MD-readiness assessment of PDB structures: supplementary code

Supplementary code for BioExcel-3 Deliverable D2.5, Final report on standards and recommendations for simulation, data generation, storage and mining.

Contents

Notebook	What it does
01_scoring_assemblies.ipynb	Extracts issue counts for assemblies, computes penalties and tiers, and ranks assemblies within each PDB entry
02_scoring_non_assemblies.ipynb	Does the same for non-assemblies, ranked across the whole collection
03_analysis_figures_tables.ipynb	Produces the figures and summary tables from the outputs of 01 and 02

Workflow

MongoDB (PDB_Checking)
   │  (or parquet cache when MongoDB is unavailable)
   ├── 01_scoring_assemblies ──────► data/assembly_md_readiness_ranked.parquet
   │                                 data/assembly_issue_config.json
   └── 02_scoring_non_assemblies ──► data/non_assembly_md_readiness_scored.parquet
                                     data/non_assembly_issue_config.json
                                            │
                    03_analysis_figures_tables ──► figures/*.png + tables

Run the notebooks in order: 01, then 02, then 03. Notebooks 01 and 02 are independent of each other. Notebook 03 needs the outputs of both.

Outputs of notebook 03

Section	Figure	Values printed below it
3. Issue prevalence	figures/issue_prevalence.png	% of structures with each issue
4. Tier distribution	figures/tier_distribution.png	Counts and % per tier
5. Penalty distribution	figures/penalty_density_log_ecdf.png, figures/penalty_density_raw.png	Penalty summary statistics
6. Penalty within tiers	figures/penalty_by_tier.png	Penalty statistics by tier
7. Penalty contribution by issue	figures/penalty_contribution.png	Share of the total penalty per issue

Other analyses in the report (ligand composition, experimental method, alternate locations, microheterogeneity, AlphaFold models) are not part of this package.

How to run

conda env create -f environment.yml
conda activate md-readiness
jupyter lab

Data source. Notebooks 01 and 02 first try to connect to MongoDB with the credentials in .env; copy .env.example to .env and fill it in.

• If the connection succeeds, the issue counts are extracted from the database and written to the parquet cache in data/.
• If it fails, the notebooks use the cached files data/assembly_issue_counts.parquet and data/non_assembly_issue_counts.parquet.
• Set USE_CACHE_ONLY = True in Section 1 to skip the database.

Consistency. Severity weights and remediation classes are defined once in notebooks 01 and 02 (Section 2), and each notebook saves them to data/*_issue_config.json. Notebook 03 reads those files, checks that the two configurations are identical, and verifies that the per-issue penalty terms add up to each structure's penalty.

Licence and citation

Code: Apache-2.0 · Data: CC-BY-4.0 · DOI: to be assigned