Integrative Analysis in pediatric AML

This is the main repository for analyses relating to our paper:

Image-based drug screening combined with molecular profiling identifies signatures and drivers of therapy resistance in pediatric AML

Ben Haladik^1,2, Margarita Maurer-Granofszky¹, Peter Zoescher¹, Raul Jimenez-Heredia^1,2,3, Alexandra Frohne¹, Anna Segarra-Roca¹, Chloe Casey¹, Felix Kartnig^2,4, Sarah Giuliani¹, Christina Rashkova^1,3, Peter Repiscak¹, Michael N. Dworzak^1,3,6,&, Giulio Superti-Furga^2,5,&, Kaan Boztug^1,2,3,6,7,&

Affiliations:

1 St. Anna Children’s Cancer Research Institute, Vienna, Austria

2 CeMM Research Center for Molecular Medicine of the Austrian Academy of Sciences, Vienna, Austria

3 Medical University of Vienna, Department of Pediatrics and Adolescent Medicine, Vienna, Austria

4 Medical University of Vienna, Department of Internal Medicine III, Division of Rheumatology, Vienna, Austria

5 Medical University of Vienna, Center for Physiology and Pharmacology, Vienna, Austria

6 St. Anna Children’s Hospital, Vienna, Austria

7 Clinic for Pediatric Immunology and Rheumatology, Center for Pediatrics and Adolescent Medicine, University Hospital Bonn, Germany

& shared senior authors

In this study, we integrate data from image-based drug screening, ATAC-seq, RNA-seq and whole exome sequencing to identify functional genomic predictors of patient risk in pediatric AML.

This repository enables the users to reproduce the main figures of the manuscript in several jupyter notebooks. Analyses to reproduce the plots from the individual manuscript figures is found in the corresponding folders in this repository.

About the paper

Abstract

Despite recent advances in the understanding of genomic landscape of pediatric acute myeloid leukemia (pedAML), targeted treatments are only available for selected genomic alterations and the functional link between genotype and outcome remains partially elusive. Functional precision medicine approaches to investigate treatment resistance and patient risk have not been applied systematically for pedAML.

Here, we describe an advanced functional screening platform combining high-content imaging and deep-learning-based phenotyping. In 45 pedAML patients, we identify BCL2 and FLT3 inhibitors and standard chemotherapy as major drivers of the chemosensitivity landscape, reveal substantial differential sensitivities between risk groups, and may effectively predict individual measurable residual disease and patient risk. Integration with genomic and epigenomic data uncovers a chemotherapy-resistant primitive state vulnerable to combined BCL2 and MDM2 inhibition and HDAC inhibition. Overall, we identify early present signatures of therapy resistance across genetic subgroups and prioritize targeted treatments for these functionally and epigenetically defined patient subsets.

Setup and installation

Code in this repository requires Python 3 and R. Please consult the setup folder for more details.

If you are specifically interested in the analysis of raw image-based screening data, please note that it requires code from 3 other repositories to achieve its goal.

• patchpy for setting up, training and using adversarial autoencoders on image patches
• imscreenpy for integrating data from CellProfiler outputs and applying patchpy models to them
• cell_segmentation for training the segmentation model and integrating it with CellProfiler

The first two are sufficient to run the postprocessing and other analyses that this repo is mainly concerned about. You can set them up with conda or mamba like this:

## create environment with initially required packages
conda create -n pedaml_screening_env --file setup/requirements.txt -c bioconda
## activate environment, download and install packages from git
conda activate pedaml_screening_env
python -m pip install git+https://github.com/HabelNadik/imscreenpy
python -m pip install git+https://github.com/HabelNadik/patchpy

Structure of this repository

Most analyses in the manuscript rely on data annotations in the annotations folder that contains annotations of both drugs and patient samples. Otherwise, analyses to reproduce manuscript figures or supplementary figures are contained in the respective folders:

• figure_1_pipeline_benchmarking contains analyses for benchmarking the pipeline as described in Figure 1
• figure_2_cohort_overview contains plotting notebooks to plot the cohort overviews
• figure_3_chemosensitivity_landscape contains notebooks to plot the chemosensitivity landscape and the corresponding analyses
• figure_4_correlation_analysis contains notebooks for the correlations in figure 4
• figure_5_molecular_underpinnings contains notebooks for the integration of ATAC-seq data as described in in figure 5
• figure_6_risk_pedictivity contains notebooks to do the analyses of risk predictiveness described in figure 6

Analyses for raw(er) data of different modalities is found in the corresponding folders:

• sequencing_data_analysis for sequencing data
• cohort_drug_response_analysis for cohort drug response analysis
• imaging_data_analysis contains scripts to run our image analysis pipeline and to train models and to prepare imaging data for training

We would like to add some more documentation and most of the actual underlying data are too large to share via GitHub, so this repo may receive some updates to even better document our approach.