SandboxAQ

/

AQAffinity

like 30

Follow

SandboxAQ 118

Thanks for open-sourcing this! Quick question regarding the cropping strategy for training on super large proteins.

The Boltz-2 paper mentions a pocket-centric cropper limited to 256 tokens total (max 200 protein tokens). Did you strictly stick to this 200-token limit for your training set, or did you adjust the crop size to accommodate larger complexes?

Also, do you have the definitions of the pockets you used available to share?

Thanks!