PROJECT SUMMARY / ABSTRACT Rhabdomyosarcoma (RMS) is a devastating pediatric soft tissue cancer with morphological features of develop- ing skeletal muscles. Although most patients with RMS achieve a complete remission, one third will develop disease recurrence which is associated with a dismal clinical outcome. These clinical challenges underscore an urgent need to identify patients at risk for resistance and develop better therapy to reduce the risk of recurrence. Our pilot study revealed that rhabdomyosarcoma tumors have developmental intratumoral heterogeneity: differ- ent cells in a single tumor harbor transcriptomic feature of different myogenic stages. Moreover, tumor cells with developmentally immature characteristics are enriched in post-therapy specimens and have the potential to rep- licate and reconstitute the entire developmental trajectory after therapy. In this application, we will develop com- putational analysis methods to unambiguously identify immature tumor subpopulations in single cell RNA-seq data and to reveal their master deregulated genes (Aim 1); In Aim 2, we will characterize the changes in those transcriptional networks and cellular states during treatment in orthoptic patient derived xenograft models (O- PDXs) in vivo. And in Aim 3, we will develop an innovative deep-learning data mining approach to evaluate the prognostic significance of the myogenic transcriptional networks that underly RMS cellular heterogeneity in pa- tient tumors. The proposed study integrates computational, statistical and experimental approaches to study the role of immature cell populations in rhabdomyosarcoma recurrence. Building upon our computational expertise, research experience in rhabdomyosarcoma, robust preliminary results and highly productive collaborations with multi-disciplinary expertise in genomics/epigenomics, machine learning, Bayesian statistics and translational re- search in rhabdomyosarcoma, we are in a unique position to achieve the goals of this research proposal. The proposed research will be impactful because it will potentially change how we treat children with rhabdomyosar- coma and the developed computational/statistical approaches will be broadly applicable to cancer research.