E-science grid facility for Europe and Latin America FISIOCOMP - Laboratory of Computational Physiology Computer Science Department Universidade Federal de Juiz de Fora (UFJF) Juiz de Fora - MG - Brazil Gustavo Miranda Teixeira Ricardo Silva Campos Heart Simulator
E-science grid facility for Europe and Latin America Group Professors Prof. Rodrigo Weber dos Santos, Dr. Math. * Prof. Marcelo Lobosco, Dr. Comp. Sci. * Prof. Ciro Barros Barbosa, Dr. Comp. Sci. Prof. Rubens Oliveira, Dr. Eng. Prof. Luis Paulo Barra, Dr. Eng. Prof. Elson Toledo, Dr. Eng. Master Students Carolina Xavier Ronan M. Amorim Franciane Peters * Grid team Undergraduate Students Caroline Costa Gustavo Miranda * Ricardo Campos * Guilherme Montebrune Former Master Students Rafael Sachetto Oliveira Fernando Otaviano Campos Bernardo Rocha Daves Martins Ely Fonseca
Itacuruça (Brazil), E2GRIS1, – Overview Computational physiology The heart Heart models Computational Framework Inverse Problems Gridification Goals 3
Itacuruça (Brazil), E2GRIS1, – Computational Physiology Physiology: The study of the (bio) functions Computational Physiology: The use and development of mathematical and computational models to describe biological functions 4
Itacuruça (Brazil), E2GRIS1, – Computational Physiology The bad news: –It is a wide gap connecting multiple scales, genes, proteins, cells, tissues, organs...; –multiple physics: quantum, molecular dynamics, chemistry, electro-mechanics…; 5
Itacuruça (Brazil), E2GRIS1, – Computational Physiology The models representation are based and depend on multiple and diverse data MODEL 6
Itacuruça (Brazil), E2GRIS1, – The Heart 7
Itacuruça (Brazil), E2GRIS1, – The Heart The blood pump Cells contract changing the organ geometry and the blood is expelled 8
Itacuruça (Brazil), E2GRIS1, – The Heart Cellular contraction: –An electric potential difference develops across the cell membrane and triggers a chain of electrochemical reactions that results in cellular contraction (intracellular Calcium spike, ATP, etc) 9
Itacuruça (Brazil), E2GRIS1, – The interior of the cells are connected by special proteins that allow the electric potential to propagate. A fast electric wave propagates and triggers heart contraction. The Heart 10
Itacuruça (Brazil), E2GRIS1, – Models of Cardiac Electro-Mechanics Cardiac disease is the #1 cause of death in the globe (30%) Today, computational models of the heart provide a better understanding of the complex phenomenon and support the development of new drugs, therapies, biomedical equipments and clinical diagnostic methods 11
Itacuruça (Brazil), E2GRIS1, – Bottom-up design –Sub-cellular and cellular mathematical models Models of Cardiac Electro-Mechanics 12
Itacuruça (Brazil), E2GRIS1, – Models of Cardiac Electro-Mechanics Bottom-up design –Tissue mathematical models: electric activity 13
Itacuruça (Brazil), E2GRIS1, – Models of Cardiac Electro-Mechanics Bottom-up design –Tissue mathematical models: mechanical coupling 14
Itacuruça (Brazil), E2GRIS1, – Models of Cardiac Electro-Mechanics Bottom-up design –Organ modeling 15
Itacuruça (Brazil), E2GRIS1, – Introduction to cardiac modelling Two basic components: 1) A cell model that describes the electric behavior of a single cell; 2) A tissue model which describes how the cardiac electric wave propagates from one cell to another 16
Itacuruça (Brazil), E2GRIS1, – Cell model Bi-lipid layer: Ionic channels: Special arrangement of proteins cut thru the membrane and allow the flow of specific ions, such as Sodium, Potassium and Calcium. Intracellular space Extracellular space Ionic channel 17
Itacuruça (Brazil), E2GRIS1, – ee ii CmCm I ion IcIc Cardiac cell models Hodgkin-Huxley based models Membrane works as a capacitor, isolating charges The ionic channel currents and the transmembrane potential satisfy a set of ordinary differential equations 18
Itacuruça (Brazil), E2GRIS1, – Cell models Canine ventricular model: Beeler-Reuter (9 eqs) Rabbit atrial model: Lindblad (27 eqs) Rat ventricular model: Pandit et al (26 eqs) Human atrial model: Nygren et al (30 eqs) Simplified ventricular model based on FHN (2 eqs) Guinea pig ventricular model: Luo-Rudy II (14 eqs) Human atrial model: Courtemanche et al (20 eqs) 19
Itacuruça (Brazil), E2GRIS1, – Cardiac Bidomain Model Tissue Model for cardiac electrophysiology Intracellurar and extracellular spaces (domains) modeled from an electrostatic point of view The coupling of the two domains is via non- linear cell modeling. Total cell membrane current spreads to both intracellurar and extracellular spaces 20
Itacuruça (Brazil), E2GRIS1, – Cardiac Bidomain Model 21
Itacuruça (Brazil), E2GRIS1, – Complex Models 22 Involves the coupling of several components (submodels) and data (geometry, biophysical parameters) Each component is a complex mathematical formulation, typically with tens of variables and hundreds of parameters New detailed models (components) are created and validated every week
Itacuruça (Brazil), E2GRIS1, – Modeling Challenges: Multi-scale and Multiphysics Computational Challenges: Simulations are computationally expensive (one heart beat = a couple of days in a parallel machine) Integration Challenge: Patient Specific Heart Model Complex Models 23
