The Algorithms of Life - Scientific Computing for Systems Biology
2 likes1,049 views
The document discusses high-performance computing (HPC) applications in systems biology, focusing on the algorithms involved in tissue formation and cellular communication. It details the development of the OpenFPM library for scalable parallel simulations and outlines various computational methods for modeling biological processes. Additionally, the document highlights ongoing research and collaborative efforts within the scientific community to enhance simulation and visualization techniques in biology.
![Our approach: 1 Platform
High-Performance Computing
Real-time
big data
numerical
simulation
AR/VR
Learning
Algorithms
(a) (b)
(d) (e)
page 7 of 20
OpenFPM core OpenFPM numerics
C++ Template Metaprogramming
Open Particle Mesh Environment (OpenPME)
Distributed memory
Shared memory
GPUs
Vector processors
DSL optimizer and code generator
Figure 3: Layers in the envisioned OpenFPM/
OpenPME stack.
2. It is an object-oriented architecture [ADS10] with
enous multi-core platforms [AS14], as well as support](https://image.slidesharecdn.com/keynoteivo-sbalzarini-190731210513/85/The-Algorithms-of-Life-Scientific-Computing-for-Systems-Biology-5-320.jpg)
![Scientific Computing for Systems BiologyIvo F. Sbalzarini !9
Simulation fits Experiment
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 1 2 3 4 5
FRAP[-]
time [s]
Simulation and
Experiment in the
same ER
Simulation
Experiment
in vivo diffusion
constant from
fit
⌫ = 34 ± 0.95 µm2
/sD =?
Sbalzarini et al., Biophys. J. 89(3):1482. 2005.](https://image.slidesharecdn.com/keynoteivo-sbalzarini-190731210513/85/The-Algorithms-of-Life-Scientific-Computing-for-Systems-Biology-9-320.jpg)
The Algorithms of Life - Scientific Computing for Systems Biology
- 1. The Algorithms of Life - Scientific Computing for Systems Biology - Ivo F. Sbalzarini MOSAIC Group Center for Systems Biology Dresden TU Dresden & MPI-CBG 1 Conference Keynote ISC High-Performance Frankfurt, June 17, 2019
- 3. Algorithms of tissue formation Cells execute programs: § Genetic programs § Communication programs § Decision making § Signaling networks § Endosomal sorting § … Cells communicate by: § chemical signals § mechanical signals § cell-cell contacts § motion § … Meyerowitz, Caltech
- 4. What we want to do… HPC for Life Simulation (<1s step) Real-time analysis (1-5 GB/s) Real-time visualization (1.77 Gpix/s) Photo: SZonline PPM_RC OpenFPM
- 5. Our approach: 1 Platform High-Performance Computing Real-time big data numerical simulation AR/VR Learning Algorithms (a) (b) (d) (e) page 7 of 20 OpenFPM core OpenFPM numerics C++ Template Metaprogramming Open Particle Mesh Environment (OpenPME) Distributed memory Shared memory GPUs Vector processors DSL optimizer and code generator Figure 3: Layers in the envisioned OpenFPM/ OpenPME stack. 2. It is an object-oriented architecture [ADS10] with enous multi-core platforms [AS14], as well as support
- 6. Learning equations (PDE) from images DATA Inference box ODE/PDE Grill lab (MPI-CBG) Drescher lab, MPG Maddu et al., NYSDS, 2019
- 7. Scientific Computing for Systems BiologyIvo F. Sbalzarini !7 Image: NEC NEC SX-5 CSCS, Switzerland 512 Processors 2002
- 8. Scientific Computing for Systems BiologyIvo F. Sbalzarini !8 Sbalzarini et al., Biophys. J. 89(3):1482. 2005. 1 billion particles 242 processors 84% efficiency @250 GFlop/s Custom F90+MPI code, 3 years 2002
- 9. Scientific Computing for Systems BiologyIvo F. Sbalzarini !9 Simulation fits Experiment 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 1 2 3 4 5 FRAP[-] time [s] Simulation and Experiment in the same ER Simulation Experiment in vivo diffusion constant from fit ⌫ = 34 ± 0.95 µm2 /sD =? Sbalzarini et al., Biophys. J. 89(3):1482. 2005.
- 10. Scientific Computing for Systems BiologyIvo F. Sbalzarini !10 Image: cray.com CRAY XT-3 CSCS, Switzerland 1664 Processors 2005
- 11. Scientific Computing for Systems BiologyIvo F. Sbalzarini !11 2005 Sbalzarini et al., Biophys. J. 90, 2006.
