CancerSig

CancerSig: This work proposes an evolutionary learning method called CancerSig to identify stage-specific miRNA signatures for cancer stage prediction. CancerSig obtains a small set of miRNA biomarkers as a signature and establishes a panel of miRNAs to predict cancer stage across 15 cancer types.

Input Data

miRNA expression (RPM log2) with CSV format. User can put their in-house miRNA expression into csv file.

Dataset	Abbreviation	miRNA panel
Bladder urothelial carcinoma	BLCA	BLCA_input.csv
Breast invasive carcinoma	BRCA	BRCA_input.csv
Colon adenocarcinoma	COAD	COAD_input.csv
Esophageal Carcinoma	ESCA	ESCA_input.csv
Head and neck squamous cell carcinoma	HNSC	HNSC_input.csv
Kidney renal clear cell carcinoma	KIRC	KIRC_input.csv
Kidney renal papillary cell carcinoma	KIRP	KIRP_input.csv
Liver hepatocellular carcinoma	LIHC	LIHC_input.csv
Lung adenocarcinoma	LUAD	LUAD_input.csv
Lung squamous cell carcinoma	LUSC	LUSC_input.csv
Rectum adenocarcinoma	READ	READ_input.csv
Skin cutaneous melanoma	SKCM	SKCM_input.csv
Stomach adenocarcinoma	STAD	STAD_input.csv
Thyroid carcinoma	THCA	THCA_input.csv
Uveal melanoma	UVM	UVM_input.csv

Getting start

git clone https://github.com/mingjutsai/CancerSig.git
cd CancerSig

build LIBSVM

cd libsvm
make

Options of CancerSig

python cancersig_main.py -h
usage: cancersig_main.py [-h] -t T -i I

CancerSig obtains a small set of miRNA biomarkers as a signature and establishes a panel of miRNAs to predict cancer stage across 15 cancer
types.

optional arguments:
  -h, --help  show this help message and exit
  -t T        choose one of the cancer type(BLCA, BRCA, COAD, ESCA, HNSC, KIRC, KIRP, LIHC, LUAD, LUSC, READ, SKCM, STAD, THCA, UVM)
  -i I        the file of miRNA gene expression for specific cancer type

Example of running CancerSig

python cancersig_main.py -t BLCA -i models/BLCA/BLCA_input.csv
svm-predict exist:libsvm/svm-predict
svm-scale exist:libsvm/svm-scale
cancer type:BLCA
expression file:models/BLCA/BLCA_input.csv
model exist:models/BLCA/BLCA.model
prediction score:0.320973

Prediction Result

Normalized probabilities between 0 to 1, with higher scores more likely to be advanced stages and lower scores more likely to be early stages.

Contact

• Srinivasulu Yerukala Sathipati: sathipathi.srinivasulu@marshfieldclinic.org
• Ming-Ju Tsai: mingjutsai@hsl.harvard.edu