Datasets:

ClarusC64
/

protein_structure_uncertainty_auditor_v0.2

id stringclasses 10 values	sequence_length int64 95 520	predicted_confidence float64 0.48 0.93	disorder_fraction float64 0.04 0.55	binding_site_uncertainty stringclasses 3 values	experimental_support stringclasses 4 values	model_disagreement stringclasses 3 values	proposed_use stringclasses 5 values	gold_uncertainty_flag stringclasses 2 values	gold_uncertainty_type stringclasses 7 values	gold_recommendation stringclasses 10 values
PSU-001	312	0.92	0.05	low	cryo_em_support	low	drug_design	no	low_uncertainty	Proceed with structure guided design
PSU-002	148	0.61	0.38	high	no_experimental_data	medium	drug_design	yes	disorder_high	Validate with NMR or disorder analysis before use
PSU-003	402	0.75	0.12	medium	low_resolution_xray	high	mutation_mapping	yes	model_disagreement	Compare models and seek experimental validation
PSU-004	95	0.58	0.55	high	no_experimental_data	low	function_annotation	yes	high_disorder	Use sequence motifs and avoid rigid structure assumptions
PSU-005	520	0.88	0.07	low	cryo_em_support	low	docking	no	low_uncertainty	Proceed with docking cautiously
PSU-006	210	0.66	0.22	medium	no_experimental_data	high	drug_design	yes	confidence_borderline	Obtain additional modeling or experimental support
PSU-007	333	0.48	0.41	high	no_experimental_data	high	binding_analysis	yes	confidence_low	Do not rely on structure for binding inference
PSU-008	275	0.81	0.09	low	xray_support	medium	mutation_mapping	no	moderate_confidence	Proceed but monitor uncertain loops
PSU-009	180	0.57	0.33	medium	no_experimental_data	medium	function_annotation	yes	confidence_borderline	Use ensemble predictions and avoid single model reliance
PSU-010	430	0.93	0.04	low	cryo_em_support	low	drug_design	no	low_uncertainty	Proceed with structure driven pipeline

Protein Structure Uncertainty Auditor

Goal
Detect when predicted protein structures are too uncertain for downstream use.

Model must output

uncertainty_flag (yes/no)
uncertainty_type
recommendation

This dataset tests whether models can audit structural confidence before use in:

drug design
docking
mutation mapping
function inference

Run scorer

python scorer.py --predictions predictions.jsonl --test_csv data/test.csv

Downloads last month: 40

Number of rows:

Total file size:

8.93 kB