- 12. Scientific Computing for Systems BiologyIvo F. Sbalzarini !12 1952 OPINION Turing’s next steps: the mechanochemical basis of morphogenesis Jonathon Howard, Stephan W. Grill and Justin S. Bois Abstract|Nearly60 yearsago,AlanTuringshowedtheoreticallyhowtwochemical species,termedmorphogens,diffusingandreactingwitheachothercangenerate spatial patterns. Diffusion plays a crucial part in transporting chemical signals and explicitly stated that both pa be taken into account. Because “ dependence of the chemical and data adds enormously to the dif he “proposed to give attention ra cases where the mechanical aspe ignored and the chemical aspect significant,” although he did sug this difficulty might be circumv “the aid of a digital computer.” Indeed, it has become clear th mechanical processes, such as tr along cytoskeletal filaments, cyt flow and endocytosis, also have roles in patterning at the cell and PERSPECTIVES 2011
- 13. Example: dorsal closure in Drosophila Elisabeth Knust wt Crb Y10A Bridge from a molecular perturbation to a tissue-level dynamic phenotype.
- 14. Biological Mechanics: active polar gels Frank Jülicher, MPI-PKS Prost, Jülicher, Joanny (Nature Physics) PROGRESS ARTICLES | INSIGHT NATURE PHYSICS kp kd kp Myosin motor Actin filament Actin monomer Figure 1 | Illustration of an active gel consisting of actin filaments, myosin motors and passive crosslinks (not shown). Filament polymerization and depolymerization processes are indicated by the rates kp and kd. a vector field as a slow variable. From there, several approaches may be chosen. The simplest approach, first used to describe active nematics, considers small perturbations with weak gradients (long equations (1) and (2)). This term can graining a microscopic description in wh are represented by force dipoles9,10 . The re tive stress is the average force dipole de nematic symmetry the anisotropic part o most important new term compared to situation where the total anisotropic stres zero by appropriate boundary conditions spontaneous shear at short times and a s (Box 1, equation (3)). The continuity equation describing the actin mass has both source and sink terms mass is conserved (Box 1, equation (4)). Al for scalar densities have not only convecti but they have additional terms character both the bend and splay of a nematic dire of a vector and, owing to the absence of they can enter the expression for the flux a system with polar symmetry, there are f this symmetry and the absence of time-rev equation (6)). First, the fluxes correspondin quantities have a term describing a spontan to the barycentric velocity along the polar connection with self-propelled systems. In First-ever simulation of this model using Particle Methods. Polarity field Density field Chemical potential field Velocity field Pressure field PDEs Simulation
- 15. Friction! Friction! ) Model illustration. Model active polar viscous layer sandwiched between two external m r spans a length of Lx in the x-direction and Ly in the y-direction. The layer is sandwiched passive membrane active material passive material Application to Embryo Ramaswamy & Jülicher, Nat. Sci. Rep. (2016)
- 16. Novel behavior predicted Static Flows Traveling vortices Chaos (a) relative myosin activity Lyapunovexponent unstable stable layer sandwiched between two external media. The sandwiched n the y-direction. The layer is sandwiched between y = 0 and (b) (a) Steady-state polarity field across the computational domain at he Franck free-energy density of the polarity field (Eq. 5) across the Ramaswamy & Jülicher, Nat. Sci. Rep. (2016)
- 17. Example: dorsal closure in Drosophila Elisabeth Knust wt Crb Y10A Bridge from a molecular perturbation to a tissue-level dynamic phenotype.
- 18. Numerical method: Particle-Mesh R. Ramaswamy et al. / Journal of Computational Physics 291 (2015) 334– Ramaswamy et al., JCP, 2015 - Particles move and interact - Hydrodynamics on the mesh - Particle-Mesh interpolation
- 19. Scientific Computing for Systems BiologyIvo F. Sbalzarini Particle Methods as a Unifying Computational Framework 19 Particle-particle interactions can be deterministic or stochastic Particles have positions and properties they carry. Continuos Discrete DeterministicStochastic Partial Differential Equation (PDE) Molecular Dynamics Stochastic Differential Equations (SDE) Agent-based Models
- 20. Scientific Computing for Systems BiologyIvo F. Sbalzarini Particle Methods for Continuous Problems 20 Bergdorf et al., J. Math. Biol. 61, 2010.
