Transcriptome profiling by combined machine learning and statistical R analysis identifies TMEM236 as a potential novel diagnostic biomarker for colorectal cancer

Ashutosh Mani,Sandeep Kushwaha,Aakash Chawade,Neha Shree Maurya

doi:10.1038/s41598-021-92692-0

Abstract

Colorectal cancer (CRC) is a common cause of cancer-related deaths worldwide. The CRC mRNA gene expression dataset containing 644 CRC tumor and 51 normal samples from the cancer genome atlas (TCGA) was pre-processed to identify the significant differentially expressed genes (DEGs). Feature selection techniques Least absolute shrinkage and selection operator (LASSO) and Relief were used along with class balancing for obtaining features (genes) of high importance. The classification of the CRC dataset was done by ML algorithms namely, random forest (RF), K-nearest neighbour (KNN), and artificial neural networks (ANN). The significant DEGs were 2933, having 1832 upregulated and 1101 downregulated genes. The CRC gene expression dataset had 23,186 features. LASSO had performed better than Relief for classifying tumor and normal samples through ML algorithms namely RF, KNN, and ANN with an accuracy of 100%, while Relief had given 79.5%, 85.05%, and 100% respectively. Common features between LASSO and DEGs were 38, from them only 5 common genes namely, VSTM2A, NR5A2, TMEM236, GDLN, and ETFDH had shown statistically significant survival analysis. Functional review and analysis of the selected genes helped in downsizing the 5 genes to 2, which are VSTM2A and TMEM236. Differential expression of TMEM236 was statistically significant and was markedly reduced in the dataset which solicits appreciation for assessment as a novel biomarker for CRC diagnosis.

Highlights

Colorectal Cancer (CRC) is very common in many countries and is one of the major causes of death worldwide[1]
A total of 695 Colorectal cancer (CRC) samples were collected from the The Cancer Genome Atlas (TCGA) database Fig. 2
The CRC gene expression dataset was reduced in dimensionality and was further analyzed through the different algorithms, named as Principal Component Analysis (PCA) and t-distributed stochastic neighborhood estimation (t-SNE)

Summary

Introduction

Colorectal Cancer (CRC) is very common in many countries and is one of the major causes of death worldwide[1]. Sun et al.[1] used GEO datasets and applied the Robust Rank Aggregation method to identify significant Differentially Expressed Genes (DEGs). They found 494 significant differential expressions containing 282 downregulated and 212 upregulated genes. Another study by Su et al.[4] has used both miRNA and mRNA datasets from GEO to identify. The studies which are mentioned above had only used the traditional approaches of R bioconductor for finding the genes responsible in CRC progression. Sometimes the traditional approaches often provide results that are inconsistent in behavior In this context, alternative methods can be implemented which can provide better and consistent results to achieve the respective goal. The classification of gene expression data can be performed through machine learning (ML) algorithms to find significant features

Methods

Results

Discussion

Conclusion