Abstract 3003: Development of a computational framework for disease knowledge modeling and clinical treatment decision support and its application to a rare cancer, malignant peripheral nerve sheath tumor (MPNST)

Abstract

Abstract The integration of increasingly diverse and disparate data about cancers into routine, clinical decision making is a major challenge for evidence-based medicine. Clearly, a comprehensive knowledge model for an indication is the mandatory basis for any evidence-based treatment prioritization. But the concepts and technologies required to collate pre-existing disease and treatment data into useful models of molecular etiology are unclear. Rare cancers represent a particularly pressing need in this context: finding, combining and enriching the few existing disease information and using it for treatment prioritization, -especially in the off-label setting-, beyond the often poorly defined standard of care. To address this, we have developed a novel framework for computational disease knowledge modeling and tested its applicability for clinical treatment decision support in the concrete case of a patient with a recurrent rare cancer, -malignant peripheral nerve sheath tumor (MPNST). Our primary goal was to construct for the first time an integrated disease knowledge model for MPNST by combining bioinformatics and text data mining methods. Text data mining on scientific literature (Pubmed) defined a basic corpus of 317 publications with a total of 145 proteins/ genes and sequence variants, associated with indication terms for MPNST. This set was enriched in factors involved in the MPNST-related tissue and cell biology, and candidate drivers of familial and sporadic MPNST forms. From this core we identified 47 factors, for which MPNST-associated genomic or transcriptomic aberrations had been reported. Utilizing information on molecular pathway association, physical protein interactions and cellular process association of these factors from an integrated oncology data warehouse, we then built a network representation of the molecular disease knowledge for MPNST. Our model highlights and links several molecular modules potentially involved in MPNST, such as deregulated receptor tyrosine kinases involved in cell proliferation, adhesion and invasion, the downstream signaling pathways PI3K/ AKT/ MTOR and RAS/ RAF/ MAPK, and their interactions with the familial MPNST candidate genes NF1 and NF2. Using drug database and clinical trial information we matched treatment options and associated open protocols to our model and mapped the target promiscuity profiles of drugs to network topography. This validated the clinical relevance of molecular entities in our model and allowed treatment prioritization for the patient based on the modeled disease knowledge. Our study demonstrates the feasibility of computational disease knowledge modeling in oncology and the utility of it in providing mechanistic evidence for treatment prioritization in poorly understood rare cancers such as MPNST. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 3003. doi:1538-7445.AM2012-3003