Unraveling Multimodality with Large Language Models.pdf
Supporting the virtual physiological human with semantics and services e science 2011
1. Supporting The Virtual Physiological
Human With Semantics And Services
Dr. Carlos Pedrinaci
KMi, The Open University
2. Virtual Physiological
Human
“... a methodological and technological
framework that, once established, will
enable collaborative investigation of the
human body as a single complex system”
3. (Some)
VPH Challenges
• Data organisation and access
• Integration and interpretation of
heterogeneous data
• Creation of composable and reusable
analysis models
• Automated evaluation of hypothesis or
theories against available data
4. Components of a
VPH Workflow
Clinical / /
Clinical Analysis
Analysis Clinical Output
Clinical Output
biomedical data
biomedical data Personalise VPH models
Personalise VPH models Run simulations
Run simulations
Privacy, security, ethics Aggregate evidence, reduce uncertainty Support decisions for better
• Select model(s ) • Infer missing items health outcomes
• Images
• Retrieve data from: • Estimate parameters • Diagnosis
• Lab
• literature • Integrate data
• Genetic data • Treatment strategy
• population data • boundary conditions
• Lifestyle • Predictions
• EHR, PHS • functional behaviours
• ... • Prognosis
• ... •...
Examples
eu Heart
Fit patient images Compute organ
Patient Segment physiological function using Diagnostic index, suggestion
Patient to virtual Diagnostic index, suggestion
images biophysically based
images population DB models of treatment strategy
images models of treatment strategy
Comparative Molecular dynamics
Query DB, produce drug ranking DB simulation
HIV genotypic assay molecular model
HIV genotypic assay Literature: comparable Treatment suggestion
of patient of mutated HIV Treatment suggestion
of patient mutations w. HIV Drug ranking
drug resistance
5. Return Users
Select Retrieve Infer missing Run Results &
Workflow Existing Data items simulation Support
Patient Data
Workflow Inputs
Workflow Outputs VPH Outreach VPH-Share
Project No: 269978
Co-ordinator:
Application
University of
Patient Avatar
Personalised
Sheffield, UK
Model
Partners:
CYFRONET, PL
Sheffield Teaching
Hospitals, UK
ATOS Origin, ES
Kings College
London, UK
Universitat Pompeu
Fabra, ES
Empirica, DE
euHeart VPH OP SCS SRL, IT
@neurIST Patient Centred Computational Workflows ViroLab
NHS IC, UK
INRIA, FR
IOR, IT
Open Univ., UK
Philips Elec., NL
TU Eindhoven, NL
Univ. Auckland, NZ
Knowledge Knowledge Discovery Decision Support Uv Amsterdam, NL
UCL, UK
Infostructure
Management Data Inference Univ. Vienna, AT
AATRM, ES
FCRB, ES
Data Services: Compute Services Storage Services
Patient/Population
HPC Infrastructure Cloud Platform
(DEISA / PRACE) (Public / Private)
8. “Linking Open Data cloud diagram, by Richard Cyganiak and Anja Jentzsch. http://lod-cloud.net/”
9. Exposing VPH
Linked Data
• Provisioning of modular vocabularies
for capturing patterns of data
• Measurements, treatments, etc
• Assisted annotation of services and
DB2RDF mapping generation
• Interlinking, mapping, indexing
10. Outstanding issues Simplicity vs Expressivity vs Support
Controlled access to data
Anonymisation of records
Co-existence of different “unique IDs” for a single
entity (e.g., patient)
Large, heterogeneous, distributed, multi-party setting
13. Linked Services
• Linked Services are services described as
Linked Data (inputs, outputs, functionality...)
• That is, Linked Data describing reusable
functionality
• With supporting machinery Linked Services are
Linked Data consumers and/or producers
• Building blocks for Linked Data Applications
14. Linked Services and VPH
• Two main roles
• Controlled publication of data as
Linked Data on demand
• Supporting the creation of VPH
workflows using Linked Services as
processing activities
15. Dealing with
Sensitive Data
• Services for controlled access to the
data sources on demand
• DBs, RESTful services, Web Services
• Services used to expose heterogeneous
data as Linked Data on demand
• Declarative descriptions cover how to
deal with heterogeneous interfaces
30. “Linking Open Data cloud diagram, by Richard Cyganiak and Anja Jentzsch. http://lod-cloud.net/”
31. “Linking Open Data cloud diagram, by Richard Cyganiak and Anja Jentzsch. http://lod-cloud.net/”
32. Service Discovery
• Simple SPARQL-based
• Inputs/Outputs logic-based using
RDFS reasoning
• Functional Classifications with RDFS
reasoning, and over SKOS
• Similarity analysis
• Composition of these discovery types
33. Linked Services
Invocation
• Generic invocation engine OmniVoke
• Based on declarative descriptions
• RDF in, RDF out
• Supports RESTful and Web services
• Automated transformation of data
• Injection of provenance data
36. Ongoing Research
• Extension of workflow engine with
embedded Linked Services support
• Improve assisted annotation
• Cross-ontology logic-based discovery of
services
37. Thanks for your
attention
Contact: c.pedrinaci@open.ac.uk
Thanks to: Guillermo Alvaro, Irene Celino,
John Domingue, Jacek Kopecky, Ning Li,
Dong Liu, Maria Maleshkova