TY - JOUR T1 - The prognostic potential of alternative transcript isoforms across human tumors JF - bioRxiv DO - 10.1101/036947 SP - 036947 AU - Juan L. Trincado AU - E. Sebestyén AU - H. Climente-González AU - A. Pagés AU - E. Eyras Y1 - 2016/01/01 UR - http://biorxiv.org/content/early/2016/01/15/036947.abstract N2 - Background Molecular signatures can improve tumor stage identification, which is essential for therapy selection and patient prognosis. These signatures are generally based on the expression changes that occur during cancer progression, which are related to the activation of specific aggressive phenotypes. However, it is not yet known whether specific transcript isoform expression patterns are informative for clinical stage and survival.Methods Using information theory and machine learning methods, we integrated RNA sequencing and clinical data from The Cancer Genome Atlas project to perform the first systematic analysis of the predictive potential of transcript relative abundances for stage and prognosis in 12 solid tumors. Additionally, we built predictive models for breast tumors with ER positive and negative status and for melanoma tumors with proliferative and invasive phenotypes.Results Tumor-specific models based of transcript isoforms accurately separate early from late stage and metastatic from non-metastatic tumors, and are predictive of survival in patients with undetermined stage. These models show comparable, sometimes better, accuracies compared with models based on gene expression or alternative splicing events, do not correlate with stromal or immune cell content of the samples, and indicate possible functional alterations in the involved genes. Additionally, we describe the transcriptome differences in breast tumors according to estrogen receptor status, and in melanoma tumors according to invasive or proliferative phenotypes, and derive accurate predictive models of stage and survival for each subtype.Conclusions Our analyses reveal new signatures that characterize tumor phenotypes and their progression independently of gene expression. Knowledge about the relative abundance of transcript isoforms in tumors can help predicting stage and clinical outcome, and thereby contribute towards current strategies of precision cancer medicine. ER -