Ontology-based prediction of cancer driver genes: predCAN

Datasets

We used the driver genes from and IntoGen other genes from cellular Phenotype Database

Dependencies

To install python dependencies run: pip install -r requirements.txt

Prediction of candicate cancer driver genes workflow

We integrate the annotates from Gene ontology (GO), Cellular Microscopy Phenotype Ontology (CMPO) and Mammalian Phenotype ontology (MP) using OPA2Vec

We used generated embeddings to test each ontology individually and evaluate their performance (AUC and F-score). Then considering the genes in which they have complete representation in GO, CMPO and MP.

python OPA2Vec_Prediction.py "filename"

And merging all three ontologies by running mergeOntologies.groovy to have outont.owl.

groovy mergeOntologies.groovy

As a result, we predict 112 new candidate driver genes within 20 cancer type Predicted112candidateDriverGenes.txt

Validation on two-cohorts

Following GATK pipline with MuTect2 in tumor-only mode by running VariantsCall-TumorOnlyMode.sh as a job with specifiying the folder name of the VCFs files:

sbatch VariantsCall-TumorOnlyMode.sh

A- First analysis (count the mutations)

Start by running Annovar annovarscript.sh:

chmod +x annovarscript.sh
./annovarscript.sh

And run combinedhist.r to plot the figures with trying to adjust limits parameter.

B- Second Analysis (pathogenicity test)

Start with PrepareU-test.sh (it will run Annovar, but independent from the previous test):

chmod +x PrepareU-test.sh
./PrepareU-test.sh

Then, ranksumtest.py to compute different 7 p-value scores and it needs those specific cancer type related files (all-driver and predicteddriver):

python ranksumtest.py

Final notes

For any comments or help needed with how to run the scripts, please send an email to: [email protected]