backbone with clashes

[ntxxt]

Hi,

Very interesting work.

I tried the sampling command 'python ~/projects/foldingdiff/bin/sample.py -l 50 128 -n 10 -b 512 --device cuda:0' and I observed a lot of steric clashes within the generated PDB structures. Any insights about where these clashes might come from? Based on the paper, it seems rebuilding the backbone from angles does not really cause problems. So it could be the training/mode/data?

Best,
Xiaotong