InstaNovo-FM training corpus
training · DDA · 4 versions
| Kind | training |
| Acquisition | DDA |
| Organisms | mixed |
| Home | https://huggingface.co/datasets/InstaDeepAI/InstaNovo |
Tandem mass spectra with PSM labels, uniformly reprocessed from public PRIDE submissions to pretrain and evaluate InstaNovo-FM. The largest labelled corpus published in this field.
Versions
LCFM (low confidence)
The broadest and least stringently filtered labelled tier, from which the stricter ones are derived. “Low” is relative to the other tiers, not absolute: every labelled tier is high-confidence PSMs.
released 2026-09-03 · introduced by InstaNovo enables diffusion-powered de novo peptide sequencing in large-scale proteomics experiments
Where it lives:
- Hugging Face · InstaDeepAI/InstaNovo#lcfm splits/lcfm and by_project/lcfm
MCFM (medium confidence)
A stricter PSM filter than LCFM, and a superset of HCFM.
released 2026-09-03 · introduced by InstaNovo enables diffusion-powered de novo peptide sequencing in large-scale proteomics experiments
Where it lives:
- Hugging Face · InstaDeepAI/InstaNovo#mcfm splits/mcfm and by_project/mcfm
HCFM (high confidence)
The most stringently filtered tier, a subset of both others.
released 2026-09-03 · introduced by InstaNovo enables diffusion-powered de novo peptide sequencing in large-scale proteomics experiments
Where it lives:
- Hugging Face · InstaDeepAI/InstaNovo#hcfm splits/hcfm and by_project/hcfm
ACFM (all confidence)
The unlabelled all-confidence superset the three labelled tiers are filtered out of. Not published as a processed tier, but reproducible: the InstaNovo-FM paper lists the raw accessions it was built from, so rebuilding it means reprocessing those rather than downloading it.
released 2026-09-03 · introduced by InstaNovo enables diffusion-powered de novo peptide sequencing in large-scale proteomics experiments
No public address: this version is named in the literature but cannot be downloaded.