Inverse FoldDir: Structure-conditioned Protein Sequence Design by Dirichlet Flow Matching
Protein engineering has important implications in the bioeconomy, enabling applications in materials, medicine, and energy. A key challenge is designing protein sequences that have a specific form and function. Protein inverse folding seeks to address this challenge by identifying amino acid sequences compatible with a desired protein backbone. This task is central to protein redesign and can…
Protein engineering plays a pivotal role in the bioeconomy, allowing for applications in materials, medicine, and energy sectors. One significant challenge in this field is designing protein sequences with a particular structure and function. Protein inverse folding aims to tackle this challenge by identifying amino acid sequences that correspond to a desired protein backbone.
This task is crucial for protein redesign and can provide a method for generating sequences from de novo backbones generated by structure-generation techniques. Ideally, inverse folding can yield diverse sequence alternatives, preserve specific residues or motifs, account for soft biochemical preferences at certain positions, and generate candidates that remain experimentally viable.
The researchers have developed a controllable inverse-folding method called Inverse FoldDir, which achieves this through iterative denoising of the amino acid probability simplex. By providing a backbone structure, the model updates all positions simultaneously through a learned Dirichlet flow, enabling full sequence generation, fixed-residue inpainting, and user-defined soft residue priors.
On the test set, Inverse FoldDir achieved a mean TM-score of 84.5 and a mean C RMSD of 1.76 Å, outperforming the ESM-IF1 baseline with scores of 83.3 and 1.86 Å, respectively. The denoising trajectory analyses revealed that different positions commit at varying rates, with some residues changing identity later in the generation process, indicating a whole-sequence refinement approach rather than one-shot prediction or irreversible sequential decoding.
The researchers tested Inverse FoldDir in an anti-GFP nanobody redesign task, where two out of 35 redesigned sequences retained reproducible sfGFP-binding signal across independent assay runs, demonstrating approximately 43% sequence divergence from the native nanobody. In conclusion, Inverse FoldDir is a structure-conditioned protein redesign method that integrates structural recovery, user control, experimental validation, and a natural progression towards future property-guided sampling.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
