AI Paper Atlas

Research map / Probabilistic and scientific modeling / Molecules and proteins

Demystifying Multimodal Biomolecular Co-design With Intrinsic Geodesic Coupling

Keyue Qiu, Xintong Wang, Zhilong Zhang, Hao Zhou, Wei-Ying Ma

ICML 2026 · Probabilistic and scientific modeling › Molecules and proteins · approach: Generative protein structure design · 0 citations (OpenAlex)

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
learn temporal coupling during training instead of random coupling with post-hoc selection (MolPilot)MolCRAFT-style BFN · CrossDocked SBDDVina Score (avg): -6.88 -> -7.16Improved
learn temporal coupling instead of synchronous coupling (MolCRAFT)BFN · CrossDocked SBDDVina Dock (avg): -7.67 -> -8.32Improved
learn coupling vs MolPilot on PoseBusters validityBFN · CrossDocked SBDDPB-Valid: 95.9% -> 94.3%No clear effect
use random coupling instead of synchronousTargetDiff · CrossDocked SBDDVina Dock (avg): -7.80 -> -8.04Improved
use random coupling instead of synchronousTargetDiff · CrossDocked SBDDPB-Valid: 50.5% -> 58.1%Hurt
freeze a-priori optimal schedule instead of adaptive bilevel optimization (ablation)GeoCoupling · CrossDocked SBDD (1000 samples)Vina Score mean: -6.97 -> -7.12Improved
add adaptive bilevel optimization over fixed schedule (ablation)GeoCoupling · CrossDocked SBDD (1000 samples)Connected (%): 91.1 -> 93.5Improved
add EMA smoothing of learned schedule (ablation)GeoCoupling · CrossDocked SBDD (1000 samples)Vina Score mean: -6.50 -> -7.12Improved
add EMA smoothing of learned schedule (ablation)GeoCoupling · CrossDocked SBDD (1000 samples)QED: 0.58 -> 0.54No clear effect
train with learned coupling instead of applying it post-hoc to MultiFlow (ablation)GeoCoupling · Unconditional protein co-designco-designability: 0.74 -> 0.79Improved
apply learned coupling post-hoc to MultiFlow checkpointMultiFlow · Unconditional protein co-designFS clusters: 0.56 -> 0.73Improved
use learned coupling for longer proteinsGeoCoupling vs MultiFlow · Protein length 500co-designability: < 0.3 -> > 0.6Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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