Simulation-based engineering and science

Prof. Frank Bruggeman

Prof. Barbara Bakker

Prof. J.C. de Munck

Prof. Ivo van Stokkun

BACKGROUND

Systems Biology is defined as the science that studies how properties that are important for biological function emerge from the interactions between components of biological systems. Systems Biology bridges molecular biology and physiology. It needs mathematical approaches to deal with the complexity of the nonlinear interactions. It also needs quantitative experimentation in actual living systems, or well-designed in vitro representations thereof. Professor Westerhoff is a driver behind the Silicon Cell program (http://www.siliconcell.net), which makes computer replicas of pathways in living cells available on the Web for in silico experimentation. He heads a transnational research group on Systems Biology that includes the Manchester Centre for Integrative Systems Biology (MCISB) in the Manchester Interdisciplinary BioCentre (MIB) and the BioCentrum Amsterdam. He is also the director of the UK’s EPSRC (Engineering and Physical Sciences Research Council) Doctoral Training Centre in Integrative Systems Biology.

The Systems Biology effort at Vrije University (VU) Amsterdam is affiliated with Integrative BioInformatics VU, the Netherlands Institute for Systems Biology, TopMaster Systems Biology, The Trans North Sea Centre (Manchester Centre for Integrative Systems Biology; Manchester Doctoral Training Centre Systems Biology), and the BioSim EU Network of Excellence. The WTEC visiting team heard about efforts in signal transduction, metabolic engineering, hierarchical control and regulation, and digital human from Professor Westerhoff. A theme of the efforts at this center is that to cure a disease, one must cure the network. This requires the understanding of multifactorial relationships. A strong case was made of why simulation is an integral part of this process.

Professor Bruggeman described the Silicon Cell project. This is an initiative to construct computer replicas of parts of living cells from detailed experimental information. The Silicon Cells can then be used in analyzing functional consequences that are not directly evident from component data but that arise through contextual dependent interactions.

Professor Barbara Bakker described efforts in network-based drug design, showing the interplay between metabolism, gene-expression, and differentiation. The premise is that the recent decline in drug discovery may be due to the focus on single molecules, neglecting cellular, organellar, and whole-body responses. Specifically, she outlined a research effort focusing on the treatment of African sleeping sickness using trypanosome glycolysis as a drug target. This entailed development of an ODE model of trypanosome glycolysis, and sensitivity analysis to identify those reactions that can best control the flux. The researchers found that the network behaviour was complicated by the fact that gene-expression response can either counteract or potentiate the primary inhibition; hence, it was necessary to model the regulation of gene expression. Some of the model parameters were obtained from the literature, and others were fitted from experiments.

Professor J.C. de Munck described efforts in brain imaging. EEG and fMRI are used simultaneously to localize pathological waves such as epileptic spikes. The project requires modelling, solution of both forward and inverse problems, signal/image processing, and visualization, all in close coordination with experiment.

Dr. Ivo van Stokkum described a problem-solving environment (PSE), to identify and quantitatively model complex biomolecular systems. The PSE takes data from multiple time-resolved spectroscopical experiments and a priori knowledge, and outputs the model structure and estimated physicochemical parameters. A partitioned variable projection method was developed for the (large-scale) parameter estimation problem. The PSE is used by many collaborating scientists to study problems including natural and artificial photosynthetic systems, photoreceptors, plant metabolism, and imaging of the corneal surface. The work is done in close coordination with experiment.

The systems biology and biophysics efforts in Amsterdam are world-class. There is an impressive integration of simulation, visualization, and experiment. They demonstrate both examples and a tremendous potential for how modeling and simulation can impact drug development and medical diagnosis and treatment. The WTEC team’s hosts expressed some frustration in finding appropriately trained students.

The panel’s hosts graciously provided answers to our written questions, which are attached below.

Silicon/ digital human (90 % complete in 15 years)

Construction, validation, analysis of large models of biological systems, most urgently cells. This should go hand in hand with technological advances for cellular measurements. The feasibility of large predictive models is still questionable given current technological approaches. This is anticipated to change significantly within the next 5-10 years making predictive biosimulation reality; allowing for rational engineering of biological systems for medical and industrial purposes.

What are the national and/or regional funding opportunities that would support research to meet these needs, and/or take advantage of these opportunities? Are these funding opportunities expanding?

National: left-overs from genomics plus small scale

Germany, UK, EU: large investments

Transnational

Funding of institutions (excellence centres) to carry out integrative biology in a setting of technology and simulation development. Such institutions should offer facilities to groups outside of the institutes. Institutes should be involved in strategic collaboration with other institutes worldwide.

