batterysnoopy / cell-death-signatures Goto Github PK

Abstract

Transcriptomics perturbation signatures are valuable data sources for functional genomic studies. They can be used to identify mechanism of action for compounds and to infer functional activity of cellular processes. Linking perturbation signatures to phenotypic studies opens up the possibility to model cellular phenotypes from gene expression data and to predict drugs interfering with the phenotype. By linking perturbation transcriptomics data from the LINCS-L1000 project with cell viability phenotypic information upon genetic (Achilles project) and chemical (CTRP screen) perturbations for more than 90,000 signature - cell viability pairs, we show that cell death signature is a major factor behind perturbation signatures. Analysing this signature revealed transcription factors regulating cell death and proliferation. We use the cell death - signature relationship to predict cell viability from transcriptomics signatures, and identify and validate compounds that induce cell death. We show that cellular toxicity can lead to unexpected similarity of signatures, confounding mechanism of action discovery. Consensus compound signatures predict cell-specific drug sensitivity, even if the signature is not measured in the same cell line and outperform conventional drug-specific features. Our results can help understanding mechanisms behind cell death, removing confounding factors of transcriptomics perturbation screens and show that expression signatures boost prediction of drug sensitivity.

The corresponding article for this project is available at Nucleic Acids Research. We also proveide an R Shiny application, CEVIChE (CEll VIability Calculator from gene Expression) to browse predicted cell viability values, and to perform predictions on any gene expression data online.

You have to clone / dowload the project, and run the different Jupyter Notebooks to reproduce our analysis

Libraries used

Beside basic scientific computing (NumPy, pandas, SciPy, scikit-learn) and plotting (Matplotlib, seaborn) we used the following libraries:

1. cmapPy to access Connectivity Map Resources
2. colormap for some coloring applications
3. adjustText for text adjustment in plots

We also use some R packages:

1. viper for regulon enrichment
2. biomaRt for translating between gene IDs
3. msigdbr for GO and KEGG
4. piano for enrichment analysis
5. pROC statistical analysis of ROC curves

Data download

We used the data from LINCS, CTRP and Achilles for this project.

cell_death_0_download.ipynb: downloads / helps to download all the necessary files.

cell_death_1_preprocess.ipynb: performs data preprocess / matching between the different datasets

cell_death_2_model_analysis.ipynb: performs model building (related to Fig.1 of manuscript)

cell_death_3_functional_analysis.ipynb: pathway enrichmnets and associations with drug response (related to Fig.2 of manuscript)

cell_death_4_moa.ipynb: mechanism of action inference

cell_death_5_prediction.ipynb: predicting cell viability for the whole LINCS-L1000 dataset

cell_death_6_ML.ipynb: Machine Learning prediction of drug sensitivity

cell_death_7_additional.ipynb: additional analysis (not in manuscript)

cell_death_8_figures.ipynb: recreating figures