RGIE: A gene selection method associated with radiotherapy resistance in head and neck squamous cell carcinoma

This repository hosts the experimental code and data for the paper "RGIE: A gene selection method associated with radiotherapy resistance in head and neck squamous cell carcinoma".

Dataset acquisition

The RNA sequencing (RNA-seq) data of HNSCC patients can be obtained from The Cancer Genome Atlas (TCGA) database, while information on whether the patients received radiotherapy and their survival can be obtained from the XENA database.

Document description

Document	Description
1_EDASeq_normalization.R	R script file for downloading the TCGA-HNSC dataset and running EDASeq normalization.
2_data_preprocessinge.ipynb	Oversampling of radioresistance samples using SMOTE.
3_ReliefF_feature_selection.ipynb	Calculate the importance of genes and rank them.
4.1_GENIE3 sorting (ASC), feature elimination and classification.ipynb	First, the regulatory importance of genes is calculated using GENIE3. Second, sort the genes in ascending order. Then, run a feature elimination algorithm to select genes.
4.2_GENIE3 sorting (DESC), feature elimination and classification.ipynb	First, the regulatory importance of genes is calculated using GENIE3. Second, sort the genes in descending order. Then, run a feature elimination algorithm to select genes.
5.1_classifier_comparison_all_genes.ipynb	Comparison of classifier algorithms using all genes as input.
5.2_classifier_comparison_9_genes.ipynb	Comparison of classifier algorithms using the selected 9 genes as input.
5.3_confusion matrix_9_genes.ipynb	Confusion Matrix Analysis of Classification Algorithms.
6_Comparison_of_feature_selection_methods	Folder for feature selection algorithm comparisons.
GENIE3.py	GENIE3 algorithm.

Quick Start

Run "1_EDASeq_normalization.R" to download the gene expression matrix and perform EDASeq normalization.
Run "2_data_preprocessing.ipynb" to oversample radioresistance samples.
Run "3_ReliefF_feature_selection.ipynb" to calculate and rank the importance of genes.
Run "4.1_GENIE3 sorting (ASC), feature elimination and classification.ipynb" and "4.2_GENIE3 sorting (DESC), feature elimination and classification.ipynb" to calculate the regulatory importance of genes, sort (ascending order or descending order) and perform feature elimination.
Run "5.1_classifier_comparison_all_genes.ipynb", "5.2_classifier_comparison_9_genes.ipynb" and "5.3_confusion matrix_9_genes.ipynb" to compare the performance of the classifiers.
Run the code in the folder "6_Comparison_of_feature_selection_methods" to compare different feature selection algorithms.

Result generation

To improve the reproducibility of the results, instructions for generating the results in the paper are given here.

Item	Overview	Instruction
Fig. 2 Performance of identifying HNSCC-RR with increasing number of genes in terms of feature importance.		Run 3_ReliefF_feature_selection.ipynb to calculate and rank the importance of genes. Then use the saved performance indicator files (files starting with ReliefF_ in the result_data directory) to draw Figure 2.
Fig. 3 Feature importance of genes in the first 20 experiments.		Run 3_ReliefF_feature_selection.ipynb and plot the result_data/sortedImp_60.xlsx file using MATLAB (See Draw_feature_importance.m for the code.).
Fig. 4 Sum of the feature importance values of the top 100 genes.		Run 3_ReliefF_feature_selection.ipynb.
Fig. 5 Classification performance of genes ranked by different methods.		Run 4.1_GENIE3 sorting (ASC), feature elimination and classification.ipynb and 4.2_GENIE3 sorting (DESC), feature elimination and classification.ipynb.
Fig. 6 GI network of the final nine signature genes.		Select the final 9 genes from the GENIE3_ranking.txt file and use OriginLab software to draw Figure 6.
Fig. 7 Gene ontology (GO) and signaling pathway analyses were performed on the signature genes.		The plotting data is stored in result_data/gopath. The figure was plotted by https://www.bioinformatics.com.cn , an online platform for data analysis and visualization.
Table 3. Classification performance comparison using all genes.		Run 5.1_classifier_comparison_all_genes.ipynb.
Table 4. Classification performance comparison using the selected nine genes.		Run 5.2_classifier_comparison_9_genes.ipynb.
Fig. 8 Confusion matrix for different classifiers.		Rum 5.3_confusion matrix_9_genes.ipynb.
Table 5. Comparison of different feature selection methods.		Run the algorithms in the 6_Comparison_of_feature_selection_methods directory.
Fig. 9 qRT‐PCR validation of the expression of 9 signature genes.		Experimental methods are noted in the Methods section of the manuscript.
Fig. 10 Comparison of classification performance in the three steps when the number of genes is nine.		It can be plotted according to the results obtained before.