Challenges

• Cancer research involves the use of loss- and gain-of-function mouse models to test hypotheses.
• A major goal is to correlate those findings to humans.

Benefits

• The public domain (GEO, TCGA, cBioPortal) consist of thousands of gene expression datasets with associated phenotypic data.
• By leveraging public data, we can validate findings in multiple cohorts as well as generate data-driven hypotheses.

Objectives

• The client requested survival analyses on gene expression data for a gene signature as part of a larger effort to study the tumor microenvironment and the role genes play in survival outcomes.
• The client requested data analysis for two types of cancer patients.
• Additionally, the client wanted to explore the impact of a gene signature in checkpoint blockade responders in a metastatic cancer dataset.

Solutions

• We obtained transcriptomic data from publicly available cancer patient cohorts including clinical metadata.
• We completed an exploratory analysis to assess distributional differences to check for sample to sample variability and presence of any outlier samples. We then cleaned, processed, and normalized the raw data as necessary to prepare it for survival analysis.
• We performed univariate survival analysis for individual genes followed by a multivariate analysis of a gene signature (multiple genes) using a non-parametric, unsupervised method, and Gene Set Variation Analysis, to summarize the gene sets.
• Lastly, we performed univariate and multivariate survival analyses using a data-driven cutoff to optimize the cutpoint that maximizes the statistical significance of the difference between low and high expressing patients that corresponds with survival outcomes.