Xevacurate Tsao 2022 — PDX Drug-Response Curation

This repository curates Tsao Lab lung cancer drug-screen data into a reproducible XevaSet object using a Snakemake workflow, Python utilities, and an R step executed inside Docker (Bioconductor 3.20).

The pipeline scans & validates raw inputs, processes and updates annotations, builds Xeva input CSVs, calculate drug response statistics, and finally creates an .rds XevaSet.

---

Table of Contents

• What You Get
• Requirements
• Quick Start
• Inputs Expected
• Configuration
• Workflow Outline
• Outputs
• Load the XevaSet in R
• Troubleshooting
• Repo Layout
• Citation
• License

---

What You Get

• A fully reproducible Snakemake workflow orchestrated via pixi (Python env manager).
• Clean CSV inputs for Xeva: model.csv, drug.csv, experiment.csv, expDesign.csv, modToBiobaseMap.csv, and per-omics *_pdata.csv.
• An XevaSet saved as results/xevaset/<NAME>.rds (name configurable).
• Logs at each step under logs/.

---

Requirements

• Docker (Desktop) : https://www.docker.com/
• pixi (to provide Snakemake + Python deps): https://pixi.sh/
• Git (to clone this repo)
• Internet access during the R step (biomaRt Ensembl gene mapping & PubChem calls if enabled).

Windows users: Docker Desktop with WSL2 backend works fine with this workflow.

---

Quick Start

# 1) Clone
git clone git@github.com:BHKLAB-DataProcessing/Xevacurate_Snakemake_UHN_Lung.git
cd Xevacurate_Snakemake_UHN_Lung

# 2) Dry run
PYTHONPATH=src pixi run snakemake -n

# 3) Execute (adjust cores to taste)
PYTHONPATH=src pixi run snakemake --cores 4 --printshellcmds

The final XevaSet appears at:

results/xevaset/UHN_Tsao_Lung_DrugResponse_2022_v1.rds   # default, configurable

Inputs Expected

Place the following under data/input/:

data/
└── input/
    ├── drug_screen/                # raw drug-response inputs (time/volume etc.)
    ├── additional/
    │   ├── models_to_remove.xlsx
    │   ├── all_model_clinical.csv
    │   └── all_drug_mapping.xlsx
    └── omics/
        ├── data_expression_mRNA.txt   # log2 microarray intensity matrix (rows: genes, cols: samples)
        ├── Tsao_Lung_WES_mutation_2025.tsv  # WES mutation matrix
        └── data_CNA.tsv               # CNV matrix

Notes

• Gene IDs in omics matrices are expected as HGNC symbols (rows). The R script will annotate via biomaRt.

• Sample (column) IDs must match what *_pdata.csv files (generated during build) will contain. The R script ensures ordering & sanity checks.

Configuration

All knobs live in config.yaml. Minimal example:

docker:
  xeva_r_image: docker.io/gfeng2023/xeva-build-r:bioc-3.20  # public image with Xeva deps

xevaset:
  name: Tsao_Lung_2022
  rds_name: UHN_Tsao_Lung_DrugResponse_2022_v1.rds

Workflow Outline

1. scan → workflow/scripts/run_scan.py

   • Scans data/input/drug_screen and writes a manifest + QC info.
   • Output: results/scan/all_file_scan.json

2. process → workflow/scripts/run_process.py

   • Validates/normalizes raw data and emits a clean TSV + QC.
   • Outputs: results/process/processed_data.tsv, results/process/qc_report.json

3. update → workflow/scripts/run_update.py

   • Maps models/drugs, applies removal lists, updates batches.
   • Outputs:
     • results/update/update_model_data.tsv
     • results/update/update_model_drug_data.tsv
     • results/update/drug_filtered_out.tsv
     • results/update/update_model_drug_batch_data.tsv

