Distributed univariable predictive modelling for Immuno-Oncology response

Authors: Farnoosh Abbas Aghababazadeh, Nasim Bondar Sahebi

Contact: farnoosh.abbasaghababazadeh@uhn.ca, nasim.bondarsahebi@uhn.ca

Description: A distributed framework for univariable predictive modeling of Immuno-Oncology (IO) response. This pipeline enables center-specific analyses while preserving data privacy and provides a central integration workflow for federated meta-analysis.

---

Project Overview

This repository implements a distributed framework for evaluating the predictive value of RNA-based signatures in response to IO therapies. It supports:

• Per-center analysis: signature scoring and univariable association with outcomes (OS, PFS, response)
• Privacy preservation: no sharing of raw or patient-level data
• Central aggregation: meta-analysis of effect sizes across centers
• Reproducibility: modular pipeline using Pixi, Nextflow, and R

---

Code Ocean Capsules

This workflow is also available as reproducible Code Ocean capsules:

• Local Node Capsule (per-center analysis)
  Run locally at each center to generate summary association results without sharing raw data.
  🔗 Federated Network Biomarker Discovery in Immuno-Oncology (Per-Center)

• Aggregation Node Capsule (meta-analysis & visualization)
  Collects outputs from local nodes and performs centralized meta-analysis and visualization.
  🔗 Federated network Immuno-Oncology Meta-analysis

Both capsules will be published in the Code Ocean Open Science Library (OSL) upon peer-reviewed acceptance, ensuring long-term reproducibility.

---

Set Up

Prerequisites

Pixi is required to run this project. If you haven't installed it yet, follow these instructions

Installation

# Clone the repository
git clone https://github.com/bhklab/predictio-uv-dist.git
cd predictio-uv-dist

# Install dependencies via Pixi
pixi install

Repository Structure

predictio-uv-dist/
├── config/           # YAML config files for each dataset and center
├── data/             # Raw, processed, and results directories
├── workflow/         # Scripts and Nextflow pipeline for analysis
├── docs/             # MkDocs-based project documentation
│   └── README.md     # Documentation index and setup instructions
└── pixi.toml         # Pixi environment specification

---

Documentation

Full documentation, including usage instructions, data setup, config templates, and pipeline stages, will be available in the docs/ folder or via published GitHub Pages.

Start by downloading and organizing the raw input datasets as described in data/rawdata/README.md.

For data download and processing, please refer to the univariable repository:
🔗 https://github.com/bhklab/PredictIO-UV-Dist