01_data_prep.ipynb

01_data_prep.ipynb — Preparing GDSC IC₅₀ and metadata

Objective

Create clean, analysis-ready tables that connect GDSC2 IC₅₀ values with COSMIC/DepMap cell-line IDs and compound SMILES strings.

---

Inputs

File	Drive location	Notes
GDSC2_fitted_dose_response_27Oct23.xlsx	GDSC2_drugsens/	IC₅₀ (µM) for 228 drugs × 987 cell lines
screened_compounds_rel_8.5.csv	GDSC2_drugsens/	Drug names, IDs, SMILES
sample_info.csv	CCLE-DepMap22Q2_geneexp/	CCLE sample ↔ DepMap IDs

---

Processing steps ↴

1. Load IC₅₀ sheet and cast to float32.

2. Map COSMIC_ID → DepMap_ID using sample_info.csv; drop rows without a match (987 → 676 cell lines).

3. Filter compounds: keep only 228 drugs with valid canonical SMILES (verified via RDKit).

4. Long-format table: reshape the drug-cell line matrix into a three-column table (DrugID, CellLineID, lnIC50).

5. Save cleaned artefacts into datasets/:

   • drug_info.csv — 228 compounds, SMILES, GDSC IDs

   • drug_name.txt — plain list of compound names for plotting

   • cell_line_info.csv — created but not reused later

   • ic50_cleaned.csv — long-format ln(IC₅₀)

---

Outputs that feed the next notebook

File	Consumed by
drug_info.csv	02_gen_dataset.ipynb
ic50_cleaned.csv	02_gen_dataset.ipynb

---

Rationale

• Aligning to DepMap IDs up front guarantees all downstream joins (expression, fingerprints) key on a single identifier.

• Validating SMILES now avoids extra API calls later when generating Morgan fingerprints.

• Reshaping to long format makes it easy to apply a ln(IC₅₀) threshold and to pivot into classification and regression matrices.

Intermediate CSVs not read again (cell_line_info.csv, etc.) are kept solely for provenance.