search

HaotianZhangAI4Science / Delete

Delete: Directly optimizing lead in protein pockets, including linker design, fragment elaboration, scaffold hopping and side-chain decoration
MIT License
20 stars 2 forks source link

Is there something fundamentally wrong with deleting large molecules #5

Open amin-sagar opened 8 months ago

amin-sagar commented 8 months ago

Thanks again for this awesome work.

I am wondering if there is something wrong in principle in trying to optimize large molecules like peptides using Delete. I would like to start with the protein-peptide crystal structure and perform fragment growing on the peptide. Since the peptide would be kept rigid, this should not pose problems in finding the correct conformation of the peptide, just the small groups that are added to it. Do you think this is doable or is there something wrong with this approach? I would really appreciate any suggestions. Best, Amin.

HaotianZhangAI4Science commented 7 months ago

Dear amin-sagar, Sorry for the delayed response. I think your approach is right, especially when you consider adding some functional groups to the original peptide. Though the method is not trained for this, it can work. The specialized considerations of peptide representation in the model could improve the performance. This year, I can try to create a peptide version that fits your intention! Warm regards, Odin

amin-sagar commented 7 months ago

Thanks @HaotianZhangAI4Science I have tried on a couple of peptides. Although, the process takes quite long, possibly because of the large size of the peptide, the initial results are quite encouraging. Is there a way to specify the attachment points for fragment growing as there are only one or two atoms where growing is expected to give good results. I guess this will increase the speed. A model that can consider the peptide representations would be awesome. In a way, this can be done by partial diffusion using RFdiffusion. But, that would be limited to natural amino acids. An approach that combines a peptide representation model with a small molecule model would be really powerful.

HaotianZhangAI4Science commented 6 months ago

Dear amin-sagar,

Thanks for your advice! I'm going to incorporate Delete into the Rosetta Full Atom model, named Rosetta-Delete. Your advice about specifying attachment points will be implemented in the released version! Please send me an email address and I will tell you when we have done!

Best, Odin

amin-sagar commented 3 months ago

Dear Odin,

This sounds great. Please let me know when you release this. My email address is aamin.sagar@gmail.com. You can close this issue if you like or keep it open.

Best, Amin.