This is the repository to reproduce the analysis carried out in our manuscript, from the data download to the figures.

The article describes how we repurposed gene-level knockdown screens across many cancer cell lines and their intrinsic heterogeneity to generate a computational framework called spotter of potential cancer-driver exons whose inclusion is highly predictive of cancer cell proliferation.

The computational framework created to perform in silico RNA isoform screening is available as a Python package called target_spotter and can also be accessed interactively at https://spotter.crg.eu.

• data will contain the raw and preprocessed data generated by running the workflows.
• images contains images for this repository.
• results will contain the results (files and figures) of the analyses.
• support contains a set of relatively small tables and files essential to run all workflows that are either handmade or had to be downloaded manually.
• workflows
  1. obtain_data generates the data/raw directory. Note that some raw datasets require special permissions and/or need to be downloaded manually. See the corresponding workflows/obtain_data/README.md file.
  2. preprocess_data generates the data/prep directory cleaning raw datasets to be analyzed.
  3. model_splicing_dependency generates the results/model_splicing_dependency directory. Perfoms model fitting and validation using existing datasets.
  4. streamlined_therapy_dev generates the results/streamlined_therapy_dev directory. Prioritizes potential cancer-driver exons with therapeutic potential from patient data.
  5. experimental_validation generates the results/experimental_validation directory. Analyzes our inhouse experimental results.
  6. exon_drug_interactions generates the results/exon_drug_interactions directory. Performs drug-exon association models using predicted splicing dependencies for cancer cell lines and explores how potential cancer-driver exons may mediate drug mechanism of action.
  7. treatment_recommendation generates the results/treatment_recommendation directory. Case study of a dataset in which ~40 patients were treated with chemotherapies and we try to predict their outcomes with our statistical models.
  8. prepare_submission generates the results/prepare_submission directory. Collects tables from all workflows and puts them in order to prepare the submission.

After running all workflows (indicated below), the figures used in the manuscript and their accompaining data will be in the subdirectory results/*/figures.

We indicate the versions used, however, you'll probably be able to run the full pipeline with different versions.

• command-line tools

  • conda==23.3.1
  • snakemake-minimal==6.10.0
  • vast-tools==2.5.1
  • csvkit==1.0.5
  • fastq-dump==2.9.1
  • gdc-client==1.6.1
  • ascp==4.1.1.73
  • axel==2.17.10
  • bowtie==1.3.0

• Python==3.8.3

  • joblib==1.1.0
  • tqdm==4.65.0
  • pandas==1.3.0
  • networkx==2.8.4
  • numpy==1.20.3
  • target_spotter==0.1.1

• R==4.1.0

  • Biostrings==2.64.1
  • BSgenome.Hsapiens.UCSC.hg38
  • cluster==2.1.2
  • clusterProfiler==4.0.5
  • ComplexHeatmap==2.9.3
  • cowplot==1.1.1
  • doParallel==1.0.16
  • extrafont==0.17
  • ensembldb==2.16.4
  • EnsDb.Hsapiens.v86
  • foreach==1.5.1
  • ggnewscale==0.4.5
  • ggplotify==0.1.0
  • ggpubr==0.4.0
  • ggraph==2.0.5
  • ggrepel==0.9.1
  • ggtext==0.1.1
  • graphlayouts==0.8.0
  • gridExtra==2.3
  • gtools==3.9.2
  • here==1.0.1
  • igraph==1.3.1
  • optparse==1.7.1
  • pROC==1.18.0
  • readxl==1.3.1
  • scattermore==0.7
  • statebins==2.0.0
  • SummarizedExperiment==1.22.0
  • TCGAbiolinks==2.25.3
  • tidytext==0.3.2
  • tidyverse==1.3.1

Refer to each workflow/*/README.md file to run each workflow. Note that obtain_data and preprocess_data are the most computationally demanding workflows.