Transcript Slides - Association for Pathology Informatics

Essential Elements For Semiautomating Biological And
Clinical Reasoning In Oncology
Roger S. Day, William E. Shirey, Michele Morris
University of Pittsburgh
Big
in Modeling of Cancer
What are cancer models good for?
–
–
–
–
Discovering general principles
Professional training
Prediction for planning experiments
Description of natural history, distinguishing
mechanisms & explanations
– Prediction for individualizing treatments
Educational Resource for Tumor Heterogeneity
“ERTH”
• Develop a computer “playground” for thinking
broadly about cancer
• Develop wide range of learning applications
• Field test, evaluate, deploy, disseminate

Oncology Thinking Cap
“OncoTCap” software
Why is tumor heterogeneity
important?
• Spatial heterogeneity  metastasis
It kills people.
• Genetic/epigenetic heterogeneity within tumors
 survival of the fittest
 immortalization, motility, invasion, metastatic potential, recruitment
of blood vessel, resistance to apoptosis, resistance to therapy
 resistance to patient’s defenses
• Natural intuition about POPULATION DYNAMICS
is poor
Tumor heterogeneity
A missing link in the big picture
“Cancer
Genome
Anatomy”
????
What happens
to patients
Population dynamics,
Toxicity,
Drug interactions,
Doctor/patient,
“Society of cells”,
…
INFORMATION SYNTHESIS
Reductionism, then holism
OncoTCap 4/Cancer Information Genie
The software platform: “Protégé”
An expert knowledge acquisition system
protégé.stanford.edu
Frame-based KB,
compliant with OKBC.
The standard “tabs”
Ontology development
Forms editor
Instance capture
OncoTCap 4:
mission creep is a good thing
• Clinical trials bottleneck:
–
–
–
–
Accrual
Time
Expense
Far “faaar” too many hypotheses to test
• Choosing which trials to do… today:
– Due diligence information gathering– by hand
– Model-building and prediction – by intuition
• What if…
– Information gathering is empowered
– Model-building/validation/prediction is empowered
Three workflows
• Knowledge capture
• Mapping from a catalog of statement templates
to computer model-driving code
• Building modeling applications like tinker toys
OncoTCap 4 “Tricorn”
Knowledge
capture
work process
Code-mapping
work process
Applicationbuilding
work process
Workflow #1:
Information capture
•Automated field capture
•Full-text location, script-driven
Workflow #1: Information capture
Assessments
An example of the work flow



.
Workflow #2: Coding catalog
Example of a statement template:
A WT gene locus for gene gene name can mutate to MUT
with rate mutrate
Representation in statement bundles:
The gene [gene name] has values WT/WT, WT/MUT, MUT/MUT.
The mutation rate for [gene name]
from WT/WT to WT/MUT is 2 times [mutrate]
The mutation rate for [gene name]
from WT/MUT to MUT/MUT is [mutrate]
Workflow #3: Model controllers
Workflow #3:
A Validation Suite model controller
NLP and OncoTCap?
• Plug in new tools for locating published
resources (like MedMiner, EDGAR).
• Parse captured text, identify concepts, map to
keyword tree.
• Provide a conduit to other Ontologies, to import
portions into our Keyword tree.
• Replace user-defined Keywords with standard
terms from other Ontologies.
• Suggest “interpretations”– mappings into catalog
of StatementTemplates.