4. build → workflow/scripts/run_build.py

   • Produces Xeva input CSVs:
     • results/build/model.csv
     • results/build/drug.csv
     • results/build/experiment.csv
     • results/build/expDesign.csv
     • results/build/modToBiobaseMap.csv
     • results/build/CNV_pdata.csv
     • results/build/mutation_pdata.csv
     • results/build/RNASeq_pdata.csv
   • Plus feature annotations created by the R step:
     • results/build/rna_fdata.csv
     • results/build/mutation_fdata.csv
     • results/build/cnv_fdata.csv

5. create_xevaset (Dockerized R) → workflow/scripts/create_xevaset.R

   • Runs inside the configured Docker image (Bioconductor 3.20).
   • Builds ExpressionSets, merges with experiment & design, computes responses, and writes the final xevaset as .rds file.

Outputs

results/
├── scan/
│   └── all_file_scan.json
├── process/
│   ├── processed_data.tsv
│   └── qc_report.json
├── update/
│   ├── update_model_data.tsv
│   ├── update_model_drug_data.tsv
│   ├── drug_filtered_out.tsv
│   └── update_model_drug_batch_data.tsv
├── build/
│   ├── model.csv
│   ├── drug.csv
│   ├── experiment.csv
│   ├── expDesign.csv
│   ├── modToBiobaseMap.csv
│   ├── CNV_pdata.csv
│   ├── mutation_pdata.csv
│   ├── RNASeq_pdata.csv
│   ├── rna_fdata.csv
│   ├── mutation_fdata.csv
│   └── cnv_fdata.csv
└── xevaset/
    └── UHN_Tsao_Lung_DrugResponse_2022_v1.rds   # configurable

Load the XevaSet in R

# From the repo root or any R session with Xeva installed:
xset <- readRDS("results/xevaset/UHN_Tsao_Lung_DrugResponse_2022_v1.rds")

# Inspect
xset
responseType(xset)       # Which response measures were computed (e.g., slope, AUC, angle, abc, mRECIST)
modelInfo(xset)[1:5,]
head(summarizeResponse(xset, by="model"))

Troubleshooting

• docker: command not found

  Install Docker Desktop (and enable WSL2 on Windows).

  Restart your shell/terminal.

• Bioconductor/CRAN packages fail to install

  The default image gfeng2023/xeva-build-r:bioc-3.20 already contains the required R/Bioc packages (e.g., Xeva, Biobase, biomaRt).

  If you build your own image, use Bioconductor 3.20 (or newer) and avoid packages unavailable in older Bioc releases (e.g., AnnotationGx is not on 3.19).

• Omics sample mismatches

  The R script performs sanity checks. Ensure column names of omics matrices match the *_pdata.csv IDs produced in the build step.

• Windows path quirks

  Always run from inside WSL2 (Ubuntu) and keep the repo on the WSL filesystem (e.g., under ~/).

  Bind-mounting Windows paths can be slower and occasionally tricky.

Repo Layout

.
├── config.yaml
├── src/                      # Python package(s): xevacurate_new/...
├── workflow/
│   ├── rules/                # Snakemake rules (*.smk)
│   │   └── curation.smk
│   ├── scripts/
│   │   ├── run_scan.py
│   │   ├── run_process.py
│   │   ├── run_update.py
│   │   ├── run_build.py
│   │   └── create_xevaset.R
│   └── Snakefile            # includes rules/curation.smk
├── data/
│   └── input/               # place your inputs here (see Inputs Expected)
├── results/                 # pipeline outputs
└── logs/                    # stepwise logs

Citation

If you use this workflow or the produced XevaSet in a publication, please cite:

• Xeva (Bioconductor)

  Mer A, Haibe-Kains B (2025). Xeva: Analysis of patient-derived xenograft (PDX) data. doi:10.18129/B9.bioc.Xeva.

  R package version 1.24.0. https://bioconductor.org/packages/Xeva.

• The Original Dataset/Source

  Weiss J, Pham NA, Pintilie M, Li M, Liu G et al (2022) Optimizing Drug Response Study Design in Patient-Derived Tumor Xenografts. doi.org/10.1177/11769351221136056.

License

MIT

Author

Guanqiao Feng

Contributions

Contributions are welcome!