Polymer Informatics (Jerry Winter and Ian Stott)

Informatics has proven to be useful in producing a methodology for effectively searching databases of molecules to select ingredients for optimized product formulations. The inverse-design problem has been solved in Unilever’s industrial context very effectively for small molecules ingredients. A €15 million investment from Unilever has created the Unilever Cambridge Centre for Molecular Science Informatics (http://www-ucc.ch.cam.ac.uk; see also the site report), and in collaboration with this centre, Unilever researchers are tackling the same approach for polymers, because polymers present a formidable opportunity for product innovation in the fields in which Unilever competes. Winter and Stott discussed their vision for the role of molecular modeling and informatics and how they complement and integrate with each other along the product design cycle, which relies on defining selection rules and selecting the ingredients.

As with the other Unilever R&D efforts, they collaborate with external academic research partners—in particular, the Centre for Materials Discovery (http://www.materialsdiscovery.com) at the University of Liverpool, where high-throughput methods are applied to materials synthesis and physicochemical measurements. The Centre for Materials Discovery is another example of a regionally funded research effort. Other issues discussed were the use of integrating platforms—such as the Scitegic Pipeline Pilot from Accelrys (http://accelrys.com/products/scitegic)—that have enabled the development of the informatics capabilities and their deployment across the company. The importance of a true estimate of the “uncertainty” in the materials data (measured or calculated) stored in the databases was also discussed at length.

CONCLUSIONS

The WTEC visit to Unilever left the WTEC team with the strong impression that the Unilever R&D modeling group is evolving in line with the directions of the parent company and in response to the availability of new tools (particularly informatics- and infornomics-based tools) relevant to the core businesses. The research efforts reflected maximum leverage of regional, national, and European funding sources, and collaborations with UK and European universities. The leadership of Dominic Tildesley, coming out of an academic environment with excellent connections to the UK academic institutions and national funding councils, may be a factor in Unilever R&D’s success in making external collaborations work for the company.

In contrasting the UK university collaborations with similar collaborations that might take place with U.S. universities, issues of negotiating intellectual property arrangements and contract terms with North American universities were cited as hurdles in starting and maintaining collaborations with leading U.S. groups.

REFeRENCES

Allen, M.P., and D.J. Tildesley. 1994. Computer simulation of liquids, 2^nd ed. Oxford: Clarendon Press.

Warren, P.B., and J.L. Jones. 2007. Duality, thermodynamics, and the linear programming problem in constraint-based models of metabolism. Physical Review Letters 99, Art. no. 108101.

Site: University College London

Prof Peter V. Coveney, Director, Centre for Computational Science

Prof Michael Gillan, Director, Materials Simulation Laboratory, UCL

Dr Paul Kellam, Reader, Division of Infection and Immunity, Medical School, UCL

Dr Maziar Nekovee, Royal Society Industrial Research Fellow (UCL) & BT Research

Prof G. David Price, Vice Provost for Research, UCL

Prof Sally L Price, Professor of Theoretical & Computational Chemistry

University College London (UCL) was established in 1826, the first university to be established in England after Oxford and Cambridge. UCL is regarded as one of the four top-ranked universities in the UK (along with Oxford, Cambridge, and Imperial College). It has more than 2,000 faculty members (including over 600 professors in established or personal chairs) in 72 departments, 19,000 students (more than a third whom are in graduate programs, with half of these pursuing research degrees), and 4000 post-doctoral researchers. Almost a third of the UCL students come from outside the UK (more than 140 countries). UCL is the largest medical school in Europe, and research in the medical and biological sciences constitutes a major institutional focus. UCL has considerable ties with industry, including a research and graduate training centre at BT’s Adastral Park, as well as a £4.6 million program to equip UCL's scientists and engineers with enterprise and business skills.

Institutionally, UCL has made a major capital investment (£3.6 million) in a new 2560-core cluster with Infiniband interconnect rated at 27TF peak. Information about the cluster, called Legion, is available at the UCL Research Computing website, http://www.ucl.ac.uk/research-computing/information/services/cluster. With the transition to full costing of research in the UK, it is important to note that UCL is applying full costing upstream of the user, so that use of Legion is free to users. (In other UK universities, as a result of the transition to full costing of research, researchers are typically being charged for usage by the core-hour or node-hour consumed.) UCL is committing £1.5 million/year for research computing, including technical refreshing of the Legion cluster.

UCL also has a Centre for Computational Sciences, http://ccs.chem.ucl.ac.uk/, headed by Peter Coveney, which provides a bridge to external computing facilities, such as the UK CSAR and HPCx. CSAR (Computer Services for Academic Research) is a national high-performance computing (HPC) service for the UK run on behalf of the Research Councils by the Computation for Science (CfS) consortium and located at Manchester University. HPCx is the capability computing center for the UK (http://www.hpcx.ac.uk), located at the University of Edinburgh. 25–30% of the cycles at HPCx are associated with consortia-led or -containing UCL faculty.

