Utilizing a Novel Machine Learning Pipeline for Single-Cell Transcriptomatic Characterization of a Remodeled Tumor Microenvironment
By Alan Chang
It was just another car ride home with my brother; I was the curious freshman asking difficult questions to the knowledgeable senior. Topic of the night: viruses. It seemed almost unfair, how viruses could inject their DNA into target cells and exploit them as host cells. I inquired further, “But Kevin, what if scientists could actually reprogram these viruses to artificially alter genomes?” He paused. “Hm … never thought of that.” After I got home, I eagerly searched “genome editing viruses,” and a phrase kept popping up: “CRISPR-Cas9.” I found exactly what I was looking for: researchers were transfecting reprogrammed bacterial plasmids into target cells to selectively mutate genomes. Moreover, I was particularly interested in cancer CRISPR screening - a contemporary method for identifying tumorigenesis drivers in the tumor microenvironment. After reading more literature and discussing my interest with teachers, I realized the massive potential this form of computational analysis holds in the field of systems biology. I was determined to teach myself R language and parallel it with my enthusiasm for the CRISPR-Cas9 system to hopefully aid in the development of improved cancer immunotherapy … Cancer death tolls are expected to continue increasing to 13 million in 2030. Despite recent advancements in cancer research, cancer cells utilize countless genetic perturbations to resist current methods of immunotherapy. Understanding the tumor mechanisms of immune escape is imperative for designing improved immunotherapies. This study lays important groundwork for elucidating the functional roles of tumorigenesis drivers. By employing diverse machine learning approaches with a high-resolution 10X Genomics Chromium scRNA-seq dataset, this study establishes a novel pipeline to separate, identify, and characterize the remodeled cell populations within the tumor microenvironment after Prkar1a knockout. With this methodology, researchers can evaluate the effects of countless protumor genetic perturbations that have remained unexplored for decades. Top differentially expressed genes identified in both the tumor and immune subpopulations not only characterized the cell populations, but also reinforced current literature. In addition, this study is the first of its kind to holistically validate the efficacy of two cutting-edge tools, PHATE and scmap, via well-established methods tSNE and canonical marker expression, respectively…