CHIEF: An Attention-based Ensemble Learning Framework for Functional Protein Inverse Folding

Abstract

This archive contains the source code for CHIEF, an attention-based ensemble learning framework for protein inverse folding. CHIEF integrates residue-level amino-acid predictions from multiple pretrained inverse folding models, including ProteinMPNN-vanilla, ProteinMPNN-soluble, ESM-IF, Frame2seq, and PiFold, using a trainable Transformer-based ensemble module.