Team Science for Precision Medicine:
The Utah PRISMS Informatics Center
Professors Kathy Sward and Julio C. Facelli
2016 Air Quality, Health and Society Symposium,
September 21st 2016
Center for Clinical &
Translational Science
Genome
Exposome (t)
Phenotype (t)
Environmental Data
Educational Data
Socioeconomically Data
Sensors
Interment of Things
…..
Precision Medicine Framework
Exposome – Emerging concept
– Total set of environmental factors to which a person is
exposed, including the complex interplay between
environmental factors and between environmental, behavioral,
& psychosocial factors
– Influence health and diseases modifying the genome
http://www.niehs.nih.gov/research/supported/dert/programs/peph/podcasts/exposome/
Exposome
• the total set of environmental
factors to which a person is
exposed
– the complex interplay between
environmental factors
– and between environmental, behavioral
& psychosocial / socioeconomic factors;
– that in turn influence health and disease
• Encompasses life-course of
exposures from prenatal period
onwards.
• Complements genome by providing
a comprehensive description of
lifelong exposure history.
General External Environment
Specific External
Environment Internal
Environment
Overlapping domains within exposome
5
Exposomics • Study of defining, generating and utilizing exposomes in biomedical
research.
• Ongoing efforts: – HELIX: Early life exposome
– EXPOsOMICS: Assess exposures
– HEALS: Studies exposure to environmental stressors and health outcomes
– NIH’s Environmental influences on Child Health Outcomes (ECHO) Program. Understanding the effects of environmental exposures on child health and development
• Requires a systems biology approach. Body's responses to environmental influences including endogenous metabolic processes that can alter or process the chemicals to which humans are exposed.
• ‘Expotying’: Exposure of a biological entity usually with reference to a specific characteristic under consideration. Also called as Exposome Informatics, Exposure Information Science.
• Provides great opportunities to Biomedical Informatics
Utah
• Inversions – days of good air vs bad air
• Lots of complementary research
– Environmental/air quality
– Sensors
– Pediatric asthma
– Informatics
• High level of interest in the community
• Technical infrastructure is robust
• Robust “knowledge infrastructure” (expertise)
http://utahpoliticalcapitol.com/wp-content/uploads/2013/04/inversion01.jpg
Posssible Sensors
• Goal: Use commercial wireless devices (IoT / smart
home)
• Air: PM, O3, NO2, SO2, CO, ..., VOC, temp, humidity
• Phone: position, INS (acc, gyro, mag), noise, light
• Wearable: Heart rate, INS, activity
• Sleep: respiration rate, heart rate, movement, duration
• Home: door/window, occupant locations, elec appliance usage,
furnace/ac control, video, audio
Variety: What is measured
• Particulate Matter
• PM2.5 ,PM10
• Small particles, large particles
• Some categorize more
granularly – suspended
particulate matter, respirable
particles, coarse/fine/ultrafine
– some separately categorize
“soot” https://en.wikipedia.org/wiki/Particulates
PRISMS Challenges
PRISMS program research will need:
The Utah PRISMS Informatics Center
Center for Clinical &
Translational Science
Team Leadership
• PD: Sward, Facelli
• CoN: College of Nursing; BMI: Department of Biomedical Informatics,
School of Medicine; CCTS: Center for Clinical and Translational Science
• Project manager: Heather Oldroyd (Pediatrics)
CoN, BMI BMI, CCTS
Admin core: project oversight, administration & reporting, evaluation,
coordination. Cross-project & “all hands” meetings.
Project 1: Sensor interface and
subject interaction
• Team
– Leads: Patwari, Meyer
– Key Collaborators: Collingwood, Kim
Electrical & Computer
Engineering Scientific Computing
& Imaging
Pediatrics Bioengineering
P1: Specific Work Areas
– Data pipeline, aka “what happens in the home”
– Bi-directional – gathering data from sensors, presenting
information to parents and kids
– Mobile apps, sensor communications, web portal, age-
appropriate messaging
Project 1 Status • We have a fully functional (version 1.0)
hardware and software system
– collects data from wireless sensors to a
raspberry pi-based device
– sends data to a cloud based server using
the home’s broadband internet
– where it can be visualized by the researcher.