- 21. Scientific Computing for Systems BiologyIvo F. Sbalzarini Particle Methods for Discrete Problems 21
- 22. Scientific Computing for Systems BiologyIvo F. Sbalzarini Particle Methods for Image Analysis 22 Paul et al., IJCV 104, 2013.
- 23. Scientific Computing for Systems BiologyIvo F. Sbalzarini 23 Particle Methods for Optimization Müller et al., IEEE CEC, 2009.
- 24. Scientific Computing for Systems BiologyIvo F. Sbalzarini Particle Methods as a Unifying Computational Framework 24 ✦ Discretize and numerically solve mathematical models in arbitrarily complex geometries. ✦ Mimic the physical interactions between the discrete constituents of a system (proteins, animals, …). ✦ Combine with meshes for far-field interactions. ✦ Moment-conserving particle-mesh interpolation is available. Particles interact with each other in order to: dxp dt = NX q=1 K(xp, xq, !p, !q) d!p dt = NX q=1 F(xp, xq, !p, !q)
- 25. Scientific Computing for Systems BiologyIvo F. Sbalzarini !25 Past 15 years: PPM Library (Fortran 90, then 2003) Sbalzarini et al., JCP, 2005; Awile et al., INCAAM, 2010; Awile et al., PARTICLES, 2013. A:14 PPM language PPM core PPM numerics single processordistributed memoryshared memoryvector Message Passing Interface (MPI) PETSc METIS FFTW PPM Environment Fig. 10. PPME is a new access layer to the underlying PPM library.
- 26. Scientific Computing for Systems BiologyIvo F. Sbalzarini Prior Use of the PPM Library !26 ✦ Vortex methods for incompressible fluids → 10B particles, 16’384 processors (BG/L), 62% efficiency1,2 ✦ Smooth Particle Hydrodynamics for compressible fluids → 268M particles, 128 processors, 91% efficiency3 ✦ Particle diffusion → 242 processors, 84%, up to 1B particles3 ✦ Discrete element methods → 192 processors, 40%, up to 122M particles4 ✦ Molecular dynamics → 256 processors, 63%, up to 8M particles5 (a) (d) Fig. 3. Visualization of individua an inclined plane. Initially, 120 00 radii between 1.00 and 1.12 mm a 0 the box is inclined at an angle avalanche down the plane 0.1, 0.2, 2. Daerr, A., Douady, S.: Two ty 241–243 1. P. Chatelain et al., Comput. Method.Appl. M., 197, 1296, 2008 2. P. Chatelain et al., Lect. Notes Comput. Sc., 5336, 479, 2008 3. I.F. Sbalzarini et al., J. Comp. Phys., 215(2), 566, 2006 4. J.H.Walther and I.F. Sbalzarini, Eng. Comput., 26(6), 688, 2009 5. unpublished
- 27. Scientific Computing for Systems BiologyIvo F. Sbalzarini !27 The OpenFPM Library (C++) page 7 of 20 OpenFPM core OpenFPM numerics C++ Template Metaprogramming Open Particle Mesh Environment (OpenPME) Distributed memory Shared memory GPUs Vector processors DSL optimizer and code generator Incardona et al., CPC, 2019
- 28. Dynamic Load balancing particles.decompose() particles.redecompose() !28 CPU 0 CPU 1 CPU 2 CPU 3
- 29. Compact scalable simulations C++ Lines: 531 C++ Lines: 85 C++ Lines: 299 C++ Lines: 492 !29 C++ Lines: 492
- 30. Performance @ZiH/TUD Time to complete 1.5 seconds Dam break simulation OpenFPM DualSPH 0 250 500 750 1,000 Dam break Time to complete 5000 timesteps for a Gray-Scott system on a 256^3 grid in OpenFPM and Amrex OpenFPM Amrex 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 100 398 Number of processor DualSPH° 24 cores dam break 150,000 particles !30 OpenFPM DualSPH Wallclocktime(s) Wallclocktime(s) 192 384 768 1536 0.0 2.5 5.0 7.5 10.0 Number of processor seconds Wallclocktime(s) Number of cores Number of cores Number of cores *S. Plimpton, Fast Parallel Algorithms for Short-Range Molecular Dynamics, J Comp Phys, 117, 1-19 (1995) °Crespo et al, DualSPHysics: open-source parallel CFD solver on Smoothed Particle Hydrodynamics (SPH). Computer Physics Communications, 2015 ^C. A. Rendleman, at al, Parallelization of structured, hierarchical adaptive mesh refinement algorithms. Computing and Visualization in Science, 3(3):147–157, 2000. 1 4 8 16 24 48 96 192 384 768 1536 0 25 50 75 100 Number of processor seconds Efficiency(%) OpenFPM LAMMPS LAMMPS* MD Lennard-Jones 216,000 particles Number of cores Incardona et al., CPC, 2019 AMREX^ Reaction diffusion 128^3 grid
- 31. Multi-GPU with minimal changes (SPH) GTX 1080: 200x speedup (for float) over E5-2670, 1 core Programming kernel-based (C++) or “numpy-like” ==> POSTER
- 32. Scientific Computing for Systems BiologyIvo F. Sbalzarini !32 Rapid Development/Coding for HPC Jeronimo Castrillon Karol et al., ACM TOMS, 2018.
