OpenCMISS & CellML Chris Bradley 1,2 & David Nickerson 1 1 Auckland Bioengineering Institute, University of Auckland. 2 Department of Physiology, Anatomy and Genetics, University of Oxford.

Why OpenCMISS? We want to solve larger scale coupled models of the heart, lungs, stomach and other biological systems than we can currently solve We need to redevelop legacy CMISS computational code. Take advantage of modern High Performance computer architectures.

OpenCMISS Re-engineering of CMISS Main website – Open source on github – Fortran 95/2003. Object “based”

OpenCMISS Design Goals 1.It should be a library based rather than a monolithic application. 2.It should use the Physiome standards FieldML and CellML. 3.It should be a general code. 4.It should be an inherently parallel code. 5.It should be used, developed, and understood by novices and experts alike. 6.It should deal with multiscale and multiphysics applications and with complicated workflows.

OpenCMISS library openCMISS librarycore library openCMISS (cm) FieldML API CellML API openCMISS (cm) FieldML API CellML API openCMISS (cm) CellML API FieldML API FieldML file CellML file Processors History file Language Bindings Fortran C/C++ Python Etc. Language Bindings Fortran C/C++ Python Etc. Language Bindings Fortran C/C++ Python Etc. Language Bindings Fortran C/C++ Python Etc. OpenCMISS (cm) FieldML API CellML API External Program or script

OpenCMISS application openCMISS (cm) FieldML API CellML API Com Layer openCMISS (cm) FieldML API CellML API openCMISS (cm) CellML API FieldML API Com Layer FieldML file CellML file OpenCOR Processors CellML API FieldML API Com Layer History file FieldML API CellML API Com Layer OpenCMISS Cmgui Network

General parallel environment Bus CC CC Cache CCCCC CCCCC CCCCC CCCCC CCCCC Memory CPU Accel CC CC Cache CCCCC CCCCC CCCCC CCCCC CCCCC Memory CPU Accel Memory Network Bus CC CC Cache CCCCC CCCCC CCCCC CCCCC CCCCC Memory CPU Accel CC CC Cache CCCCC CCCCC CCCCC CCCCC CCCCC Memory CPU Accel Memory Network Hierarchy of heterogeneous computing elements

Multi-scale & Multi-physics In order to accurately model biological systems OpenCMISS must model multiple physical systems at different time and space scales. OpenCMISS deals with this via: –A flexible system for describing multiple models, solvers and workflows. –Methods for coupling different physical systems together. –Handling different spatial and temporal scales via FieldML and CellML

Multi-physics: Equations Sets Can be broken down into –Linear/Nonlinear –Dynamic/Static General form of OpenCMISS equations: DynamicLinearNonlinearSourceRHS

Example – Electrical Activation (Bidomain representation) Cell Membrane Intracellular Space  i,  i Extracellular Space  e,  e I ion Cell model =

Bidomain equations Where: –V m = transmembrane potential –  i = intracellular potential –  e = extracellular potential –A m = membrane surface/volume ratio –C m = membrane capacitance –  e = intracellular conductivity tensor –  i = intracellular conductivity tensor –I ion = ionic source current from cellular model –I m = transmembrane stimulus current –i i = current injection per volume intro intracellular space –i e = current injection per volume intro extracellular space

Cellular processes

Problem/Field description Fields: Geometric Fibre Material (  i,  e, A m, C m etc.) Potential (V m,  e, etc.) Source (I ion, etc.) Cellular state variables Cellular intermediate variables Cellular parameter variables

Field parameter vector Field variable, u Component, u 1 Component, u 2 Parameter vector, x Global DOFs

Electrophysiology - CellML 300% 100% gNa Distribution +50 mV -85 mV Stimulus off +50 mV -85 mV Normal activation +50 mV -85 mV gNa activation

Finite Elasticity – Breast Mechanics Courtesy of Prasad Gamage Skin Breast Tissue Undeformed Deformed Under gravity loading

Fluid Mechanics – 3D NS Carotid Artery Flow, Re=100 MeshVelocityPressure Courtesy of David Ladd

CellML as a DOF black box Field variable Parameter DOF Vector Component 1 Component 2 Component 3 InputsOutputs

OpenCMISS CellML Import void computeRates(double VOI, double *PARAMETERS, double *RATES, double *STATES, double *INTERMEDIATE); { const double FIXED[..]; : } StateParametersIntermediate StateParametersIntermediate CellML Variables CellML Environment CellML Model Import Set as known Set as wanted Set as known and wanted

CellML Interface ModelsStateParametersIntermediate CellML environ OpenCMISS Field CellML Fields

CellML/Field Maps ModelsStateParametersIntermediate CellML environ CellML to field MapField to CellML MapCellML integrate Cellular model Transmembrane Voltage Field

CellML/Field Maps ModelsStateParametersIntermediate CellML environ Strain Field Stress Field CellML to field Map Field to CellML Map CellML evaluate Constitutive Law

FPGA Acceleration Investigating using Field Programmable Gate Array (FPGAs) as accelerators for CellML equations (see Ting’s talk next).

GPU Acceleration

Acknowledgements A (hopefully complete) list of those individuals who have contributed to OpenCMISS (so far….) Bojan Blazevic Andy Bowery Chris Bradley Randall Britten Vincent Budelmann Phani Chichapatnam Richard Christie Andrew Cookson Andrew Crozier Prasad Gamage Arne Gjuvsland Thomas Heidlauf Alice Hung Peter Hunter Jagir Hussan Chloe Irwin Whitney Jessica Jor Sebastian Krittian David Ladd Sander Land Jack Lee Caton Little Xaio Bo Lu Kumar Mithraratne Christian Michler Jennine Mitchell Martyn Nash David Nickerson Steven Niederer Poul Nielsen Øyvind Nordbø David Nordsletten Stig Omholt Ali Pashaei Vijayaraghavan Rajagopal Adam Reeve Oliver Rohle Ishani Roy Ole W. Saastad Soroush Safaei Vickie Shim Matt Sinclair Nic Smith Martin Steghofer Merryn Tawhai Mark Trew Jon Olav Vik Tim Wu Robert Jiahe Xi Nancy (Xiani) Yan Ting Yu Heye Zhang Ju Zhang