Predicting clinical drug response from model systems by non-linear subspace-based transfer learning

[Mourragui et al 2020]

Biorxiv pre-print : https://www.biorxiv.org/content/10.1101/2020.06.29.177139v2
This folder contains the code necessary to:

• Download data (GDSC, PDXE and TCGA).
• Reproduce the main figures of the mananuscript (Fig 1,2,3,4 and 5) and the table (Table 2)

Setting up

You need to have conda 4.8 or higher. Based on that, you can settup your environment using the following command:

conda create -n transact_figures python=3.6
conda activate transact_figures
sh setup.sh

Downloading the data

We provide a script to download the three public datasets used (GDSC, TCGA and PDXE). First you need to set-up the download environment:

git clone https://github.com/NKI-CCB/TRANSACT_manuscript.git
cd TRANSACT_manuscript
cd ./download_data/
sh setup_download.sh
sh download_data.sh

HMF data can not be publicly shared. If you wish to reproduce our results, you must get in touch with the Hartwig Medical Foundation (https://www.hartwigmedicalfoundation.nl/en/) to get approval on data.

Reproduce the results

Each folder contains the codes and notebook used to produce the figures. Each folder contains a sub-folder called './figures/' where all figures will be plotted in high resolution.

Of particular interest is the figure 4 that can be reproduced from subfolder "figure_4", but also contains the code to reproduce Table 1.

Contact

In case you do not manage to reproduce the figures, you found a bug, or you have any question/suggestion on the code, please contact me either by GitHub, or by mail s [dot] mourragui [at] nki [dot] nl.