- 33. I AM A VIDEO OF THE CAVE ==> POSTER In-situ Visualization and Steering in AR/VR
- 34. Scientific Computing for Systems BiologyIvo F. Sbalzarini !34 Real-time distributed image segmentation (a) (b) (c) (d) Image (a) (b) (c) (d) (e) (f) Segmentation(a) (b) (c) (d) (e) (f) TWANG (Stegmaier et al., PLoS ONE, 2014) (a) (c) (d) On one node: Visualization: ClearVolume Royer et al., Nat. Meth., 2015. Afshar & Sbalzarini, PLoS ONE 2016.
- 35. Fun Stuff!
- 36. Present Group Collaborators at CSBD External Collaborators External Funding: • Johannes Bamme • Bevan Cheeseman • Benjamin Dalton • Susann Gierth • Aryaman Gupta • Krzysztof Gonciarz • Ulrik Günther • Michael Hecht • Karl Hoffmann • Pietro Incardona • Vojtech Kaiser • Surya Maddu • Anastasia Solomatina • Justina Stark • Tina Subic • Chen-Ho Wang • Frank Jülicher, MPI-PKS • Jeronimo Castrillon,TU Dresden, INF • AxelVoigt,TU Dresden, MATH • Stephan Grill,TU Dresden, BIOTEC • Jens Walther, DTU Copenhagen, Denmark • Christian Müller, Simons Center, NYC, USA • Jan Huisken, Morgridge Institute, USA • Jonathon Howard,Yale University, USA • Urs Greber, University of Zürich, Switzerland • Yuanhao Gong, Shenzhen University, China • Carl Modes • Dora Tang • Jan Brugues • Elisabeth Knust • Gene Myers • Pavel Tomancak • Christoph Zechner • Marino Zerial • Scientific Computing Facility • Light Microscopy Facility Acknowledgements Former Group Members • Yaser Afshar • Josefine Asmus • Omar Awilé • Georgios Bourantas • Janick Cardinale • Ömer Demirel • Nicolas Fiétier • Yuanhao Gong • Nélido Gonzalez-Segredo • Jo Helmuth • Alexander Mietke • Christian Müller • Grégory Paul • Rajesh Ramaswamy • Sylvain Reboux • Aurélien Rizk • Sophie Schneider • Birte Schrader • Arun Shivanandan • Xun Xiao • AlejandroVignoni
- 37. Scientific Computing for Systems BiologyIvo F. Sbalzarini !37 Open-Source Community Software Image-Analysis: MOSAICsuite for Fiji/ImageJ • Tracking, Segmentation, Interaction Analysis, Enhancement • mosaic.mpi-cbg.de Simulation: OpenFPM • Scalable parallel hybrid particle-mesh simulations • openfpm.mpi-cbg.de Virtual Reality: Scenery • Virtual Reality for biology with user interaction • https://github.com/clearvolume/scenery Real-time Microscopy: ClearVolume • Real-time volumetric visualization during acquisition • https://github.com/clearvolume Development: PPME • Rapid code development, projectional editing • bitbucket.org/ppme/
- 38. Scientific Computing for Systems BiologyIvo F. Sbalzarini !38 Further Reading Conference Keynote ISC High-Performance Frankfurt, June 17, 2019 mosaic.mpi-cbg.de For Publications, Code, and additional materials: @MOSAICgroup1 Follow our announcements on Twitter: