Eikor / InstructPLM

The first large protein language model trained follows structure instructions.
Apache License 2.0
71 stars 5 forks source link

Fix residues #3

Open ioioba opened 3 months ago

ioioba commented 3 months ago

Hi thank you for your work. Is it possible to add constraints for the generation, espacially fix certain resiudes? thanks in advance.

Eikor commented 3 months ago

Dear ioioba, Fix-residue design is a vital feature. However, unlike other pLMs such as ESM, InstructPLM is an auto-regressive language model, it can only support one-direction decoding, i.e. from the N-terminus of a protein sequence to the C-terminus. So, if the fixed residues are all placed close to the N-terminus side of the protein (before the redesign part), you can pass it with structure as a new instruction. If fixed residues are placed close to the C-terminus side, those residues will not affect the design process of InstructPLM. Hope this can solve your question.