Skip to content


Repository files navigation

A Multifactorial Model of T Cell Expansion and Durable Clinical Benefit in Response to a PD-L1 Inhibitor

This repository contains the source code for reproducing the experiments and figures from Leiserson, et al. (bioRxiv, 2017). See the references section below for more information on the paper.

We use continuous integration to automatically regenerate the figures whenever there is a push to the master branch. You can see the current figures below. Note that, for runtime purposes, by default we do not run the permutation test, and thus the automatically generated Figure 1b will be different from the one in the paper.



The methods and experiments are written in Python 3. We recommend using Conda to manage dependencies, which you can do directly using the provided environment.yml file:

conda env create -f environment.yml
source activate multifactorial-immune-response-env


We use snakemake to run a pipeline of commands to download and process data, run experiments, and create the figures. To run the entire pipeline, simply run:

snakemake all


Configuration for the entire pipeline is controlled by the variables in configs/default.yml. The following variables can be set:

Variable name Choices Default Value used in paper
model 'en', 'rf' 'en' Both 'en' and 'rf'
n_permutations Positive int 0 1000
max_iter Positive int 1000000 1000000
tol Positive real 1e-7 1e-7
random_seed Positive int 12345 12345
n_jobs Positive int 1 46
figure_format Standard image formats png png

The configuration values used in the paper are stored for convenience in configs/paper.yml. To use that configuration, change the first line of the Snakefile accordingly.


Please report bugs and feature requests in the Issues tab of this GitHub repository.

For further questions, please email Max Leiserson and Lester Mackey directly.


Mark DM Leiserson, Vasilis Syrgkanis, Amy Gilson, Miroslav Dudik, Dean F Bajorin, Jonathan Rosenberg, Samuel Funt, Alexandra Snyder, Lester Mackey. (2018) A Multifactorial Model of T Cell Expansion and Durable Clinical Benefit in Response to a PD-L1 Inhibitor. In submission [bioRxiv preprint].


Figure 1: (a) Predicted log TIL expansion versus ground-truth log TIL expansion for patients held out using LOOCV. Predictions are formed using the elastic net. (b) Histogram of LOOCV error when patient responses are permuted uniformly at random 1000 times (for runtime purposes, we do not perform these permutations by default for these automatically generated figures). The overlaid dotted line displays the LOOCV error obtained on the original dataset.

Figure 2: Learned elastic net coefficients and feature types.

Figure 3: Distributions of biomarker values in patients with and without durable clinical benefit (DCB, defined as ≥ 6 months of progression-free survival): (a) predicted number of expanded TIL clones; (b) missense SNV count; (c) expressed neoantigen count; and, (d) percentage of tumor infiltrating immune cells found to be PD-L1-positive. When each biomarker alone is used for triage, the patients highlighted in red must be treated to ensure all DCB patients are treated.


Multifactorial modeling of response to checkpoint inhibitor immunotherapy from tumor, immune, and clinical features








No releases published


No packages published