Unveiling Molecular Markers and Prediction of Stomach Cancer from Transcriptomic Profile: A Comprehensive Study of Feature Mining and Learning-based Algorithm

Abstract

RNA-Seq technology is one of the most authoritative technologies among all Next-generation Sequencing technologies to produce the expression of transcripts in bulk and at a single-cell level. The bulk RNA-Seq data are high-dimension in nature and comprise large features or transcripts genes across a smaller number of samples. The interpretation of bulk RNA-Seq data is significant for identifying the hidden molecular insights of specific disease prognosis and treatment. In recent times, feature mining has played a crucial part in dimensionality reduction of high-dimensional datasets. In this study, we proposed a framework which is able to predict Stomach cancer and identify molecular markers for Stomach cancer prognosis and treatment. In this study, we have utilized seven different rank and algorithm-based feature selection techniques to find the optimal features set while integrating six different types of classifiers for downstream analysis of the Gene Expression Quantification transcriptomic dataset. Further, we have also performed bioinformatics interpretation of selected top transcript genes viz; survival interpretation, pathological stage-wise expression, GO, and Reactome pathway prediction network pharmacology. However, we premeditated drug-repurposing and natural compounds interaction study with the targeted top genes. Finally, we have applied the selected best feature selection techniques on a multi-omics dataset of Stomach cancer. Boruta (AUC = 0.988–0.994%), MRMR (AUC = 0.958–0.994%), and LASSO (0.982–0.994%) feature selection techniques outperformed other feature selection techniques when combined with six classifiers for the Gene Expression Quantification dataset of Stomach cancer. However, we have identified UBE2D2, HPCAL4, JCHAIN, SF1, ANKRD13C, and NCKAP1 six novel molecular markers from the Stomach cancer Gene Expression Quantification dataset that can serve as potential molecular markers for Stomach cancer. However, we observed that the FDA-approved drug “Everolimus” highly interacted with ANKRD13C and NCKAP1 genes, and the natural compound “UDP-D-galactose” highly interacted with gene HPGD which can be a potential drug target for Stomach cancer treatment.