Materials / Energy and sustainability / Life sciences and medicine

What major breakthroughs in these fields will require SBE&S; and which are you and or your colleagues pursuing? Within your institution, region, or country are there identified targets of opportunity for applications of simulation either for scientific research or for engineering applications in these fields?

Digital human

Self-sustaining systems (Systems Biology to green planet)

Quantitative analysis of signal and metabolic networks; theory and experiment

We are engaged in: 1. SMAD, MAPK signalling 2. Central carbon and energy metabolism in yeast, E coli and extremophiles, 3. Modelling/experimentation of eukaryotic transcription initiation (PPAR induced transcription of human PDK4 gene), 4. nucleocytoplasmic shuttling of signalling proteins, 5. Nuclear receptor signalling

Formulation of concepts/approaches to facilitate comparative systems biology (discovery of principles) related to control, regulation and adaptation of intracellular processes.

Single-cell stochastic simulation of transcription and signalling; we are currently setting involved in initiating experiments

Which problems could benefit most from a 1-2 order of magnitude increase in computational power?

Distributive digital human

Web services based linkage of groups each responsible for pathway/organ

Analysis of populations of models for experimental design, system identification, stochastic simulation of large networks

SUCCESSES: Application of models in biotechnology and bioengineering, improved understanding/ exploitation of control of metabolism; hybrid modelling approaches; applications of engineering concepts/approaches to biology; role of simulations in bioinformatics/molecular dynamics.

Network-based drug design

FAILURES: small number of predictive models (depend much on quantitative experimentation of enzyme kinetics (enzymology) which is out of fashion and has no high-throughput method at the moment); ab initio prediction of kinetic constants from molecular dynamics simulations; solutions to deal with multi-scale simulation of biological systems (how to cross levels?; formal methods to allow for coarse graining);

No

Have any start-up companies spun-off based on simulation efforts in your lab? If so please describe them.

No; potential was there; but by nature of the projects too much interconnection required

Multiscale Simulation

Describe efforts and advances within your institution to couple multiple simulation methods in order to bridge multiple length and/or time scales.

Modular approaches for control and response analysis

No?

Silicon cell/JWS: models are recalculated before put onto live web site and collaborating journals

30% Mathematica code

50%; Mathematica

50% Copasi, JDesigner

0%

What are the biggest issues in using models/simulations developed by others?

Link to experimental data; judgement of biological quality

How easy/difficult is it to link codes to create a larger or multifaceted simulation environment?

Hard for large models; SBML is promising effort but not yet sufficient multiscale supportive; no good multiscale software available for novice

Not relevant for us now

Computer replica of reality

Site: Vrije University Theoretical Chemistry Section

De Boelelaan 1105

1081 HV Amsterdam, The Netherlands

http://www.chem.vu.nl/en/sec/tc/
Date Visited: February 25, 2008
WTEC Attendees: S. Glotzer (report author), L. Petzold, C. Cooper, J. Warren
Hosts: Prof. Evert Jan Baerends, Head, Theoretical Chemistry Section, Department of Chemistry and Pharmaceutical Sciences
Tel: +31-20-598 7621; Email: baerends@chem.vu.nl

Dr. Stan van Gisbergen, CEO, Scientific Computation and Modeling NV (SCM)

Prof. Luuk Visscher,

Prof. Evert Jan Baerends leads the Theoretical Chemistry group at Vrije University, which is comprised of 30–35 people, including 5 permanent members (Dr. Luuk Visscher, Dr. Matthias Bickelhaupt, Dr. Oleg Gritsenko, Dr. Drew McCormack, and Prof. Baerends) plus postdocs and graduate students. An additional five senior researchers (including Dr. Stan van Gisbergen, CEO of SCM) are employed by Scientific Computation and Modeling NV (SCM), a spin-off company that supports the Amsterdam Density Functional (ADF) simulation software developed by the Baerends group.

Research within the Baerands group deals primarily with development and application of density functional theory. The computational advantages offered by DFT are being exploited in computational method development. Further research emphases within the group include theoretical aspects of DFT (Kohn-Sham potentials, meaning and status of Kohn-Sham orbitals, orbital-dependent exchange-correlation functionals); MO analysis of bonding and spectroscopy, mostly in transition-metal (organometallic) complexes; molecule-surface interactions (scattering, dissociation, heterogeneous catalysis); spectroscopic (response) properties; and relativistic effects in heavy element compounds. This group is one of the leading theoretical chemistry groups in Europe; its ADF simulation software is highly popular and used around the world.

