clinical-additivity

Author: Haeun (Hannah) Hwangbo

This repository contains source codes for "Additivity predicts the efficacy of most approved combination therapies for advanced cancer". Data is deposited in figshare.

Getting Started

Installing all required packages. You would need conda to access the virtual environment.

git clone https://github.com/palmerlabunc/clinical-additivity.git
conda env create -f env/environment_short.yml

conda activate surv

Dependencies: numpy, scipy, pandas, scikit-learn, lifelines, matplotlib, seaborn, snakemake

Reconstructing the analysis

Download all data and place them into a directory of your choice. The original data directory was organized as

data
├── all_phase3_trials
├── approved_trials
├── experimental
├── placebo
├── raw
│   ├── all_phase3_trials
│   ├── approved_trials
│   ├── experimental
│   └── placebo

• raw/ contains files before preprocessing.
• raw/placebo/ contains digitized KM curves from clinical trials for placebo or best supportive care treatments.
• raw/approved_trials/ contains digitized KM curves from all FDA-approved combination therapies and constituent monotherapies clinical trials used for the analysis.
• raw/all_phase3_trials/ contains digitized KM curves from all positive and negative phase III trials (2014-2018) of combination therapies and constituent therapies clinical trials used for the analysis.

All data files should be placed in their approprite directories as described in config.yaml file.

You can reconstruct all tables and figures in the article by running the following code. We recommend using at least 4 cores because the code utilizes parallel computing. If you have limitied computation power, reducing the NRUN in the src/all_phase3_predictive_power.py to 100 or 1000 will significantly reduce the run time.

# specify number of cores {N} you want to use
snakemake --cores {N} all

Note on Methods

For each patient the combination's progression-free survival (PFS) time is:

• Highest single agent model (a.k.a. independent drug action model): $max(PFS_A + PFS_B)$
• Additivity model: $PFS_A + PFS_B - first scan time$

Note, to ensure that additivity does not result in shorter PFS than the monotherapy, we took the maximum between $PFS_A + PFS_B - first scan time$ and $PFS_A$.