Skip to content

Benchmarking study of feature extraction methods for cancer diagnosis using blood-based biomarkers. Feature extraction methods are compared both in terms of their performance and robustness

Notifications You must be signed in to change notification settings

abhivij/bloodbased-pancancer-diagnosis

Repository files navigation

Blood-based transcriptomic signature panel identification for cancer diagnosis: Benchmarking of feature extraction methods

Citation

If you use this repository, please cite our publication in Briefings in Bioinformatics : Blood-based transcriptomic signature panel identification for cancer diagnosis: benchmarking of feature extraction methods

Problem

Compare feature extraction methods for binary classification of cancer types and subtypes using blood-based biomarkers.

Approach

Build a generic pipeline to run any biomarker dataset on multiple feature extraction methods and classification models

Type of data used

  • microRNAs from Extra Cellular Vesicles
  • Total RNA from Tumour Educated Platelets
  • microRNAs from blood
  • microRNAs from serum

Pipeline

pipeline

The Feature Extraction Method comparison pipeline code is made available as an R package, inside the directory FEMPipeline.

To use this in your project :

devtools::install_github("abhivij/bloodbased-pancancer-diagnosis/FEMPipeline")

And within R :

library(FEMPipeline)

The function to call the pipeline is execute_pipeline.

To obtain information regarding the arguments, within R, use

?execute_pipeline

Main inputs to the pipeline are :

  • Read count file in (transcripts x samples) format. Other omics datasets can also be used.
  • Phenotype file - tab separated file with column named 'Sample' with each of the samples in read count file, and their corresponding meta-data that includes a classification criteria column
  • Classification criteria column name

Code & Directory Structure

The R script files outside the FEMPipeline directory calls the FEMPipeline package for datasets relevant to this study

  • pipeline_executor.R : starting point to call pipeline
  • dataset_pipeline_arguments.R : list of datasets and its meta-data, used by pipeline_executor.R as arguments to call pipeline
  • katana_scripts/ : scripts to call pipeline_executor.R in Katana computational cluster
  • data/ : contains source data, extracted data and preprocessed data
  • phenotype_info/ : contains currently used phenotype files and the script used in some steps of phenotype file creation
  • data_extraction/ : data extraction step in the pipeline
  • results_processing/ : scripts to generate plots from results, statistically analyze results, compute pairwise Jaccard Index, combine results, analyze results specifically of that of Ranger feature selection method
  • install.R : list of packages to be installed to run this pipeline

About

Benchmarking study of feature extraction methods for cancer diagnosis using blood-based biomarkers. Feature extraction methods are compared both in terms of their performance and robustness

Resources

Stars

Watchers

Forks

Packages

No packages published