We are exploring optimal computational metrics for design selection and experimentally characterizing designed aptamers to advance structural-level nucleic acid design. Check out our preprint here: https://t.co/wst3V54EAZ
We are excited to announce that HalluDesign-NA is now available on GitHub: https://t.co/zeiu328ft9
We have expanded the HalluDesign framework to nucleic acids, enabling de novo design of nucleic acids by leveraging the hallucination ability of AF3-like models.
We are excited to announce that HalluDesign, an all-atom protein optimization and design platform, is now open-sourced and we have also updated our preprint:
https://t.co/tvzThrZBg9
https://t.co/BHRAzavb5O
The new version includes more detailed benchmarks and our newly developed ssRNA binders.
In addition, we are actively exploring other applications, including nanobody binding affinity optimization and cyclic peptide design.
Latest publication of our lab! 🎉
Amantadine-controlled protein-designed assembly, as an ideal amantadine-inducible gene switch for therapeutic applications
@KevinKaichuang@WChentong Thanks, Kevin! We’ll be dropping the code and an updated preprint soon—hoping it moves the needle for the field. Would love your take once it’s out.
As we were about to submit, we were pleased to see parallel work from the Sergey team, which adopted a similar philosophy in exploring
AlphaFold3-like models for protein design. @ChoYehlin@sokrypton
Great to see more research in this space.
https://t.co/sqQGbARin9
🚀 New Paper: We’ve just released HalluDesign – a novel framework for protein optimization and de novo design using AlphaFold3-style models, without finetuning or gradient backpropagation! 🎉
https://t.co/tmqKqPMp2b
We also used a framework-fixed strategy to design a phospho-specific Aβ targeting antibody, with a binding affinity in the low hundred nM range, specifically binding only to the phosphorylated Aβ.
We designed seven small molecules or ion binders: ATP, CA, DTG, AD, cAMP, TRP, and Zn2+. For each target, fewer than 16 designs were tested, with at least one binder identified. Five of these binders were further characterized by crystal structures.
We replace HalluDesign prediction model with protenix, an AF3-style model built on protein language embeddings, which enables successful de novo design of monomers and diverse biomolecular binders by directly initializing random sequences.
Structure conditioning at different noise level in the structure prediction stage allows precise control over the sampling space, facilitating tasks from local and global protein optimization to de novo design.
A breakthrough in glioblastoma treatment! Qihan Jin of our lab, de novo designed binders on CAR-T cells to target tumor antigens. Remarkable results in mice, clinical trials are underway. #proteindesign#glioblastoma#CART@lxcaosd
https://t.co/8wzr3KktsH