PLM Subnetworks

A framework to study structural knowledge factorization within Protein Language Models (PLMs). It uses trainable subnetworks to isolate model weights corresponding to specific protein structural categories (CATH taxonomy) for improved structure prediction.