Research map / Image generation / Autoregressive and latent generation
GOLD: Geometric Optimized Latent Diffusion for Structure-Aware RNA Inverse Folding
Qi Si, Xuyang Liu, Penglei Wang, Shuo Su, LIMEI HAN, Mengzhu Li, Xin Guo, Yuan Qi, Yuan Cheng
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes GOLD, a geometric latent diffusion model for RNA inverse folding that explicitly incorporates continuous side-chain orientation geometry alongside sequence generation.
Problem
RNA inverse folding (designing sequences that fold into a target 3D structure) typically conditions on backbone coordinates but generates only discrete 1D sequences, ignoring continuous side-chain orientations that matter for structural stability.
Approach
Explicitly introduces side-chain geometry into RNA inverse folding and uses latent diffusion over a geometry-aware latent space so that generation is no longer limited to discrete sequences alone. Details beyond the abstract are not available.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add joint sequence and side-chain torsion modeling (SO(2) fusion + geometric losses) (ablation) | GOLD · GOLD Benchmark, no guidance | sequence recovery: 0.5897 -> 0.6249 | Improved |
| add dual-guided sampling (SS pairing + clash gradients) | GOLD (Base) · GOLD Benchmark | SS recovery: 0.7785 -> 0.8186 | Improved |
| add dual-guided sampling | GOLD (Base) · GOLD Benchmark | AlphaFold3 RMSD: 13.26 -> 12.31 | Improved |
| add dual-guided sampling | GOLD (Base) · CASP15 | sequence recovery: 0.6187 -> 0.6624 | Improved |
| remove geometric coordinate loss L_coord from GeoVAE (ablation) | GeoVAE · GOLD Benchmark | coordinate MSE: 0.0287 -> 0.1611 | Improved |
| remove angle loss L_angle from GeoVAE (ablation) | GeoVAE · GOLD Benchmark | angle MSE: 0.0109 -> 0.0546 | Improved |
| add RNA-FM alignment instead of none (ablation) | GOLD · GOLD Benchmark | sequence recovery: 0.6217 -> 0.6380 | Improved |
| use RNA-FM instead of Evo2 for alignment (ablation) | GOLD · GOLD Benchmark | SS recovery: 0.7922 -> 0.8186 | Improved |
| add clash guidance only with weight 1.0 (ablation) | GOLD · GOLD Benchmark native outputs | sequence recovery: 0.6378 (0.1) -> 0.6269 (1.0) | Hurt |
| add SS guidance only (ablation) | GOLD · GOLD Benchmark native outputs | clash score: 4.51 -> 4.32 | No clear effect |
| add clash guidance only (weight 0.5) (ablation) | GOLD · GOLD Benchmark native outputs | clash score: 4.51 -> 2.28 | Improved |
| add both SS and clash guidance (1.0, 0.1) instead of SS only (ablation) | GOLD · GOLD Benchmark native outputs | clash score: 4.32 -> 2.67 | Improved |
Closest papers by meaning
- Beyond Sequence: Impact of Geometric Context for RNA Property Prediction (ICLR 2025 · similarity 0.90)
- gRNAde: Geometric Deep Learning for 3D RNA inverse design (ICLR 2025 · similarity 0.89)
- RNAFlow: RNA Structure & Sequence Design via Inverse Folding-Based Flow Matching (ICML 2024 · similarity 0.89)
- TANGO: RNA Topology and Geometry Co-Design (NeurIPS 2026 · similarity 0.88)
- Equivariant Blurring Diffusion for Hierarchical Molecular Conformer Generation (NeurIPS 2024 · similarity 0.86)
- ManifoldCache: Training-Free Diffusion Acceleration via Constraint Manifold Caching (NeurIPS 2026 · 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.