Project 1 Status • The configuration of the gateway is automated
via a script and a configuration file which can
be edited to meet researcher specifications
18
Dylos DC1100 Pro Air
Quality Monitor
Web Interface showing a graph of a single day
during a home deployment Web Interface showing a pie chart for
daily activities for Monitor x
Air Quality Application showing events of
air quality activities on a smart device
Home Smart Device
Air Server
Project 1 Deployments
• Deployed system in two homes for 3-4
months each.
– Debugging, performance testing,
examination of data quality
• We conducted experiments regarding
calibration to increase data reliability and
reduce variability between sensors
Deployment B: Time Series of PM2.5 & Events
Dustin
g
Cleaning
w/
cleaning
solution
Vacuuming
Many
people in
the room
All 11 sensors in one room: 6 of them plotted here
Grafana: Graph and dashboard builder for visualizing time series metrics
Prototype data visualization
Project 2: Informatics architecture
(Federation-Integration Platform)
• Team
– Leads: Gouripeddi, Facelli
– Technical lead: Madsen
– Key Collaborators: Kelly, Horel
BMI, CCTS CCTS
Chemical
Engineering Atmospheric
Sciences
Project 2: Specific Work Areas
• Core infrastructure
• Builds on and extends work from the CCTS
OpenFurther architecture
• Model driven, standards based, open source
• Makes digital objects FAIR (Findable, Accessible,
Interoperable, Reusable) – OpenFurther has FAIR as its main requirement and has been so since its
inception
Project 2: Specific Work Areas
– Sensor Data Harmonization Framework
• Integration and Federation of data from multiple sensors, sensors &
clinical/research data
• logical data model to store and harmonize metadata from environmental
sensors
– Sensor - Environmental Mathematical Modeling.
• High resolution spatial-temporal grid along with uncertainties.
– Data integration and interim storage.
• Integrate mathematically modelled data with collected data.
• Preprocessing before passing information on to the DCC
Challenges and Informatics
Methods and Solutions Data Sources
Mathematical Modeling
Uncertainty Characterization
Data Integration
• Semantics
• Metadata
• Time & Event Modeling
• Infrastructure for multi-scale, multi-omics integration
Presentation/Visualization
Opportunities
• Leverage CTSA biomedical informatics
core
• While the PRISMS emphasis is currently
on the incorporation of data from mobile
and stationary sensors, our infrastructure
will be adaptable and scalable to other
emergent measurement modalities that
are becoming part of the “internet of
things”.
Semantics for Data Integration
• Stored in Terminology/Ontology Server
• Examples – Semantic Sensor Network Ontology: Describes sensors and
observations, and related concepts.
– Sensor Model Language (SensorML): Standard models and XML schema for describing sensors systems and processes associated with sensor observations.
– PhenX Phenotypic Terms: Standard measures related to complex diseases, phenotypic traits and environmental exposures.
– Exposure ontology (ExO): Facilitate centralization and integration of exposure data to inform understanding of environmental health.
– Standard biomedical ontologies and terminologies: Gene Ontology, UniProt, SNOMED
Metadata • Stored in Metadata
Repository
• Relational or graph stores
• Stores
– Source and Central Data
Models
• Harmonized sensor
data model
– Data provenance and
associated uncertainty
– Inter-model
transformative functions
Air Quality Mathematical Models
•Validated on the east coast
•Doesn’t consider Altitude
•12 kilometer resolution
•Hierarchical Bayesian model
Environmental Protection Agency – Center for Disease
Control Model
•Describe regional and small-scale spatial and temporal gradients
•Uses measured PM concentrations, monitoring site location, GIS-based location-specific characteristics and location-and month-specific meteorological data, and spatial smoothing of monthly and long-term averages
Generalized Additive Mixed
Models
McMillan, Nancy J., David M. Holland, Michele Morara, and Jingyu Feng. “Combining Numerical Model Output and Particulate Data Using Bayesian Space–time Modeling.”
Environmetrics 21, no. 1 (February 1, 2010): 48–65. doi:10.1002/env.984.
