Identification of Clinically Relevant Genes Using Machine Learning Techniques

Machine learning (ML) has demonstrated great potential and considerable effectiveness in big data analysis. It is capable of extracting novel features, including indicators or biomarkers in biology, that are unachievable with previously existing methodologies. Biomarkers are used to measure and evaluate biological and pathological processes, or pharmacological responses to assess patient status. In clinical settings, biomarkers enable the prediction and detection of many aspects, such as prognosis, drug response, and tumor recurrence. Stratification using RNA expression is the gold standard for patient prognosis in cancer. In this chapter, we describe a practical guide for identifying clinically relevant biomarkers using ML. First, integrated multi-omics data are used as input data for the analysis. Then, an autoencoder is used for dimension reduction, and a bottleneck feature space of 100 dimensions is used to extract clinically relevant features. Finally, we identify biomarkers related to the outcomes from the transcriptome data.