GSIAR: gene-subcategory interaction-based improved deep representation learning for breast cancer subcategorical analysis using gene expression, applicable for precision medicine

Tumor subclass detection and diagnosis is inevitable requirement for personalized medical treatment and refinement of the effects that the somatic cells show towards other clinical conditions. The genome of these somatic cells exhibits mutations and genetic variations of the breast cancer cells and helps in understanding the characteristic behavior of the cancer cells. But their analysis is limited to clustering and there is requirement to analyze what else can be done with the data for identifying the tumor subcategory and the stages of subclasses. In this work, we have extended the work with similar data (consisting of 105 breast tumor cell lines) to solve other detection and characterization problems through computation and intelligent representation learning. Most of our work comprises of systematic data cleaning, analysis, and building prediction models with deep computational architectures and establish that the transformed data can help in better distinction of the respective categories. Our main contribution is the novel gene-subcategory interaction-based regularization (GSIAR) based data selection and analysis concept, alongside the prediction, proven to enhance the performance of the classification techniques.