UC Irvine

Spatial transcriptomics (ST) plays a pivotal role in cancer research, offering a unique perspective on gene expression within the cancer microenvironment, further revolutionizing our current understanding of the subject. From addressing the limitations of traditional bulk RNA sequencing by preserving spatial context, this review discusses the importance of integrating machine learning (ML), artificial intelligence (AI), and statistical methods for interpreting ST data within oncology. Herein, we use examples from studies involving Raf kinase inhibitor protein (RKIP) and Ying Yang 1 (YY1) to illustrate applications for some of the ST techniques discussed. We explore how applying supervised learning techniques, such as Support Vector Machines (SVMs) and Random Forests (RFs), can significantly help further cancer classification and prediction of clinical outcomes and advance personalized medicine. Additionally, exploring unsupervised learning approaches like clustering and dimensionality reduction methods (PCA, t-SNE, UMAP) allows us to see hidden structures in ST data that may be overlooked. This review discusses recent tools and techniques that have been introduced within the last few years, underlining the transformation brought into ST by ML, AI, and statistical methods that provide new insight into oncogenic drivers such as YY1 and RKIP, cancer heterogeneity, and avenues for personalized medicine approaches in cancer treatment.