1. Joel Saltz MD, PhD Director Center for Comprehensive Informatics
Data Archive Issues: Integrative Analysis of Pathology, Radiology and High Throughput Molecular Data
Joel Saltz MD, PhD
Director Center for Comprehensive Informatics

2. Objectives of Integrative Analysis
Create categories of jointly classified data to describe pathophysiology, predict prognosis, response to treatment
Reproducible anatomic/functional characterization at gross level (Radiology) and fine level (Pathology)
Integration of anatomic/functional characterization with multiple types of “omic” information
Data modeling standards, semantics
Curation and propagation of results

3. Informatics Approach
Integrative analysis of complementary data sources to improve ability to forecast survival & response
Leverage public “omic” datasets
Create, analyze microscopy and Radiology datasets
Radiology
Imaging
Patient Outcome
“Omic”
Data
Pathologic Features

4. In Silico Center for Brain Tumor Research
Specific Aims:
Influence of necrosis/
hypoxia on gene expression and
genetic classification.
Molecular correlates of high
resolution nuclear morphometry.
Gene expression profiles
that predict glioma progression.
4. Molecular correlates of MRI
enhancement patterns.

5. In Silico Center for Brain Tumor Research
Key Public Data Sets
REMBRANDT: Gene expression and genomics data set of all glioma subtypes
The Cancer Genome Atlas (TCGA): Rich “omics” set of GBM, digitized Pathology and Radiology
Vasari Feature Set: Standardized annotation of gliomas of all subtypes

6. TCGA Research Network
Digital Pathology
Neuroimaging

7. Pathology Image Analysis and in Silico Example: Distinguishing Among the Gliomas
“There are also many cells which appear to be transitions between gigantic oligodendroglia and astrocytes. It is impossible to classify them as belonging in either group”
Bailey P, Bucy PC. Oligodendrogliomas of the brain.
J Pathol Bacteriol 1929: 32:735

8. Nuclear Qualities
Oligodendroglioma
Astrocytoma

9. Interaction between gene expression classification and Pathology Characterization
Hierarchical clustering of 176 Rembrandt samples using TCGA classification genes defines four major subtypes.
Proneural
Neural
Mesenchymal
Classical
(Lee Cooper and Carlos Moreno)

10. Predicting Recurrence/Survival
Lee Cooper Carlos Moreno

11. TCGA: GBM Frozen Sections
179 cases were assessed for % necrosis on frozen section slides for TCGA quality assurance.
Cox-based regression analysis for % necrosis vs. gene expression (795 probe sets; 647 distinct genes)

12. Network Analysis based on % Necrosis
Carlos
Moreno
David
Gutman

13. Identify correlates of MRI enhancement patterns in astrocytic neoplasms with underlying vascular changes and gene expression profiles.
Rim-enhancement
Vascular Changes
Rapid progression
No enhancement
Normal Vessels
Stable lesion ?

14. Angiogenesis Segmentation
H&E
Image
Color
Deconvolution
Hematoxylin
Eosin
Eosin intensity image
Eosin
Image
Spatial
Norm.
Density
Image
Calculation
Boundary
Smoothing
Density
Image
Object ID
Segmented
Vessels
Angiogenic Segmentation

15. Sharing and Archiving Multi-scale Integrative In Silico Experiments
Metadata
Identifiers
Federated Archives
Semantic Query Support
9/17/20189/17/2018

16. Metadata
Precisely and unambiguously describe an in silico experiment (reproducibility and reuse for new analyses)
Experiment design and protocol
Input datasets: context, characteristics, provenance
Analysis methods and workflows: at different granularities
Analysis results
Semantic information: datasets, results, workflows
Standards based, common data elements, and controlled vocabularies
Versioned, scalable, and computable semantics
Different metadata models for different data types
Linkages through backbone models, common attributes, ontologies
9/17/20189/17/2018

17. Metadata: Semantic Information
Data and semantic models to represent virtual slide related image, annotation, markup and feature information.
context relating to patient data, specimen preparation, special stains etc,
human observations involving Path classification and characteristics and
algorithm and human-described segmentations, features and classifications.
Semantic description of the computation being carried out, semantic and concrete identification of input and output datasets.
Desirable to have computable semantics
9/17/20189/17/2018

18. Identifiers
Databases of in silico experiments may be updated over time
Data may be moved to different archives
Globally uniquely identify an in silico experiment and its components
Should be possible to validate an identifier
Scalable mechanisms for generation, assignment, and resolution of identifiers
Leverage standards such as IHE PIX, LSID, URI
9/17/20189/17/2018

19. Federated Archives
May not be possible to store in silico experiments and its components in a centralized location
Multiple types of input data and results may be hosted in different systems
caArray
AIM and PAIS services
PACS and NBIA
Interoperability is critical
Common interfaces
Common information models: common data elements and controlled vocabularies
9/17/20189/17/2018

20. Semantic Query Capabilities
Ability to query metadata on experiments
Ability to query data and analysis results
Analysis results are semantically complex
Features, shapes, annotations
Use of domain ontologies
Scalable mechanisms for query
Large volumes of analysis results
Millions of nuclei in high-resolution microscopy images
9/17/20189/17/2018

21. Challenges
Structural and semantic metadata management: how to manage tradeoff between flexibility and curation
Semantic modeling infrastructures and policies able to scale to handle distributed systems with an aggregate of 109 or more data models/concepts.
How to curate changes to ontologies/data models and how to curate instance data in face of changing semantic models

22. Challenges Describe, store and replay analyses
Modeling needs to include a semantic description of the computation being carried out, curation/semantic description of input and output datasets.
Identifiers and version control
Computer assisted annotation and markup for very large datasets - large collection of cascaded algorithm/parameter dependencies challenge reproducibility

23. Thanks to:
In silico center team: Dan Brat (Science PI), Tahsin Kurc, Ashish Sharma, Tony Pan, David Gutman, Jun Kong, Sharath Cholleti, Carlos Moreno, Chad Holder, Erwin Van Meir, Daniel Rubin, Tom Mikkelsen, Adam Flanders, Joel Saltz (Director)
caGrid Knowledge Center: Joel Saltz, Mike Caliguiri, Steve Langella co-Directors; Tahsin Kurc, Himanshu Rathod Emory leads
caBIG In vivo imaging team: Eliot Siegel, Paul Mulhern, Adam Flanders, David Channon, Daniel Rubin, Fred Prior, Larry Tarbox and many others
In vivo imaging Emory team: Tony Pan, Ashish Sharma, Joel Saltz
Emory ATC Supplement team: Tim Fox, Ashish Sharma, Tony Pan, Edi Schreibmann, Paul Pantalone
Digital Pathology R01: Foran and Saltz; Jun Kong, Sharath Cholleti, Fusheng Wang, Tony Pan, Tahsin Kurc, Ashish Sharma, David Gutman (Emory), Wenjin Chen, Vicky Chu, Jun Hu, Lin Yang, David J. Foran (Rutgers)

24. Thanks!