omics-reg

This repository is supporting material for the paper: Machine learning model of somatic multiomics data reveals HRD regulators beyond BRCA-centric model.

Setup

Install the project's dependencies with conda:

$ mamba env create --file conda.env.yaml

Data Input & Configuration

To start the learning process, you need to load the somatic multiomics data that will be used as input data, as well as the data for the target score.

Somatic Mutation

You must download the public maf from MC3 and edit the Gerador_arquivo_MUT.r script to point to the mc3.v0.2.8.PUBLIC.maf.gz file.

The MAF can be found here: https://gdc.cancer.gov/about-data/publications/mc3-2017

Copy Number

You must download the TCGA CNV data and edit the Gerador_arquivo_CNV.r script to point to the TCGA-*.CNV.tsv.gz files.

These files are available on the GDC data portal: https://portal.gdc.cancer.gov We used data from TCGA release V31

Gene Expression

You must download the TCGA gene expression TPM data, and edit the Gerador_arquivo_GEL.r script to point to the TCGA-*.TPM.tsv.gz files.

These files are available on the GDC data portal: https://portal.gdc.cancer.gov We used data from TCGA release V36

Methylation

You must download the TCGA methylation data of both HM27 and HM450 assays, and edit the Gerador_arquivo_MET.r script to point to the TCGA-*.genes.tsv.gz files.

These files are available on the GDC data portal: https://portal.gdc.cancer.gov We used data from TCGA release V31

Target score

You need to edit the target.r script to point to a CSV table containing the target scores to be used in the training of the model regression.

We used the panHRD score, as published by: 'Integrating homologous recombination deficiency signatures enhances deep learning prediction from whole-slide images'.

Pipeline Organization

The pipeline is organized in the snakemake files Snakefile_agnostic.py (agnostic model) and Snakefile_tumor-specific.py (tumor-specific model). The R scripts are organized in the following order:

• target.r: reads the file with the target scores for the regression.

• Gerador_arquivo_.r: reads the file with the omics input data.

• (IND_)Gera_train_test.r: splits the data between train and test. The 'IND_' prefix is only used in Snakefile_tumor-specific.py, while the version without that prefix in Snakefile_agnostic.py.

• boruta.r: uses the training dataset to reduce the model's dimensionality with the Boruta technique, retaining only the variables related to the output variable panHRD.

• Regressor_boruta.r: Using the variables selected by Boruta, the model is trained using the XGBoost technique for the regression task of the panHRD score. The output file is in the Rdata format, which stores the trained model information and can be used in downstream analyses.

Run the workflow

Run the ./runpipe.slurm script to execute all the workflow steps.