ModCREDB

ModCREDB is a human transcription-factor database combining PWM evidence, protein–DNA structural models, domain/family annotations, and a FIMO DNA scanner. The web application is a small Python/Jinja2 service backed by SQLite.

The complete implementation and release workflow are documented in docs/MODCREDB_TECHNICAL_GUIDE.md.

Local release database

The publication-final working database is intentionally not stored in Git:

data/tf_webdb_local_staging.sqlite

Its corresponding model cache is also local:

/home/patricia/TF_database_Baldo_data/modcredb_model_cache

Set TF_WEBDB_MODEL_CACHE_DIR when starting the application so active PDBs are served from the extracted cache rather than repeatedly decompressed from their archives.

cd /home/patricia/tf_webdb
export TF_WEBDB_MODEL_CACHE_DIR=/home/patricia/TF_database_Baldo_data/modcredb_model_cache
python3 app.py --host 127.0.0.1 --port 8094 \
  --db data/tf_webdb_local_staging.sqlite

Open http://127.0.0.1:8094/ locally, or use an SSH tunnel from another machine.

Runtime dependencies

python3 -m pip install -r requirements.txt

The /scan route additionally requires MEME Suite/FIMO in PATH. Structural audit scripts that parse mmCIF use the separately documented audit dependency:

python3 -m pip install -r requirements-audit.txt

Important data rules

• Original Baldo/manuscript TSVs and model archives are immutable source data.
• Primary PWM annotation follows the evidence hierarchy documented in the technical guide; later verified direct evidence may legitimately promote a TF without deleting its supporting predicted motifs/models.
• A model is not considered DNA-interacting merely because protein and DNA are present. The stored interface QC uses explicit protein–DNA atom contacts.
• The 108 curated accessions with PWM evidence but no validated DNA-bound structure are represented explicitly; the label is never inferred merely from a zero model count.
• Public PDB links use the active stable structure_file.model_id: /model?model=<model_id>. Numeric row IDs are internal and deliberately not accepted as public model identifiers after a database rebuild.

Safety

Database-changing scripts default to dry-run where implemented and validate their frozen inputs. Read each script's --help, create a verified backup, and validate on a staging copy before applying it. Production databases, large model archives, extracted caches, backups, and generated reports are excluded from Git.