Transfer Learning Project

This repository contains the official implementation of Julian van Toledo's MSc Project, focusing on transfer learning in survival analysis.

Setup and Installation

Prerequisites

• Python 3.7.12
• R version 4.2.0
• Docker

Data prerequisites

RNA-sequencing data and matching clinical data should be downloaded:

• TCGA: Use TCGA-Assembler 2 https://github.com/compgenome365/TCGA-Assembler-
• ICGC: Data portal closed per June 2024, no longer available.

Installation

1. Docker Setup:

   • Run the following commands to set up the environment using Docker:

     docker build -t transferlearning .
     docker run -it transferlearning

   • This will install all necessary dependencies and create the right environment.

2. Running the Script:

   • After setting up the Docker environment, execute the main script:

     python transferlearning.py --approach [1,2,3,4,5,6]

   • The approach will determine which transfer learning approach is used for training the model, this is described in the paper (Figure 3)

Files and Scripts

Main Script

• transferlearning.py: The primary script implementing the transfer learning model.

Configuration and Other Files

• Dockerfile: Docker configuration for setting up the environment.
• requirements.txt: Lists all Python dependencies.
• .gitignore: Specifies files to ignore in version control.
• transferlearning.yaml: Configuration file for the transfer learning process.

Archive

These scripts are not necessary for the main transfer learning procedure and are archived:

• benchmark.py
• cox_nnet_v2.py
• hyperparam_opt.py
• nnet_survival.py
• pre_dataset.py
• Preprocessing_icgc.R
• preprocessingscript.R
• TensorBoardNotebook.ipynb
• gpu_test.ipynb
• nnetsurvivaltransfer.ipynb
• transferlearning_1.py
• transferlearning_2.py
• transferlearning_3.py
• transferlearning_4.py

Authors

• Julian van Toledo
• David van Zessen