Machine Learning Approaches for Identification of Potential Biomarkers from Cancer Omics Data

Abstract Machine learning (ML) techniques have widely been used to analyze and interpret multi-omics data. It allows researchers to uncover complex relationships and patterns within molecular features. In the present comprehensive work, we performed text mining of biomedical literature data against selected ten cancer types (breast, colon, cervical, CNS, leukemia, lung, melanoma, ovarian, prostate and renal) using the BioNLP python package. We also constructed a gene-drug interaction network to find the potential biomarkers. The major 30 hub genes were identified to drive more effective and targeted cancer therapies and relevant oncogenic pathways. Using the text mining and network-based approach we were able to identify 49 genes. These were unique and significant against cancer types and are not updated in the cancer omics databases such as TCGA and cBioPortal. Further, we employed machine learning t-SNE clustering for the identification of putative biomarkers based on cancer omics profile and to understand the complex molecular landscapes within the ten types of cancer. To gain insight into the survival outcomes of cancer patients, a Kaplan-Meier plot of the Cox coefficient was performed to get the survival correlation against TCGA data. Multi-omics data analysis has shown a significant potential to transform cancer research and clinical practice by providing a holistic view. Also, it enables to design of precision cancer medicine and drives advances in prevention, diagnosis, and treatment strategies.