UC San Diego

Despite extensive research characterizing cancer pathology, it’s heterogeneity still challenges our ability to design consistently successful therapeutic strategies for some cancer types. Indeed, divergent outcomes in patients presenting with the same tumor type, and receiving the same treatment, are commonplace. The variety of pathogenic pathways in cancer is a challenge, but it also presents an opportunity to approach treatment from a more personalized perspective. Fundamentally, cancer can be thought of as a type of aberrant shift in the transcriptional network driving the cell’s molecular machinery. Characterizing this altered transcriptional network offers the possibility of enhanced patient stratification and ultimately, more effective therapeutic strategies.

Here we use Taiji, a multi-omics method that leverages chromatin accessibility and gene expression data to create and analyze a model of the cell’s active transcriptional network. We apply Taiji to a universe of patient samples, identifying the key transcription factors (TFs) driving certain cancers. Using the importance of TFs and the strength of TF-gene regulatory relationships as criteria, we stratify patients and identify subclusters within classically-defined cancer subtypes. We demonstrate that those subclusters correlate with clinically relevant phenotypes and computationally predict oncogenic pathways driving their divergent physical characteristics. Taken together, we demonstrate the ability of a novel multi-omics method of transcription network analysis to characterize the tumors of individual cancer patients, offering the potential for more personalized diagnosis and treatment.