Itacuruça (Brazil), E2GRIS1, – Results 24
Itacuruça (Brazil), E2GRIS1, – We have a 2D simulator We needed a computational framework that would facilitate, stimulate and broadcast the use and benefits of cardiac modeling. The framework combines: The parallel simulator for bidomain-based models Cluster Computing An automatic code generator for models described by CellML User-Friendly Graphical Interfaces Web Server Results 25
Itacuruça (Brazil), E2GRIS1, – CellML XML based language (machine-readable) Describes mathematical models (MathML) Repository contains over 300 biophysical models A model is described via the connection of units, variables and components, in a hierarchical fashion 26
Itacuruça (Brazil), E2GRIS1, – CellML The goal: Accelerating the development of new models Computational Frameworks and tools On the way: Ontology and web semantic Grid Computing 27
Itacuruça (Brazil), E2GRIS1, – A couple of tools exist for edition, validation and simulation of models described in CellML Today two CellMl-based frameworks provide both cell and tissue level simulations: COR, a MS-Windows based environment, from the University of Oxford (cor.physiol.ox.ac.uk) AGOS, A web-based framework from FISIOCOMP- UFJF CellML-based tools 28
Itacuruça (Brazil), E2GRIS1, – Goal: Reach the biologists Computational Framework that hides many of the technical issues of cardiac modeling Agos Framework 29
Itacuruça (Brazil), E2GRIS1, – It provides support to cardiac electrophysiology modeling A editor to CellML language A translator of CellML code into C++ code A user-friendly Web form to setup parameters and visualize results Web Server Cluster Computing The Computational Framework 30
Itacuruça (Brazil), E2GRIS1, – API Generator to ODE Solutions Cellular models are described in CellML/MathML It translates CellML code into a object oriented C++ code Through the API generated, it is possible to simulate the model and setup parameters Agos Translator 31
Itacuruça (Brazil), E2GRIS1, – Tissue Model 32
Itacuruça (Brazil), E2GRIS1, – Inverse Problem 1.Forward Models of Cardiac Physiology 2.Inverse Problem 33
Itacuruça (Brazil), E2GRIS1, – Inverse Problem The forward problem –The user has to know all parameters, such as geometry of the organ and values of conductivity –It returns the potential diference along the time and space Inverse problem –The user knows the potential diference –He or she may want calculate the geometry and all another parameters 34
Itacuruça (Brazil), E2GRIS1, – Inverse Problems Estimate the values of electrical activity on the cardiac tissue Given a number of observed transmural electrograms estimate possible changes on the conductivity ( , ) of a known and specific region of the heart. 35
Itacuruça (Brazil), E2GRIS1, – Pathological Tissue Region with altered conductivity (,)(,) Motivation: focal variations of tissue conductivity values (both intra and extra) are observed in many different cardiac diseases: Acute ischemia, Infarct, Chagas Disease, Myocarditis Inverse Problems 36
Itacuruça (Brazil), E2GRIS1, – Inverse Problem More computational costly than the foward problem It solves the forward problem lots of time sequentionally InvCell and InvTissue 37
Itacuruça (Brazil), E2GRIS1, – INVCell We are adjusting a model which GA takes one day long to run. Asynchronous x Synchronous. –Heterogeneity x Homogeneity. It uses the AGOS API lots of times –ODEs are solved sequentionally 38
Itacuruça (Brazil), E2GRIS1, – InvCell Genetic algorithm –Based on Darwin’s evolutionary theory –Aims to optmization (maximize/minimize) –It works simulating the process of natural reproduction, mutation, and selecting the fittest individual
Itacuruça (Brazil), E2GRIS1, – INVCell GA implementation: –The individuals are the parameters –We know the solution – calculated by the simulator –Each iteration gets more closer to the final solution –Parallel GA – master-slaves. –Floating point representation; –Elitist selection; –The initial population is randomly generated ; –A new generation depends of their parents; 40
Itacuruça (Brazil), E2GRIS1, – INVTissue It solves an inverse problem associated to the simulation of cardiac tissue models. It also has an implementation of a Genetic Algorithm parallelized with MPI. It runs the simulator to each individual Quite slow! 41
Itacuruça (Brazil), E2GRIS1, – INVTissue Investigate the solution of an inverse problem associated to cardiac electrophysiology The goal is to estimate values for the electrical conductivity of cardiac tissue, taking as known some information concerning the electrical activity of the heart Asynchronous non generational GA Parallelized using master-slave 42
Itacuruça (Brazil), E2GRIS1, – Goals Porting InvCell –It should be the easiest; Porting InvTissue –More complicated – lots of dependencies; Porting of a basic version of the Heart Simulator –Hardest problem; 43
Itacuruça (Brazil), E2GRIS1, – Goals The heart simulator uses : –C code –Petsc library –MPI Numerical methods to solve lots of equations Each iteration have lots of dependencies on the previous one 44
Itacuruça (Brazil), E2GRIS1, – Questions … 45