[GSoC 2026] Add Python PSI-MI TAB v2.7 parser, CSV support and validation layer for openPIP 2.0

[Abhishek-Kumar-Rai5]

Companion to issue #96.

The existing data-upload/uploader.py is a Selenium script that
submits files through the live website UI — it contains no parsing
logic. requirements.txt had only selenium. This PR adds the
missing parsing, validation, and data modeling layer to data-upload/.

---

Files Added

models.py

Python dataclasses mapped field-by-field to the openpip.sql schema.

Dataclass	DB table
Protein	protein
Organism	organism
Dataset	dataset
InteractionCategory	interaction_category
ParsedInteraction	interaction + junction tables

Every field name matches the actual DB column name from openpip.sql.

mitab_parser.py

Full PSI-MI TAB v2.7 parser. Output maps directly to models.py.

• All 42 PSI-MI TAB v2.7 columns handled
• db:value(description) format with pipe-separated multiples
• Quoted PSI-MI ontology terms e.g. psi-mi:"MI:0018"(two hybrid)
• Files with fewer than 42 columns padded gracefully
• UniProt, Ensembl, Entrez → correct protein table columns
• Gene name from alias field value where description = "gene name"
• Taxon ID + common name → organism table
• Score → interaction.score capped to varchar(10) per schema
• PubMed ID → dataset.pubmed_id
• Interaction type → interaction_category.category_name

Two entry points:

• parse_mitab27(filepath) — reads from disk
• parse_mitab27_from_string(content) — reads raw string content,
  for FastAPI endpoints receiving uploaded files in memory

csv_parser.py

New format support per the GSoC project goals. Minimum required
columns: protein_a, protein_b. Optional: interaction_type,
score, publication, author, dataset, year.

Normalizes to the same ParsedInteraction model as the PSI-MI TAB
parser. Both formats feed the same validation and insertion path.
Auto-detects UniProt accession vs gene name and sets correct fields.

validator.py

Per-row validation before DB insertion.

Errors (block insertion):

• Interactor A or B has no identifier of any kind

Warnings (logged, insertion proceeds):

• No UniProt ID — UniProt REST annotation will be skipped
• Score present but not numeric

tests/

34 tests, all passing.

pytest tests/test_parser.py -v
# 34 passed in 0.67s

Covers: field parsing, multi-value fields, quoted PSI-MI terms,
UniProt/Ensembl/gene name extraction, score handling, organism taxid,
interaction category, dataset pubmed mapping, CSV parsing, missing
column detection, validator errors and warnings.

requirements.txt

Added pytest.

---

Core Design Decision

Both PSI-MI TAB and CSV normalize to the same ParsedInteraction
dataclass. Validation and DB insertion are written once regardless
of input format. Adding new formats later only requires a new
normalizer — the rest of the pipeline is unchanged.

---

Testing

cd data-upload
pip install pytest
pytest tests/test_parser.py -v

[Abhishek-Kumar-Rai5]

@MoHelmy @gbader Please look into this and give your suggestions and feedback on this. Your further guidance will be highly appreciated.