.👉
5
19
124
Our recent work on protein conformation sampling. We proposed to deploy LLMs in the discrete latent space of protein structures. Code: Paper:
Structure Language Models for Protein Conformation Generation. • This study introduces Structure Language Modeling (SLM), a novel framework for efficient protein conformation generation by leveraging language models in a discrete latent space. • SLM provides a major
0
9
49
Billion-dollar markets in the future: autonomous vehicles, humanoid robotics, quantum computing (which he says is 10-15 years away) and digital biology.
Jensen Huang says a number of industries will soon go from 0 to billion-dollar markets: autonomous vehicles, humanoid robotics, quantum computing (which he says is 10-15 years away) and digital biology
0
0
13
RT @AIHealthMIT: 🚨Just 1 WEEK left to go before paper submissions close for #MoML2024! Students + postdocs can receive FREE admission to #M….
0
8
0
(3)Single-cell foundational models: Cell-ontology guided transcriptome foundation model, Please help retweet!.
0
2
11
Some of our latest work:.(1) Flow matching for protein design: (2) Pretraining GNNs for antibody optimization: Pretrainable geometric graph neural network for antibody affinity maturation, Nature Communication
1
2
14
We have access to thousands of GPUs through our industrial collaborators to train large generative foundational models on biological data, and collaborate with the best biologist in the world.
1
1
5
My research lab is recruiting multiple PhD positions next Fall at @Mila_Quebec. I'm looking for students to work on:.(1) Generative models for protein design;. (2) LLMs for multiomics (genomics, single-cell RNA seq, proteomics). Please DM or email me through tangjian@mila.quebec.
3
84
317
Check out our new generative-AI platform for antibody design, GeoBiologics (. GeoBiologics are developed based on a lot of proprietary data from our partners and some latest generative AI models. Welcome to try it out!.
Connect your biologics development to AI with GeoBiologics!. 🔬 Harness cutting-edge AI for de novo protein design.🎯 Achieve multi-objective optimization w/ ease.🛠️ Customize solutions to fit your needs.✨ Enjoy an intuitive user experience. Try now at
0
1
10
Excited to deepen our collaboration with the global leader @Sino_Biological on protein/antibody production to push the frontier of generative AI in protein design!. Another successful example of our protein generative AI foundational model GeoFlow in real-world applications.
Excited to deepen our strategic cooperation with @Sino_Biological! Combining our protein foundation model GeoFlow with Sino Biological's protein expression and wet lab platform, we aim to set new standards in protein R&D. Learn more at
0
0
9
Can we steer the transcriptome foundation model (TFM) trained on scRNA-seq data?. Check our recent work scCello, a transcriptome foundation model (TFM) guided by cell ontology. SOTA results on identifying new cell types, prediction of marker genes and drug responses.
Studying on developing foundation models on single cell RNA-seq (scRNA-seq) data? The serge continues!💪. We're excited to announce our new work "scCello", a single cell, Cell-ontology guided transcriptome foundation model (TFM) on scRNA-seq data. Paper:
0
1
19
Glad to participate in this exciting ML conference for drug discovery. I shared our recent progress on "Geometric Deep Learning for Protein Understanding". Slides available at:
In June, we hosted the molecular machine learning conference @Mila_Quebec. With 180+ attendees and 45+ posters, it was amazing to see the progress in ML research for drug discovery. Thank you everyone for a great conference! See you at the next MoML 👋
0
12
53
RT @jianfcpku: Congratulations! We have tested this model on a collection of 160 SARS-CoV-2 RBD mAbs that have not been published before, a….
0
2
0
🔥🔥 Check our latest generative model for antibody design GeoFlow, which can be used for both antigen-antibody complex structure prediction and de novo antibody design. SOTA performance on Ab-Ag complex prediction, comparable to AF3. Try it here:
💡Introducing GeoFlow: cutting-edge protein generative model with:.1️⃣SOTA Ab-Ag docking success rates: ≈ AF3, 2x AF2.2️⃣Exceptional design capabilities: GeoFlow generated mAbs w/ novel HCDR3 that have up to 3x binding to the same epitope.🔗Try now at
1
6
37