Abstract Background: Differential transcription analysis may reveal transcriptome aberrations associated with cancer progression and help identify important drivers in molecularly distinct subtypes of a disease. Methods: We developed a data-driven bioinformatics pipeline (DiffSplice) that performs differential transcription analysis at the alternative splicing (AS) level and allows accurate transcriptome reconstruction through joint analysis of all RNA-seq samples. In this study, we applied DiffSplice to RNA-seq data from TCGA CRC (n = 686) and evaluated the expression patterns of the detected AS events across the four consensus molecular subtypes (CMS) of the disease (n = 511): CMS1 - MSI Immune (hypermutated, n = 76); CMS2 - Canonical (epithelial, WNT/MYC activation, n = 220); CMS3 - Metabolic (epithelial, enriched for KRAS mutations, n = 72); and CMS4 - Mesenchymal (stromal invasion, TGFB activation, n = 143). Results: A total of 15,613 AS events were detected, including 5,128 events that involved novel splice junctions not cataloged in UCSC GAF2.0 transcriptome annotation. Exon skipping was the most common pattern with 6,482 occurrences, followed by 4,962 alternative splice sites, 2,689 alternative 5′/3′ transcription sites and 1,262 retained introns. In the subset of samples with CMS classification, a cross-subtype comparative analysis revealed 1,835 differentially transcribed loci from 1,634 genes, including 759 novel junctions (FDR<0.01). Of particular interest, we found that KRAS 4A expression levels were higher in CMS3 samples (p = 0.0070), in line with the known effect of this isoform as a negative regulator of mutant KRAS alleles. In addition, expression of the CD44v isoform, a marker of epithelial cancer stem cells, was higher in CMS2 and CMS3 samples (p = 0.015). Alternatively, CD44s isoform expression levels were higher in CMS4 samples (p<2.2e-16), consistent with previous literature showing that cancer stem cells of tumors that have undergone mesenchymal transition predominantly express the standard CD44 isoform that contains no variant exons. Finally, variance of the transcription profiles suggested potential intra-subtype heterogeneity and an opportunity to further refine these global patterns. Discussion: Here we present an in-depth transcriptome analysis of TCGA CRC data across the molecular subtypes of the disease. We were able to identify differential transcription patterns in an unbiased way by using a scalable and efficient computational approach that models both annotated and novel splice variants simultaneously. Our pipeline revealed a number of promising candidates that may be acting as key modulators in specific disease subsets and deserve further investigation. Citation Format: Yin Hu, Rodrigo Dienstmann, Justin Guinney. RNA-seq differential transcription analysis of TCGA colorectal cancer (CRC) transcriptomes reveals subtype-specific isoform usage. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 1932. doi:10.1158/1538-7445.AM2015-1932