A Hybrid Statistical–Machine Learning Approach Uncovers Distinct Immune and Proliferative Pathways in Lung Cancer Transcriptomes..

Abstract

Lung cancer is the greatest cause of death in the world among all types of cancer. It is brought about by complicated genetic variations, variation in transcription and immune system interactions. RNA-seq data (N= 739 samples and 59,429 genes) and machine-learning algorithms (Random Forest, LightGBM and Elastic Net) and statistical methods of differentialexpression (limmavoom and DESeq2) assist us to identify strong molecular signatures. Important pathways within tumors which regulate the cell cycle, copy DNA, and preserves good shape of chromosomes were found through statistical research. Meanwhile, the machine learning methods identified non-linear immune responses such as like T-cell stimulation, cytokine signaling, and antigen presentation which are frequently missed in the fold-change based tests. The four complementary sets of genes have been obtained through the attempts to merge the two strategies (Stat-only, ML-only, Common, Union), and each shows specific enhancements in the pathway but adds architectural power to identifying biomarkers, analyzing the pathways, and the prospective tool in the field of precision medicine.