Sim TK software for physics-based simulation of biological

Sim. TK: software for physics-based simulation of biological structures Michael Sherman, Chief Software Architect Simbios NCBC at Stanford NCBC Software and Data Integration Meeting 8 July 2005, North Bethesda, MD

Topics • • Scope of Simbios Center & Sim. TK SW Architecture of Sim. TK Computational considerations Opportunities for software & data integration • Licensing approach 2

Simbios mission Enable biomedical scientists to use, develop and share accurate models & simulations of biological structures—from molecules to organisms. Strategy: – Develop, disseminate, support an open-source “biosimulation toolkit” Sim. TK – Support several “driving biological problems” to ensure accuracy & utility 3

Predict structures & co-structures MDM 2/p 53 Complex Formation • Proteins • RNA • Larger complexes • Molecular machines Imagiro® Molecular Dynamics Courtesy Locus Pharmaceuticals 4

Scaling up: Molecular motors Conceptual visualization R. A. Milligan Myosin/actin interaction 5 Assembled from EM data Walker, et al. , Nature 405 (2000)

Simulation – a size problem Calmodulin “morph” ~1200 atoms David Parker (Simbios) 6 Myosin coarse simulation ~50, 000 atoms (yellow blobs are calmodulin)

Scaling down • Large mechanical models are oversimplified • Need to incorporate more details, fine structure • Biomaterials are very complex Knee implant 7 Knee!

Neuromuscular F. C. Anderson, M. Pandy 8 J. Teran, E. Sifackis, C. Lau, R. Fedkiw

Simulation-Based Medical Planning (Cardiovascular) Patient-specific models constructed from diagnostic imaging data 9 Computer simulations of blood flow to evaluate alternate treatments Charles Taylor (Simbios)

Some goals for Sim. TK • Integrate field of physics-based modeling in biomedicine • Avoid duplication of effort • Permit multiscale modeling • Accelerate research in – Biocomputation – Biomodeling – Laboratory biology – Medicine 10

Necessary math for biosimulation Initial Value Problems Boundary Value Problems • ODE • elliptic PDEs • time-varying PDE – Diffusion, fluid flow, nonlinear elastic response • Algebraic constraints (DAE) • Discrete events • Monte Carlo sampling Algebraic • Solve linear/nonlinear systems • Minimizations/optimizations • Local/global search • Kinematics 11 – (quasi)statics, electrostatics • ODE – Path fitting/planning Characteristics • Continuous/discrete • Differential/algebraic/difference • Stiff/nonstiff/stochastic • Linear/nonlinear • Combinations/hierarchies • Huge systems!

Not much of a simplification! • Biostructure simulation is general simulation (HPDAE) • Similar situation with other “subsets” – Mechanical systems – Electronics • Sim. TK has to help – build correct models – deliver them in “biological” applications – permit user to execute them efficiently • Must serve several distinct user communities 12

Sim. TK User Communities 1. Algorithm inventor 2. Modeler 3. Scientist/clinician But: generality is not a benefit to a specialist! So … 4. Application developer 13

How to focus? • • 14 Low hanging fruit? Pressing needs? How can we attract initial users? What can we offer soon?

Sim. TK Tactics • Separate “modeling” from “computation” • Support many small, purpose-built, narrow applications • Build centralized infrastructure at Sim. TK. org 15

Sim. TK Top-level Architecture Applications Problem solving Modeling Physics, mathematics, logic Computation Resource management Sim. TK. org • “Buy in” at several levels 16

Sim. TK. org Source. Forge–like “federated” model (uses GForge) Applications Modeling Computation Sim. TK. org 17

SDI for Sim. TK. org • Familiar Source. Forge federated model via GForge – Friendly seach, browse, download & install for end users – Self-governing projects within a defined framework • Best-in-class project hosting – – Sub. Version source code control CMake/Dart 2 for multiplatform build/test Installation & download support Mailing lists, project management, etc. • Inviting to collaborators at different levels of “openness” – private open binaries open source open development – Control over who can access what when • Peer review/certification for SW & people • Curation is a major task and great service 18

SDI for Applications Layer • Goal: best computational methods delivered to end-users through narrow, domain-specific, researcherfriendly GUIs • Issues – Curation & quality control – Delivery • Tools – Easy to combine elements into a narrowlyfocused app – Exploit available hardware for speed – Example apps for developers 19 Applications Modeling Computation Sim. TK. org

SDI for Modeling Layer • Goal: robust, shareable models • Issues – Many • Model building from imaging is possible SDI area – But not the current bottleneck • A later discussion … 20 Applications Modeling Computation Sim. TK. org

Computation Layer Goals • • Support both production & research Reliable, best-of-class numerical methods Interchangeable components Encapsulation and hiding of computational expertise • Exploit parallel hardware • Put high performance in the hands of bench scientists & modelers 21

Hardware focus • Avoid impossible problems! – For now tackle technical, not people problems – Sharing is hard (big cluster, grid) • Sole ownership by researcher/clinician – – Off-the-shelf notebooks, PCs, small clusters Windows, Mac, Linux Effective use of dual- and dual-dual core machines Goal: end-user sees 10 X speedup for $20 K • Typical node – 4 tightly coupled 64 -bit CPUs (e. g. 2 X dual core Opterons) – Acceleration via GPUs (e. g. 2 X NVIDIA Quadro) – $12 K now, much cheaper soon 22

Scientists are used to instruments Mass Spec Biostructure simulation 23

SDI for Computation: HPC • High Performance Computing – Physics-based, multiscale, computationally intense – E. g. , fusion simulation, astrophysics, climate modeling – Similar computational structure across domains, incl. biology Applications Modeling Computation Sim. TK. org • Focus: how to speed up a single computation using parallel hardware • Extensively pursued by DOE (DARPA & NSF too) • Much available public domain software & expertise – DOE “ACTS” collection – Common Component Architecture (CCA) 24

Licensing • Anything we write or fund: open source, BSD-like license – Anyone can do anything; just don’t blame us! • Commercial involvement is crucial to long-term health and broad dissemination! • Contributors have legitimate reasons for privacy – Welcome at all levels; easy to open up once in Sim. TK • Open source ≠ open development! – Much great software already exists in physics-based simulation – Reliable access to it is a huge problem 25

Acknowledgments • Simbios – Executive Team: Russ Altman, Scott Delp, Jeanette Schmidt, David Paik – Sim. TK staff: Clay Anderson, Ayman Habib, Jack Middleton, Bryan Keller, Chris Bruns, Pete Loan – Collaborators: Michael Levitt, Vijay Pande, Ron Fedkiw, Charles Taylor, Oussama Khatib – Students & postdocs: Silvia Blemker, Joey Teran, David Parker, James Warren • DOE – Radu Serban, Ben Allan, Rob Armstrong, David Bernholdt • Funded through NIH Roadmap for Medical Research, grant U 54 GM 072970 – Program officer: Peter Lyster – Lead science officer: Peter Highnam 26

Thank you! Contact information – – – 27 Center: simbios. stanford. edu Software: simtk. org Journal: journal. simtk. org Magazine: Biomedical. Computation. Review. org Sherm: msherman@stanford. edu