Research map / Graph learning / Graph self-supervised learning
Learning Adaptive Topology with FiLM-Guided Distillation for Tertiary Structure-Based RNA Design
Zixun Zhang, Yuncheng Jiang, Yuzhe Zhou, Jiayou Zheng, Shuguang Cui, Zhen Li
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add vanilla feature-alignment KD from RNA-FM teacher (ablation) | MPNN baseline · RDesign split | recovery rate (%): 41.54 -> 41.59 | No clear effect |
| replace vanilla KD with FiLM-guided distillation (ablation) | MPNN baseline (no ATL) · RDesign split | recovery rate (%): 41.59 -> 43.68 | Improved |
| add Adaptive Topology Learning alone (ablation) | MPNN baseline · RDesign split | recovery rate (%): 41.54 -> 44.70 | Improved |
| add ATL on top of FGD (ablation) | FGD model · RDesign split | recovery rate (%): 43.68 -> 45.49 | Improved |
| add FGD on top of ATL (ablation) | ATL model · RDesign split | recovery rate (%): 44.70 -> 45.49 | Improved |
| add FGD and ATL together to the baseline (ablation) | MPNN baseline · RDesign split | recovery rate (%): 41.54 -> 45.49 | Improved |
| add FGD and ATL together to the baseline (ablation) | MPNN baseline · RDesign split | Macro F1 (x100): 40.50 -> 44.42 | Improved |
| increase kNN k from 10 to 60 in fixed graph, then beyond 30 (ablation) | fixed kNN baseline · RDesign split | recovery rate (%): 41.40% -> 42.87% | Mixed |
| use ATL instead of fixed kNN at k=32 (ablation) | ATL-FGD · RDesign split | recovery rate (%): 44.86% at k=32, +2.44% over fixed topology | Improved |
| increase candidate graph size k up to 128 with ATL (ablation) | ATL-FGD · RDesign split | recovery rate (%): 44.86% (k=32) -> 46.28% (k=128) | Improved |
| enlarge ATL-only candidate graph from k=96 to k=256 (ablation) | ATL only · RDesign split | recovery rate (%): 45.72 -> 44.80 | Hurt |
| enlarge ATL-only candidate graph from k=96 to k=512 (ablation) | ATL only · RDesign split | recovery rate (%): 45.72 -> 44.83 | Hurt |
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.
- gRNAde — gRNAde: Geometric Deep Learning for 3D RNA inverse design (ICLR 2025) · better on 5 of 5 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
- TANGO: RNA Topology and Geometry Co-Design (NeurIPS 2026 · similarity 0.90)
- RDesign: Hierarchical Data-efficient Representation Learning for Tertiary Structure-based RNA Design (ICLR 2024 · similarity 0.90)
- gRNAde: Geometric Deep Learning for 3D RNA inverse design (ICLR 2025 · similarity 0.89)
- Beyond Sequence: Impact of Geometric Context for RNA Property Prediction (ICLR 2025 · similarity 0.89)
- RNAFlow: RNA Structure & Sequence Design via Inverse Folding-Based Flow Matching (ICML 2024 · similarity 0.88)
- Enforcing Constraints in RNA Secondary Structure Predictions: A Post-Processing Framework Based on the Assignment Problem (ICML 2024 · similarity 0.86)
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.