The Baerends Theoretical Chemistry group is part of a new center, the Amsterdam Center for Multiscale Modeling (ACMM), which includes this group, Computational Chemistry and Physics at the University of Amsterdam, and Theoretical Biophysics at AMOLF Amsterdam.¹⁴ The mission of the ACMM is to model systems “from electrons to electricians,” spanning all of the scales needed to bridge the physics of electrons to the biology of living systems. Over the next 20 years, ACMM researchers aim to focus on building a bridge from quantum mechanics to thermodynamics.

Other groups within the Netherlands involved in molecular modeling and simulation include groups at Eindhoven, Groningen, Leiden, Nijmegen, Twente, and Utrecht. These groups all participate in activities of the European Centre of Atomic and Molecular Computations (CECAM).

SBE&S Research

Simulation-based science research in theoretical chemistry at Vrije University includes the following:

• 
  Fundamental research and development of density functional (DFT) and dynamic mean field theory (DMFT)
  
• 
  Code development
  

ADF has several competitors, including NWChem, QChem, Turboval, and Jaguar, with Gaussian being the largest. ADF was the first of these codes to include numerical integration. ADF strengths include its unique ability to accurately treat transition metals, heavy metals, and spectroscopic properties; its ease-of-use as an analysis tool; as well as the longevity and stability of the code. ADF is very popular around the world, particularly in Japan. SCM has sold several hundred licenses, which represents many more users. In industry, the primary users include oil, automotive, and electronics companies. The code is built on OPEN MPI, which enables it to run efficiently within a multicore architecture (currently under Windows), and it can operate on a single processor or many processors in a parallel environment. The development of ADF has benefited from a long-term collaboration with the Ziegler group at Calgary University of over 30+ years, e.g., they developed the popular NMR module for ADF.

• 
  Time dependent density functional theory (TDDFT), response theory
  
• 
  Embedding (frozen density), QM+MM multiscale methods. Wesolowski, Neugebauer. One can now easily calculate important properties on systems containing up to 150 atoms. More complex systems, such as molecules in water and undergoing reactions, require multiscale methods that bridge quantum mechanics and classical mechanics. The group has a large grant from the NL NSF to develop these methods.
  
• 
  Structure, reactivity, and catalysis (homogeneous). UvA (Reek) Leiden (Buda), top research school in the Netherlands, and Eindhoven (Van Santen, Jansen).
  
• 
  Relativistic effects and their inclusion in density functional methods (Gronigen, Broer, Filatov)
  
• 
  Under the leadership of Dr. Visscher, the group develops two other codes complementary to ADF—MOLFDIR and Dirac—that deal with relativistic effects for heavy atoms. MOLFDIR is not open source per se, but the group will provide it (primarily to academics) upon request. Dirac is a collaborative, bottom-up effort involving five groups throughout Europe that currently operate with only a small amount of funding from the EU.
  
• 
  Molecule-surface interactions (Leiden: Kroes, Olsen McCormack)
  
• 
  Theoretical biochemistry, molecular recognition, DNA. Questions addressed, in particular within ACMM, include, what are the reactions that lead to the formation of DNA?
  

Some universities with the Netherlands, including Vrije University, acquire and support the equivalent of Tier 3 computational resources locally to faculty, researchers, and students. These resources are paid for by the university with central funds. According to Baerends, the Universities of Groningen and Amsterdam are today at Tier 3, while Vrije University is at the low end of Tier 3. National computing facilities provide access to Tier 2 resources. Faculty can apply for grants (proposals are peer reviewed) from NCF and HPCC programs to support postdocs to develop and port simulation codes to the newest machines at NCF. These codes must then be freely shared with other NCF users. This sharing requirement occasionally discourages some Dutch groups from using NCF resources, since a particular code may represent a significant competitive edge for that research group. There is also a constraint that no single group can use more than 10% of the computational resources at NCF. Grand-challenge-type problems requiring substantially more resources can sometimes be investigated at the very early installation stages of new supercomputers at NCF. Note: NSF is no longer supporting the NCF, and now this resource is a mixture of commercial and academic use.

Regarding the impact of SBE&S and looking to the future, many important problems involving small molecules can be solved today via SBE&S. As one example, gas molecules (N₂, O₂, CO₂, H₂, CO,… ) constantly hit surrounding surfaces, with large effects. Using codes like ADF, we can answer today’s questions such as: Why does iron rust? Why do copper cupolas turn green? Why is gold a noble metal? Why is its neighbor, platinum, a good hydrogenation catalyst? Why do we need a three-way catalyst in our car? However, major challenges remain with larger molecules and complex systems, such as large molecules embedded in a solvent environment. There, much faster computers, such as petascale computers, are required, along with the ability to efficiently run codes on those new platforms. Petaflop computing is also required to treat dynamics, which is important, for example, in problems in biology, energy, and those involving nuclear waste (such as the separation of actinides from lanthinides by judicious design of selective ligands). Such problems will require the coupling of molecular dynamics and time-dependent density functional theory, which in turn will require significant method and algorithm development.

