Research map / Image generation / Fast diffusion sampling
FRIGID: Scaling Diffusion-Based Molecular Generation from Mass Spectra at Training and Inference Time
Montgomery Bohde, Hongxuan Liu, Mrunali Manjrekar, Magdalena Lederbauer, Shuiwang Ji, Runzhong Wang, Connor W. Coley
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 10 rounds of ICEBERG-guided refinement (ablation) | FRIGID-base · NPLIB1 | Top-1 accuracy: 13.31% -> 14.16% | Improved |
| add 10 rounds of ICEBERG-guided refinement (ablation) | FRIGID-base · NPLIB1 | Top-10 accuracy: 20.15% -> 25.27% | Improved |
| add 10 rounds of ICEBERG-guided refinement (ablation) | FRIGID-base · MassSpecGym | Top-1 accuracy: 8.21% -> 9.05% | Improved |
| add 10 rounds of ICEBERG-guided refinement (ablation) | FRIGID-base · MassSpecGym | Top-10 accuracy: 9.82% -> 12.23% | Improved |
| apply ICEBERG-guided refinement to DiffMS (ablation) | DiffMS · NPLIB1 | Top-1 accuracy: 8.34% -> 13.33% | Improved |
| apply ICEBERG-guided refinement to DiffMS (latency) (ablation) | DiffMS · NPLIB1 | inference compute (sec./sample): 131.21 -> 1423.30 | Hurt |
| condition directly on spectra instead of MIST fingerprint (E2E FT-Base) (ablation) | FRIGID-base · NPLIB1 | Top-1 accuracy: 13.31% -> 5.13% | Hurt |
| condition directly on spectra instead of MIST fingerprint (E2E FT) (ablation) | FRIGID · NPLIB1 | Top-1 accuracy: 14.16% -> 7.69% | Hurt |
| replace known formula with top-5 MIST-CF predicted formulae (fixed FP) (ablation) | FRIGID · NPLIB1 | Top-1 accuracy: 14.16% -> 12.33% | Hurt |
| vary p_max from 0.75 to 1.00 (ablation) | FRIGID refinement · NPLIB1 | Top-1 accuracy: 14.16% -> 15.02% | No clear effect |
| raise hallucination penalty gamma from 1 to 5 (ablation) | FRIGID refinement · NPLIB1 | Top-1 accuracy: 14.16% -> 13.80% | No clear effect |
| lower p_max from 0.75 to 0.50 (ablation) | FRIGID refinement · NPLIB1 | Top-10 accuracy: 25.27% -> 23.93% | No clear effect |
Closest papers by meaning
- DiffMS: Diffusion Generation of Molecules Conditioned on Mass Spectra (ICML 2025 · similarity 0.90)
- MS-BART: Unified Modeling of Mass Spectra and Molecules for Structure Elucidation (NeurIPS 2025 · similarity 0.88)
- MassSpecGym: A benchmark for the discovery and identification of molecules (NeurIPS 2024 · similarity 0.87)
- GLACIER: Rethinking Mass Spectrum Prediction as an Object Detection Problem (NeurIPS 2026 · similarity 0.87)
- Atomic Diffusion Models for Small Molecule Structure Elucidation from NMR Spectra (NeurIPS 2025 · similarity 0.86)
- MADGEN: Mass-Spec attends to De Novo Molecular generation (ICLR 2025 · 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.
