AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Efficient Prediction of Large Protein Complexes via Subunit-Guided Hierarchical Refinement

Chixiang Lu, Yunhua Zhong, Shikang Liang, Xiaojuan Qi, Haibo Jiang

ICLR 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Generative protein structure design

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use hierarchical subunit-guided refinement instead of end-to-endAlphaFold3 · Recent PDBDockQ success rate Top-1 (%): 70.4 -> 69.0No clear effect
add multi-chain context refinement over pairwise assemblyCombFold+AF3 · Recent PDBDockQ success rate Top-1 (%): 43.2 -> 69.0Improved
use HierAFold on complexes beyond memory limitAlphaFold3 · Large complexes (>5k tokens)DockQ success rate Top-1 (%): OOM -> 43.9Improved
use full diffusion model for coarse prediction instead of distilled (ablation)HierAFold · Recent PDBDockQ success rate Oracle (%): 73.1 -> 73.3; time 46 -> 125 minNo clear effect
use mini-rollout coarse prediction instead of distilled model (ablation)HierAFold · Recent PDBDockQ success rate Oracle (%): 73.1 -> 71.0Hurt
use residue-level segmentation (tau_split=0) (ablation)HierAFold · Recent PDBDockQ success rate Top-1 (%): 69.0 -> 67.8Hurt
vary tau_split between 5 and 20 (ablation)HierAFold · Recent PDBDockQ success rate Top-1 (%): 68.9 to 69.1No clear effect
use PDE instead of PAE for segmentation (ablation)HierAFold · Recent PDBDockQ success rate Top-1 (%): 69.0 -> 68.2Hurt
combine PAE and distance criteria for subunit selection (vs PAE-only) (ablation)HierAFold · Recent PDBDockQ success rate Top-1 (%): 66.9 -> 69.0Improved
combine PAE and distance criteria for subunit selection (vs distance-only) (ablation)HierAFold · Recent PDBDockQ success rate Top-1 (%): 66.3 -> 69.0Improved
loosen selection thresholds (tau_p=10, tau_d=30) (ablation)HierAFold · Recent PDBDockQ success rate Top-1 (%): 69.0 -> 69.3; time 46 -> 50 minNo clear effect
use full model rather than distilled model in the fine stage (ablation)HierAFold · Recent PDBDockQ success rate Top-1 (%): 64.9 -> 69.0Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy