Secondary Logo

Journal Logo

Prostate Cancer: Markers (MP60): Moderated Poster 60: Monday, September 13, 2021

MP60-08 DEEP TRANSCRIPTOMIC PROFILING OF PROSTATE CANCER WITH QUANTITATIVE MEASURES: A SPECTRA APPROACH

Hanson, Heidi; Ambrose, Jacob; O'Neil, Brock; Lee, Greg; Leiser, Claire; Waller, Rosalie; Ramsay, Joemy; Madsen, Michael; Das, Rupam; Dechet, Christopher; Avery, Brian; Camp, Nicola

doi: 10.1097/JU.0000000000002095.08
  • Free

INTRODUCTION AND OBJECTIVE:

Methods that embrace the complexity of prostate cancer (CaP) gene expression are necessary for accurate phenotypic characterization. We previously presented a novel agnostic computational framework, SPECTRA, to describe RNA sequencing (RNAseq) data using multiple quantitative expression variables, or transcriptomic spectra (TrS). Here, we implement this technique to derive a set of TrS variables for CaP tumors from The Cancer Genome Atlas (TCGA). We compare the identified TrS with clinical and sample characteristics, progression-free interval, and previously established CaP subtypes.

METHODS:

Gene-level RNAseq data were downloaded from the Genomic Data Commons (GDC) Data Portal. Data were preprocessed using the SPECTRA protocol (nsamples=480, ngenes = 10,438) and matrix factorization was used to derive quantitative measures for each TrS. Linear, logistic, multinomial logistic, and Cox regression were used to assess the relationship between TrS and age, Gleason Risk Score, tumor stage, and progression free interval. Two approaches were used to select TrS associated with demographic, clinical, and molecular features of the tumors; “hard-thresholding” (Bonferonni corrected p-value) and lasso regularization. We compare model fit statistics to determine the difference in predictive power between our TrS and previously defined molecular subtypes defined by fusion of ETS family genes.

RESULTS:

We identified 21 TrS in our data that together explain 65.5% of the variance in global gene expression across CaP tumors in TCGA. Many of the TrS were associated with demographic and clinical characteristics of the patients and progression-free interval, with TrS3 having the strongest associations (Figure 1). The difference in predictive power of the TrS vs the previously identified molecular subtypes was substantial, with the difference in pseudo R-square in the range of 30% on average. This suggests that more information can be gained from quantitative vs. qualitative measures of gene expression.

CONCLUSIONS:

This approach had substantially better model fit than traditional qualitative tumor classifications and could lead to innovative ways to understanding gene expression patterns driving individual risk, treatment response and survival.

Source of Funding:

5K07CA230150-03

© 2021 by American Urological Association Education and Research, Inc.