Exploring Efficient Gene Set Alternatives for Improved Breast Cancer Subtype Classification

Abstract

In recent years, advances in breast cancer research have underscored the need for more efficient methods to handle the vast genomic data associated with its diverse subtypes. The signature of the PAM50 gene, although successful as a prognostic tool, poses challenges due to its reliance on a large number of genes, impacting both cost and complexity. This thesis aims to explore alternative gene sets for breast cancer subtype classification that maintain or improve the accuracy of the PAM50 signature. Our research begins with the "Few-Shot Genes Selection" method, which was specifically designed to confront the signature of the PAM50 gene. By identifying and evaluating smaller subsets within the PAM50’s own gene selection, we show that even with a reduced number of the same genes, it is possible to achieve equal or superior classification performance. This finding challenges the notion that a larger gene set is inherently better, showing that streamlined subsets can be just as effective. Building on this, we explore a new approach with the second research focus, ``Using Multi-layer Classification to Improve Worst Prognosis Breast Cancer Subtypes Outcomes''. This method seeks to identify alternative gene sets, distinct from those in PAM50, but with a specific emphasis on improving classification accuracy for the worst prognosis subtypes, such as Basal and Her2. Through a layered classification framework, our objective is to improve the outcomes of these subtypes by selecting genes that are more predictive of poor prognosis cases. In conclusion, this thesis demonstrates that by refining existing gene signatures or discovering new gene sets focused on specific subtypes, we can develop more efficient and cost-effective tools for breast cancer classification. This research contributes to the ongoing effort to improve diagnostic precision and treatment outcomes in clinical settings, particularly for those subtypes with the most challenging prognoses.