How to generate more small molecule candidates?

[kenbyler]

I provided by own PDB structure with a bound ligand as a template for generating novel structures. The example in the notebook produced two complexes. I thought to increase that number by "--design_samples 100", but that gave me many strange protein structures with identical small molecules. What is the best way to increase the number of generated small molecules, while retaining a good structure prediction for the protein?

[yehlincho]

Hi, thank you for the question. What do you mean by "with identical small molecules"? Also, since we currently can’t use a template for the small molecule, I’m wondering what your input and output are.

[kenbyler]

I was starting from a PDB of a protein-ligand complex for which I had modeled a loop. I may have misinterpreted what BoltzDesign is doing from the biorxiv paper. It suggests that it designs small molecule ligands to bind to the provided protein.