yuanyuanli66/gbm.ensemble

Tumor purity prediction using TCGA RNAseq data

This is a small demo based on using TCGA's Tripple Negative Breast Cancer (TNBC) RNA-seq to build an ensemble of gradient boosting machines and use the enseble model to predict tumor purities in TNBC single-cell data.

Installation

Make sure you have R 3.4 or newer.

You can install caret, xgboost,MLmetrics,data.table,vcd, and e1071 packages under R.

The following pipeline runs on Linux (or Linux-like) environment using a Makefile.

Running the demo

Make sure you set up the classpath in env_var.sh file

tb_loc="~/your_path/gbm.ensemble-master/"        # code path
db_loc="~/your_path/gbm.ensemble-master/input/"  # input path
gb_loc="~/your_path/gbm.ensemble-master/output/" # output path
 

After classpath is set, you can use make to check if your running commands are set correctly.

LINUX> make  prepareTNBCbulkRNAandSingleCell
LINUX> make  findXgbParamTNBCbulkRNAandSingleCell
LINUX> make  predictViaRepeatedCvXgbTNBCbulkRNAandSingleCell
 

To run the 3-step pipeline, you can use the following commands:

LINUX> make  prepareTNBCbulkRNAandSingleCell | bash
LINUX> make  findXgbParamTNBCbulkRNAandSingleCell | bash
LINUX> make  predictViaRepeatedCvXgbTNBCbulkRNAandSingleCell | bash
 

Note: The step 2: make findXgbParamTNBCbulkRNAandSingleCell | bash is optional. This step may take a long time. You can skip this step and go to the prediction (step 3) directly.

Running this pipeline with your own gene expression data

You will need to download the TCGA RNAseq data for your tumor type. In the ./input subfolder, you need to generate your own .value and .label files in the same format as common_tcga_singlecell_tnbc.value and common_tcga_singlecell_tnbc.label files.

• .value file contains the gene name and gene expression values of both TCGA RNAseq data and your gene expression data that you want to predict. Each row is a gene, and each column is a sample. Your gene expression data should always come after the TCGA RNAseq data.

• .label file contains 2 columns: column 1 is the sample ID name. The TCGA sample ID (minimum of 16 characters) followed by your expression data sample ID. Column 2 is integers vector (can be any number). These numbers will be with TCGA purity values during the step 1 in the pipeline.