De novo structural elucidation on MassSpecGym (test)
32Top-1 AccuracyMSFlow
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MSFlowImplementation=Our implementation, Conditioning=CDDD embeddings2026.02 | 32 | 6.38 | 0.69 | 42.53 | 5.25 | 0.74 | |
| MIST + CDDD decoderImplementation=Our implementation, Conditioning=CDDD embeddings2026.02 | 24.1 | 13.4 | 0.56 | — | — | — | |
| MSFlow (ECFP)Implementation=Our implementation, Conditioning=Morgan fingerprints (ECFP)2026.02 | 5.17 | 17.01 | 0.31 | 6.71 | 13.87 | 0.41 | |
| DiffMS2025.02 | 2.3 | 18.45 | 0.28 | 4.25 | 14.73 | 0.39 | |
| DiffMSEvaluation setting=Spectral training, Status=SOTA2026.01 | 2.3 | — | 0.28 | 4.25 | — | 0.39 | |
| DiffMSSource=Bohde et al.2026.02 | 2.3 | 18.45 | 0.28 | 4.25 | 14.73 | 0.39 | |
| MADGEN2025.02 | 1.31 | 27.47 | 0.2 | 1.54 | 16.84 | 0.26 | |
| MADGENSource=Bohde et al.2026.02 | 1.31 | 27.47 | 0.2 | 1.54 | 16.84 | 0.26 | |
| MS-BARTSource=Han et al.2026.02 | 1.07 | 16.47 | 0.23 | 1.11 | 15.12 | 0.28 | |
| MADGENEvaluation setting=Spectral training2026.01 | 0.8 | — | — | 1.6 | — | — | |
| SMILES Transformersource=reproduced from MassSpecGym2025.02 | 0 | 79.39 | 0.03 | 0 | 52.13 | 0.1 | |
| MIST + MSNovelistimplementation=re-implemented by current paper2025.02 | 0 | 45.55 | 0.06 | 0 | 30.13 | 0.15 | |
| SELFIES Transformersource=reproduced from MassSpecGym2025.02 | 0 | 38.88 | 0.08 | 0 | 26.87 | 0.13 | |
| Spec2Molimplementation=re-implemented by current paper2025.02 | 0 | 37.76 | 0.12 | 0 | 29.4 | 0.16 | |
| MIST + Neuraldecipherimplementation=new baseline introduced in this paper2025.02 | 0 | 33.19 | 0.14 | 0 | 31.89 | 0.16 | |
| Random Generationsource=reproduced from MassSpecGym2025.02 | 0 | 21.11 | 0.08 | 0 | 18.26 | 0.11 | |
| SMILES TransformerEvaluation setting=Spectral training2026.01 | 0 | — | 0.03 | 0 | — | 0.1 | |
| MIST + MSNovelistEvaluation setting=Spectral training2026.01 | 0 | — | 0.06 | 0 | — | 0.15 | |
| SELFIES TransformerEvaluation setting=Spectral training2026.01 | 0 | — | 0.08 | 0 | — | 0.13 | |
| Spec2MolEvaluation setting=Spectral training2026.01 | 0 | — | 0.12 | 0 | — | 0.16 | |
| MIST + NeuralDecipherEvaluation setting=Spectral training2026.01 | 0 | — | 0.14 | 0 | — | 0.16 | |
| Random GenerationEvaluation setting=MassSpecGym baseline2026.01 | 0 | — | 0.08 | 0 | — | 0.11 | |
| Claude-3.5-SonnetEvaluation setting=Zero-shot, Prompting strategy=CoT-guided2026.01 | 0 | — | 0.09 | 0 | — | 0.1 | |
| GPT-4o-miniEvaluation setting=Zero-shot, Prompting strategy=CoT-guided2026.01 | 0 | — | 0.09 | 0 | — | 0.12 | |
| Llama-3-8BEvaluation setting=Zero-shot, Prompting strategy=CoT-guided2026.01 | 0 | — | 0.09 | 0 | — | 0.1 | |
| Llama-3-70BEvaluation setting=Zero-shot, Prompting strategy=CoT-guided2026.01 | 0 | — | 0.1 | 0 | — | 0.12 | |
| SMILES TransformerSource=Bohde et al.2026.02 | 0 | 79.39 | 0.03 | 0 | 52.13 | 0.1 | |
| MIST + MSNovelistSource=Bohde et al.2026.02 | 0 | 45.55 | 0.06 | 0 | 30.13 | 0.15 | |
| SELFIES TransformerSource=Bohde et al.2026.02 | 0 | 38.88 | 0.08 | 0 | 26.87 | 0.13 | |
| Spec2MolSource=Bohde et al.2026.02 | 0 | 37.76 | 0.12 | 0 | 29.4 | 0.16 | |
| MIST + NeuraldecipherSource=Bohde et al.2026.02 | 0 | 33.19 | 0.14 | 0 | 31.89 | 0.16 | |
| Random GenerationSource=Bohde et al.2026.02 | 0 | 21.11 | 0.08 | 0 | 18.26 | 0.11 |