Open Problems Perturbation Prediction task data analysis

This repository is a supplement to the manuscript "A benchmark for prediction of transcriptomic responses to chemical perturbations across cell types", the benchmark github.com/openproblems-bio/task_perturbation_prediction and the platform https://openproblems.bio/results/perturbation_prediction.

Cell type annotation and initial filtering

Use the conda environment environment.yaml. Execute data_preprocessing/celltype_analysis_1.ipynb and data_preprocessing/celltype_analysis_2.ipynb in sequence to reproduce the cell type annotation and filtering of the three compounds: 'CEP-18770 (Delanzomib)', 'MLN 2238', 'Oprozomib (ONX 0912)'.

The resulting file sc_counts_reannotated_with_counts.h5ad is further processed on the platform github.com/openproblems-bio/task_perturbation_prediction. The analysis that justifies the resulting filtering is shown in data_preprocessing/molecule_filtering.ipynb .

The notebook data_preprocessing/multiome_cell_type_annotation.ipynb applies this same cell type annotation approach to the multimodal snRNA-seq/scATAC-seq baseline data.

Fig1d

The umap plots shown in Fig1d are generated with data_preprocessing/dataset_umap.ipynb

Fig2

representation_analysis/representation_analysis.ipynb reproduces the plots from figure 2

Fig3

results_scripts/1_download.sh
results_scripts/2_combine_results.R
results_scripts/figure3.R

reproduce figure 3

Fig9

results_analysis/results_analysis.ipynb, results_analysis/results_stability_plot.ipynb reproduce the plots from figure 9

Supplementary table with T cell coefficient of variation

data_preprocessing/celltype_analysis_2.ipynb