Yanosky, Jeff D., Christopher J. Paciorek, Francine Laden, Jaime E. Hart, Robin C. Puett, Duanping Liao, and Helen H. Suh. “Spatio-Temporal Modeling of Particulate Air Pollution in
the Conterminous United States Using Geographic and Meteorological Predictors.” Environmental Health 13, no. 1 (August 5, 2014): 63. doi:10.1186/1476-069X-13-63.
• Fill gaps in measured data with mathematically models
• A library of AQ data models to provide high spatio-temporal resolution with a framework validate the model output.
Uncertainty Quantification
PRISMS Federated Data Integration Architecture
Study Generation Tools
Experiment
Parameters
Code for
devices
(if needed)
Code for
gateway
Code for
phone
app
Code for
server
Study
Design
Sensors, Data Rates,
Subject Feedback,
Subject Surveys
New types of studies
Behavior, Sensing, and Feedback
Two (complementary) options for
feedback:
1. Based on raw data
2. Based on (estimated) activities
– inferred from sensor data
Feedback:
Data, Alerts
Behavior,
Activities
Sensor
Data
Analysis
Project 3: Research Platform
• Team
– Leads: Sward, Nkoy
– Key Collaborators: Stone, Cummins, Wong, Meyer
(Facelli)
Pediatrics, BMI
Pediatrics CoN, BMI CoN
CoN
Project 3: Specific Work Areas
Help asthma researchers to conduct studies
– Requirements and use cases
– Sensor library
– Research platform.
• View into the sensor library - for researchers, parents, sensor
developers. What does each group want/need to know?
• Data visualization, big data analytics
– End to End testing
Project 3: Perspective
• Usability, ease of use. Work with projects 1 and 2
for iterative user-centered design approach
• Parent/child advocacy
– What do parents want to know before participating in this
sort of research?
• We will develop ways to link data from different
types of sensors, with other kinds of information
– How do we support research?
– How do we make the information meaningful for
families, clinicians, researchers, and sensor
developers?
Results from Project 3
Results from Project 3 • Use case archetypes
Use cases identify information flow…
and data requirements Use Case C. (This use case is derived from a recently published manuscript by
Rosalyn Singleton of the Alaska Native Tribal Health Consortium and colleagues,
“Housing characteristics and indoor air quality in households of Alaska native
children with chronic lung conditions”).
This study focused on association between childrens’ respiratory
symptoms and measurements of indoor air quality. In order to measure
indoor air quality, the investigators deployed the following sensors: TSI
Inc. DustTrac Models 8530/8533 (PM2.5), Ultra III Passive Badge(VOCs),
Radiello 130 Passive Badge (VOCs), TelAir 7001 Sensor with a HOBO
U12 data logger (CO2, temperature, and relative humidity), ExTech
CO210 Indoor Air Quality CO2 Monitor (CO2, temperature, and relative
humidity), and Lascar Electronics Lascar300 (CO). …The sensors
generated data at varied time intervals …
Synergy
• Collaboration across projects
• Collaboration across PRISMS centers
– Data Modeling Working Group
– Discussions with Environmental Defense
Fund and other groups (examining
standards)
• Collaboration across similar work on
campus (DEQ/PRISMS)
New Collaborations Emerging
• Lots of interest in broader
community
• Lots of interest in
academic community
• Opening doors to new
collaborations
Our vision
A center where researchers and
practitioners integrate genetic,
environmental, behavioral and
clinical information to advance
precision medicine and health care
transformation for better
management of pediatric asthma.
Our Dream
Imagine this scenario…
The hospital command center predicts, with 87%
accuracy, that according to environmental
conditions tomorrow your clinic will have 47 patients
with exacerbated asthma conditions. According to
their genetic profile 59% should be treated with
protocol A, 37% with protocol B, and you should
make arrangements to admit the remaining 4 for
which serious complication can be expected.
PRISMS informatics center is the first step
Acknowledgements
This effort was supported by U54EB021973, National Institute of
Biomedical Imaging and Bioengineering, NIH and University of Utah Air
Quality Seed Grant Program. OpenFurther supported by NCRR/ NCATS
Grants UL1RR025764 and 3UL1RR025764-02S2, National Center for
Clinical and Translational Science 1UL1TR001067, University of Utah
Research Foundation, grant 1D1BRH20425 (DHHS), and R01
HS019862 from AHRQ, (DHHS).