防御800结构

# dg800structure Protenix structure predictions for 850 prokaryotic proteins from `prok_selected_dg.faa`. ## Source - Input FASTA: `prok_selected_dg.faa` - Model: `protenix_mini_esm_v0.5.0` - Seeds / samples: seed=101, sample=1 - Format: mmCIF (`.cif`) ## Contents - `structures/`: one predicted structure per protein (`*_sample_0.cif`) - `README.md`: dataset summary and QC statistics ## QC summary | Metric | Min | Max | Mean | Median | |--------|-----|-----|------|--------| | pLDDT | 33.46 | 94.06 | 81.02 | 81.85 | | pTM | 0.232 | 0.967 | 0.816 | 0.877 | | gPDE | 0.287 | 4.053 | 0.691 | 0.592 | ### pLDDT tiers | Tier | Threshold | Count | |------|-----------|-------| | High confidence | pLDDT >= 90 | 111 | | Medium confidence | 70 <= pLDDT < 90 | 672 | | Low confidence | pLDDT < 70 | 67 | ## Confidence fields Each structure was predicted with Protenix mini-ESM. Summary metrics above were computed from per-protein confidence JSON files: - `plddt`: predicted local distance difference test (0-100, higher is better) - `ptm`: predicted TM-score - `gpde`: global predicted distance error - `ranking_score`: model ranking score - `has_clash`: whether the predicted structure has clashes Atom-level pLDDT values are stored in the mmCIF B-factor column. ## Usage ```python from modelscope.msdatasets import MsDataset ds = MsDataset.load('lyndons/dg800structure') ```