Transcriptome Spectra: Agnostic Expression Variables To Empower Genomic Epidemiology Studies

SPECTRA is a new data framework to describe variation in a transcriptome as a set of unsupervised quantitative variables. Spectra variables provide a deep dive into the transcriptome, representing both large and small sources of variance, and are ideal for modeling alongside other variables for any outcome of interest. Each spectrum can also be considered a phenotypic trait, providing new avenues for disease characterization or to explore disease risk. We applied the SPECTRA approach to multiple myeloma (MM), the second most common blood cancer. Using RNA sequencing from malignant CD138+ cells, we derived 39 spectra in 767 patients from the MMRF CoMMpass study. We included spectra in prediction models for clinical endpoints, compared to established expression-based risk scores, and used descriptive modeling to identify associations with patient characteristics. Spectra-based risk scores added predictive value beyond established clinical risk factors and other expression-based risk scores for overall survival, progression-free survival, and time to first-line treatment failure. Significant spectra in models may provide mechanistic insight via gene set enrichment based on their gene weights. Gene set enrichment in CD138+ spectrum S5, which was significant for all prognostic endpoints, indicated enrichment for genes in the unfolded protein response, a mechanism targeted by proteasome inhibitors, common first line agents in MM treatment. We also identified significant associations between CD138+ spectra and tumor cytogenetics, race, gender, and age at diagnosis. The SPECTRA approach provides measures of transcriptome variation to deeply profile tumors with greater flexibility to model clinical outcomes and characteristics. AUTHOR SUMMARYComplex diseases, including cancer, are highly heterogeneous, and large molecular datasets are increasingly part of describing an individuals unique experience. Gene expression is particularly attractive because it captures genetic, epigenetic, and environmental consequences. Transcriptome studies are gaining momentum in genomic epidemiology, and the need to incorporate these data in multivariable models alongside other risk factors brings demands for new approaches. The SPECTRA approach is a new intrinsic quantitative data framework for transcriptomes. A tissue is described by a set of quantitative measures (or spectra variables) to deeply profile gene expression in a tissue. Spectra variables are independent and offer flexibility for use in predictive or descriptive modeling. We applied the SPECTRA approach to multiple myeloma, the second most common blood cancer. A set of 39 spectra variables were derived to represent the myeloma tumors. Outcome modeling provided SPECTRA-based risk scores that added predictive value for clinical outcomes beyond established risk factors.