ConforFlux: Particle-Guided Trunk Repulsion for Diverse Protein Conformations
- 1University of Tsukuba
Abstract
Deep learning models for protein structure prediction achieve near-experimental accuracy on individual folds, yet their default inference samples concentrate around a single dominant conformation. We introduce ConforFlux, an inference-time procedure for Boltz-2 that couples parallel structure-prediction trajectories through a pairwise C-RMSD repulsion gradient on the trunk's single and pair embeddings. The updated trunk conditions every subsequent denoising step without retraining the model. On four conformational-change categories, ConforFlux improves the per-state success rate over default Boltz-2 by 3-17 percentage points, with increased clashscores. The procedure also improves coverage on OpenFold3 and Protenix. On targets whose two reference structures were released after each model's training cutoff, guidance improves coverage of both states on all three models. On the human dopamine transporter, ConforFlux samples inward conformations absent from 500 default samples.
Related
- Designed nanofiber immunogens prime antibody responsesto diverse conformations of the HIV-1 fusion peptide
- Filtering Disordered Protein Conformations by NMR Residual Dipolar Couplings
- Systematic discovery of protein kinase-like domains reveals diverse evolutionary strategies in the human oral microbiome
- A diverse family of protein antibiotics inhibits the BAM complex by β-signal mimicry
- The actin-binding protein drebrin interacts with the respiratory syncytial virus matrix protein and promotes virus particle release.
BibTeX
@misc{suzuki2026conforflux,
title = {{ConforFlux: Particle-Guided Trunk Repulsion for Diverse Protein Conformations}},
author = {S. Suzuki and T. Amagasa},
howpublished = {bioRxiv},
year = {2026},
month = may,
doi = {10.64898/2026.05.16.725138},
url = {https://www.biorxiv.org/content/10.64898/2026.05.16.725138v2},
note = {CC BY},
}
RIS
TY - UNPB
TI - ConforFlux: Particle-Guided Trunk Repulsion for Diverse Protein Conformations
AU - S. Suzuki
AU - T. Amagasa
PY - 2026
DA - 2026/05/16
PB - bioRxiv
DO - 10.64898/2026.05.16.725138
UR - https://www.biorxiv.org/content/10.64898/2026.05.16.725138v2
AB - Deep learning models for protein structure prediction achieve near-experimental accuracy on individual folds, yet their default inference samples concentrate around a single dominant conformation. We introduce ConforFlux, an inference-time procedure for Boltz-2 that couples parallel structure-prediction trajectories through a pairwise C-RMSD repulsion gradient on the trunk's single and pair embeddings. The updated trunk conditions every subsequent denoising step without retraining the model. On four conformational-change categories, ConforFlux improves the per-state success rate over default Boltz-2 by 3-17 percentage points, with increased clashscores. The procedure also improves coverage on OpenFold3 and Protenix. On targets whose two reference structures were released after each model's training cutoff, guidance improves coverage of both states on all three models. On the human dopamine transporter, ConforFlux samples inward conformations absent from 500 default samples.
ER -