Harnessing machine learning technique to authenticate differentially expressed genes in oral squamous cell carcinoma.

Objective: Advancements in early detection of the disease, prognosis and the development of therapeutic strategies necessitate tumor-specific biomarkers. Despite continuous efforts, no molecular marker has been proven to be an effective therapeutic tool for the early detection of cancer. The study aims to determine diagnostic and prognostic signature genes that may be involved in cancer pathology and hence, may serve as molecular markers.

Study design: Eight candidate genes were selected based on our prior study of transcriptomic sequencing and validated in 100 matched pair samples of oral squamous cell carcinoma (OSCC). We further utilized machine learning approaches and examined the diagnostic presentation and predictive ability of the OSCC genes retrieved from publicly available The Cancer Genome Atlas (TCGA) database and compared with our results.

Results: We conducted qPCR analysis to validate the expression of each gene and observed that each gene was present in the majority of OSCC samples. The predictive ability of selected genes was stable (with an average accuracy of 84%) across different classifiers. However, on validation with our dataset, it showed 75% accuracy, which might be because of the demographic variation of the samples.

Conclusions: The present research outlines cancer-associated molecular biomarkers that might eventually contribute to an enhanced prognosis of cancer patient by identifying novel therapeutic targets.