HeliXonProtein/binding-ddg-predictor

Does the mutation sites are designed by manual?

meetxinzhang opened this issue · 2 comments

Hi, I have read your research paper " Deep learning guided optimization of human antibody against SARS-CoV-2 variants with broad neutralization" and have a question that where is the Mutation Libraries from?

Where the description in citation are:

  • Variant-specific amino acid substitutions in the RBD were first created to build complexes with different antigen variants. For each antigen variant, single-point mutations in CDR of the antibody were enumerated.
  • While the initial structure of P36-5D2 bound with the RBD, the neutralizing activity of P36-5D2 against SARS-CoV-2 WT, Beta, Gamma, and Delta was then utilized to analyze and select CDR mutations. Based on those data, our method generated an in silico mutation library of antibody CDRs

So what are the details?
Thank you.

I think you can used amber software to mutate the residue of the Ab (such as : JACS DOI: [10.1021/ja027200q]https://doi.org/10.1021/ja027200q), then run MD simulations to get optimized structure (this is very important)..I have tested the Urelumab-41BB ^^

@c00jsw00 Thanks for ur help. So, the deep learning model proposed by the citation only predicts free binding energy but does not design the antibody. Like the title said "guided".