caGrid, Fog and Clouds Joel Saltz MD, PhD Director Center for Comprehensive Informatics

Overview Introduction Tools Use Case Fog and Clouds

Fog Computing Grid interoperable enterprise virtualization

caGrid, Enterprise and Clouds

Biomedical Middleware: caGrid, TRIAD, i2b2 caGrid Components –Security (GAARDS) –Language (metadata, ontologies) –Semantic/Federated query –Workflow –Grid Service Graphical Development Toolkit (Introduce) –DICOM, IHE compatibility –Advertisement and Discovery

Preferred Name Synonyms Definition Relationships Concept Code Vocabulary/Ontology

Interoperability –Registered metadata –Ontology concept codes used to annotate models –XML schemas that define data structures also registered –Thus both data semantics AND data structures are registered. That is how we achieve (relative) interoperability.

Use Case: Will Treatment work and if not, why not? Avastin and Glioblastoma in RTOG-0825 Treatment: Radiation therapy and Avastin (anti angiogenesis) Predict and Explain: Genetic, gene expression, microRNA, Pathology, Imaging RT, imaging, Pathology markup/annotations/query Active Data workflows

Avastin and GBMs in RTOG-0825 Analysis on pre-treatment tissue to extract imaging and molecular biomarkers that are indicative of Outcome/Avastin response. whole genome mRNA and microRNA expression profiling of GBM tumor specimens to identify outcome/Avastin response biomarkers Analyzing the Pathology imaging and diagnostic imaging registered with the therapy plan to extract any biomarkers that can indicate Avastin response. Does advanced imaging (eg: diffusion weighted imaging) provide markers that can predict patient response? RT, Diagnostic Imaging and Pathology: Support Human/Algorithm analyses, annotation, markup Data management and display framework that integrates the pathology with the radiology, therapy treatment information and the clinical data. This involves integrating platforms that manage imaging data at ACRIN, pathology at UCSF and molecular data from Emory.

For the sake of quality control, reproducibility and data sharing, results of RT, imaging, Pathology observations and analyses need to be described in a well defined manner Finding: mass Mass ID: 1 Margins: spiculated Length: 2.3cm Width: 1.2cm Cavitary: Y Calcified: N Spatial relationships: Abuts pleural surface; invades aorta

Distinguish (and maybe redefine) astrocytic, oligodendroglial and oligoastrocytic tumors using TCGA and Rembrandt Important since treatment and Outcome differ Link nuclear shape, texture to biological and clinical behavior How is nuclear shape, texture related to gene expression category defined by clustering analysis of Rembrandt data sets? Relate nuclear morphometry and gene expression to neuroimaging features (Vasari feature set) Genetic and gene expression correlates of high resolution nuclear morphometry and relation to MR features using Rembrandt and TCGA datasets.

Annotation and Markup of Pathology Data needs Human/Algorithm Cooperation Astrocytoma vs Oligodendroglima TCGA finds genetic, gene expression overlap Pathologists have also long seen overlap Relationship between Pathology, Molecular, Radiology Relationship to Outcome, treatment response

Example: Compute Intensive Workflow

Fog Computing

Attributes of Fog Computing For legal and organizational reasons, data location is constrained Virtual machines used to create software stacks that use caGrid + other middleware to expose data services (we regularly ship VMs with caGrid based software stacks) Enterprises such as Emory increasingly rely on virtualization architectures Mid-range active storage platforms (eg Emory 1PB archival, 100TB fast disk, 2K cores, infiniband interconnect) will rely heavily on virtualization

Relationship between Fog and Cloud Computing Workflows have enterprise, caGrid and compute intensive components Interoperability between enterprise and caGrid software stacks major current NCI/ONR issue Given virtualization, HPC and large scale data requirements can be tackled with cloud approach

Issues Organizational, legal requirements create constraints on placement of datasets, where VMs can be run Mapping of VMs, datasets: –Huge variation in communication, I/O bandwidth, compute capabilities in Fog/Cloud continua –Multiple software stacks, heterogeneous hardware architectures Security: Fog to Cloud: multiple distinct but overlapping security related requirements and constraints

Thank you