Amir Shanehsazzadeh
@amirshanehsaz
Followers
504
Following
90
Media
23
Statuses
98
Using generative AI to design antibodies at Absci (@abscibio)
New York, NY
Joined September 2021
OpenFold3-preview (OF3p) is out: a sneak peek of our AF3-based structure prediction model. Our aim for OF3 is full AF3-parity for every modality. We now believe we have a clear path towards this goal and are releasing OF3p to enable building in the OF3 ecosystem. More👇
3
66
221
@abscibio Finally, stay tuned for some exciting news on our latest antibody inverse folding model, IgDesign2 đź‘€
1
0
2
@abscibio As always, this work is a collaborative team effort, involving AI scientists and engineers, computational biologists, protein engineers, and wet lab scientists. If you're interested in joining such a team to advance the state of therapeutic antibody design, please reach out!
2
0
0
@abscibio We hope these datasets will enable benchmarking efforts! For example, we evaluated the ability of models such as ABodyBuilder2 and ESMFold to predict binding vs. non-binding using self-consistency RMSD (scRMSD).
1
0
0
@abscibio For a quick refresher on IgDesign, see my thread from last year: https://t.co/TGnt7qLbK6 We demonstrated in vitro success on antibody inverse folding for HCDR3 and HCDR123 design against 8 therapeutic targets. Model success rates were higher than a baseline as well.
We @abscibio are excited to unveil IgDesign™ @NeurIPSConf and present our work at @workshopmlsb @AI4D3 @genbio_workshop! IgDesign is an antibody inverse folding model that we have experimentally validated in our lab. Read the paper here: https://t.co/H5xDtNt0NE
#NeurIPS2023
1
0
0
We @abscibio are excited to open source the code and datasets for IgDesign, with over 1,000 SPR datapoints against 7 targets! https://t.co/Qzl7nW4Pzk With these data, we benchmark the ability of folding models to predict binding. More details below and in
biorxiv.org
Deep learning approaches have demonstrated the ability to design protein sequences given backbone structures [[1][1], [2][2], [3][3], [4][4], [5][5]]. While these approaches have been applied in...
4
27
121
I'm excited to release the code and datasets from our IgDesign work! Stay tuned for a longer post. For now check out our github repo: https://t.co/Qzl7nW4Pzk I'll be @workshopmlsb today. If you're interested in structure-based drug design, especially for antibodies, let's chat!
IgDesign: In vitro validated antibody design against multiple therapeutic antigens using inverse folding @abscibio 1. The first in vitro validation of antibody inverse folding for designing binders to therapeutic antigens. IgDesign achieves high success rates across eight
1
10
42
Come check out IgFlow! @workshopmlsb @NeurIPSConf
Excited to share IgFlow, a flow matching-based method for antibody design at @NeurIPSConf tomorrow! Come check out our poster at MLSB during the poster sessions in East meeting room 11/12. @abscibio @workshopmlsb
0
0
4
Thrilled to announce Boltz-1, the first open-source and commercially available model to achieve AlphaFold3-level accuracy on biomolecular structure prediction! An exciting collaboration with @jeremyWohlwend, @pas_saro and an amazing team at MIT and Genesis Therapeutics. A thread!
42
416
2K
I'm at UW for MLCB 2024! If anyone is here and interested in talking antibody data and deep learning for antibody design, feel free to reach out!
0
1
10
Welcome to the team!
I'm excited to share that I'll be spending the summer in NYC as an intern at @abscibio! Looking forward to work with their awesome team over the next few months 🎉 If you're in NY, send me a message! I'd love to connect both scientifically and socially (let's grab a beer!).
0
0
6
Last month, my colleague @EricDai_BioE and I attended dozens of Bio x ML workshops at @NeurIPSConf 2023. We've curated a list of papers, authors, and takeaways in a three part roundup series. Part 1 is on Generative Protein Design. https://t.co/4nbRKOQMYf TL;DR Below ⤵️
research.dimensioncap.com
Generative protein design took center stage at NeurIPS 2023. In part one of a three part series, we highlight several papers, authors, and takeaways about the intersection of ML and proteins.
3
31
93
I read this paper from $ABSI. Its really fascinating stuff on using generative AI to develop and predict antibody antigen binding. Very speculative, but a game changer if they get it to work. https://t.co/hxW72lckDx
biorxiv.org
Deep learning approaches have demonstrated the ability to design protein sequences given backbone structures [[1][1], [2][2], [3][3], [4][4], [5][5]]. While these approaches have been applied in...
0
5
20
@abscibio If you’re interested in joining such a team check out our current openings ( https://t.co/kQEPZ4tP3d) or DM me for more information.
absci.com
Join Absci and help reinvent drug discovery with AI and synthetic biology. Explore open roles and grow your career at the intersection of tech and biotech.
1
0
3
I want to acknowledge the many talented contributors involved in this work. This latest data on functionality and developability highlights the multidisciplinary team we have built @abscibio; a team with the knowledge, skills, and motivation to go end-to-end in AI drug creation!
1
0
6
We’ve also open-sourced a dataset of over a thousand trastuzumab variants screened in surface plasmon resonance (SPR) against human HER2! We hope this dataset will help advance new methods for AI antibody design ( https://t.co/B6htidmFgE).
1
0
4
We screened these mAbs against a panel of HER2 orthologs and paralogs and found matching cross-reactivity profiles to trastuzumab. One design (Candidate #3) had better species cross-reactivity than trastuzumab with 5x higher affinity to cynomolgus and canine HER2.
1
0
3
We ran cell-based functional assays and found that one of our designs (Candidate #2) showed 3x higher potency than trastuzumab in cell-surface antigen binding and antibody-dependent cell cytotoxicity (ADCC).
1
0
2
We screened a subset of our designs in mAb format for functionality and developability. We found several developability profiles comparable to trastuzumab, a clinically approved antibody, and some with superior functionality. We’ve open-sourced this data for the community to use!
1
0
2