ADF has been implemented on quad-core chips. However, implementation of ADF by SCM on chips on multicore chips with substantially more than four cores will be a major undertaking and will require a cost-benefit analysis taking into account the likely number of massively multicore users.

Regarding human resource development, education, and training in SBE&S, deep, solid knowledge of a core discipline, as well as computational skills, will be required to solve “grand challenge” problems with computational chemistry. (Half of the PhD students in the Baerends group currently are physicists, and the other half are chemists.) Baerends noted that many users of computational chemistry codes are not very good at mathematics and have limited knowledge in theoretical chemistry. To be innovative, computational chemists need to be trained in both theoretical chemistry and computational methods. Students in his group generally take graduate courses in computational chemistry and programming. A computer center at Vrije University, supported by the NSF, provides training to students in parallel programming. In addition, the national computer center offers computing courses supported by NSF.

Regardless of country, students pursuing a PhD must contribute original ideas to their field of study. However, in the Netherlands, Baerends noted that in theoretical chemistry, code development and application are not considered “new.” As a result, students generally focus on theoretical developments and not code development. He contrasted this situation with that in Germany or the United States, where code development is viewed as original.

At Vrije University, students pursuing a bachelor’s degree in chemistry take courses in computational chemistry, where they are exposed to codes like Gaussian and Spartan. The Pharmaceutical Chemistry program offers modeling courses for master’s students. All undergraduates take a course in computer programming. It requires 5 (3+2) years to obtain the master’s degree, and then an additional 4 years for the PhD.

Regarding Dutch government support of SBE&S, Professors Baerends and Visscher noted that in the Netherlands, it is very difficult to obtain research funds to support code development and long-term maintenance and support, since the peer-review process for grant applications is based on science and not code. Even more generally, code maintenance languishes because there is no reward system for non-science activities. They noted that due to lack of coordination and support of code development, there is an enormous amount of duplication of effort within academia where individual groups develop their own codes or parts of codes, in order to avoid having to rely on other’s codes that may not be supported after a few years. They further noted that the UK stands out has having paid attention to code maintenance through long-term investment.

The researchers noted that in the Netherlands, most grants cover the full period of a student’s PhD, which enables continuity of code development and science projects. Young investigator grants support 5–6 students for the duration of the PhD thesis, and “top grants” to established investigators support three students for the duration of the PhD thesis, plus equipment. At Vrije University, similar to other universities within the Netherlands, students cost approximately €40,000/year.

The theoretical chemistry community within TheNetherlands is close-knit and meets at annual meetings. When asked to compare computational chemistry code development in Europe with the rest of the world, the WTEC visiting team’s hosts stated that Europe leads in the development of relativistic QM codes, whereas the United States leads in the development of QM codes based on couple-cluster methods. They felt that algorithm development is centered in Europe. They further noted that contributions to computational chemistry at the U.S. Pacific Northwest National Laboratory (PNNL), namely the package NWChem, was innovative not in the theoretical developments in the code, but rather in their implementation. Amber and Charmm are still strong U.S. programs. Europe has strong Romax.

Over 50% of Vrije University graduate students are from outside the Netherlands, with many from Eastern Europe. Our hosts noted that, in theoretical chemistry, the strongest and most disciplined students come from Germany, with whom they have easy communication. Trade with Germany and China has increased in The Netherlands, while at the same time, the country is clamping down on immigration, making it harder to recruit students from abroad.

Our hosts noted many success stories for simulation in terms of impact on industry. ADF is used to calculate, for example, solubilities and phase diagrams. A large unnamed Dutch company uses ADF for this. Most companies use ADF for homogeneous and heterogeneous catalysis. Azko-Nobel used ADF to scan the hyperpolarizabilities of hundreds of molecules to narrow down potential targets for synthesis, and in so doing discovered molecules with enhanced nonlinear optical properties. Another Dutch company used ADF to improve the properties of an existing catalyst.

In the pharmaceutical industry, there is increased interest today in virtual screening, including dynamics and finite temperature (for entropic effects) to predict binding affinities. Many HIV transcription inhibitors have important contributions from simulation, and one compound went in production based on only 31 iterations, with significant savings.

Many pharmaceutical companies within The Netherlands have modeling groups, but these groups do not necessarily contain expert simulators. In contrast, BASF has a wide range of simulation capabilities, including a modeling group very strong in QM and MD. Astra-Zeneca is now building up its modeling and simulation capabilities, with an eye toward computational methods development.

Site: Zuse Institute Berlin (ZIB)