Much of the high-performance computing at UCL is related to research in the biological and medical sciences and research at the life sciences/medical interface, such as bioinformatics. An interesting aspect of this is that value-added tax (essentially a national sales tax of 17.5%) is exempt on biomedical-research-related purchases. This contrasts with the United States, where sales tax at the state level is generally not levied on any purchases by nonprofit educational institutions.

R&D ACTIVITIES

The WTEC visiting team heard presentations from each of the hosts, as summarized below.

Peter Coveney began with the computing environment, describing both local resources (Legion and future anticipated systems at UCL, including 48-core SMP nodes) and connections with external resources. Researchers at UCL are connected to the UK National Grid Service (NGS, http://www.grid-support.ac.uk), the Distributed European Infrastructure for Supercomputing Applications (DEISA, http://www.deisa.org), and the TeraGrid in the U.S. (http://www.teragrid.org). The grid middleware enabling grib-based applications is GridSAM (http://gridsam.sourceforge.net), Globus (http://www.globus.org) and UniCORE (Uniform Interface to Computing Resources, http://www.unicore.eu). In view of the role of medicine at UCL, the data-driven “omics”—transcriptomics, proteomics, metabalomics, physiomics, phenomics and pathomics—play a prominent role. The European data center for core biomolecular data is located in the Wellcome Trust Genome Campus south of Cambridge, and is called the European Bioinformatics Institute within the European Molecular Biology Laboratory (EMBL-EBI, http://www.ebi.ac.uk). UCL researchers are marrying “omics” data to dynamical models for signaling networks, protein dynamics, and molecular dynamics to understand diseased states. The goal is to develop grid-enabled patient-specific medical simulation. As part of their efforts, UCL researchers are leading the European-wide effort EU FP7 Virtual Physiological Human (http://ec.europa.eu/information_society/activities/health/research/fp7vph/index_en.htm) that began in 2007 and runs to 2013, with first-year funding of €72 million. UCL researchers, with Peter Coveney as PI, also participate in the GENIUS (Grid Enabled Neurosurgical Imaging Using Simulation) project that aims to use grid computing technology to provide real-time simulation tools for neurosurgeons to dry-run a surgery before performing it on a real patient (http://gow.epsrc.ac.uk/ViewGrant.aspx?GrantRef=EP/F00561X/1).

Coveney also gave the UCL perspective on questions raised by the WTEC study. He reported on multiscale simulation methodologies being developed at UCL (Delgado-Buscalioni et al. 2005) to combine particle-based methods (such as molecular dynamics for liquids or direct simulation Monte Carlo for rarified gases) and continuum methods (computational fluid mechanics) to derive a multiscale description of flow phenomena. On the question of validation and verification, he pointed out that this is an especially acute problem for grid-based computing – does an application produce the same results in grid-enabled form as it produces on a serial machine or tightly coupled parallel machine. Results for the grid-enabled version must be tested against serial and/or purely parallel implementations. Another aspect of validation is ensuring that model-based simulations of biomedical and clinical scenarios are “reliable” for use in medical decision-making. With regard to software, 50% of the software used by UCL researchers is developed in-house. Examples include HYPO4D (high Reynolds number fluid flow), LB3D (mesoscale 3-D complex fluid simulations based on lattice Boltzmann methods), and HemeLB (blood flow in the intracranial vasculature). The remainder of the codes used are external, and are a combination of open-source, community/academic and commercial codes. For example, for molecular dynamics, LAMMPS is used for materials simulations with 10M+ atoms), and NAMD, AMBER, CHARMM for biomolecular simulations. The applications housing environment at UCL is that developed under the GNU license by the UK-funded Open Middleware Infrastructure Institute (OMII-UK) project (http://www.omii.ac.uk).

In reference to big data and visualization, the UK’s education and research network is JANET (http://www.ja.net). Within the overall network (having capacity of 40Gbit/s) there are so-called Lightpaths, 1Gbit/s dedicated links to various resources, such as the TeraGrid and HPCx. The preference and trend at UCL is for in situ visualization (such as in HemeLB, http://wiki.realitygrid.org/wiki/HemeLB_Deployment), combined with computational steering, rather than post-processing visualization. There are plans for UCL-wide visualization facility, implemented on the 10 Gbit/s UCL-wide infrastructure, to facilitate real-time steering and visualization from Legion. In terms of next generation algorithms and HPC, UCL researchers are, and will continue to be, at the forefront of exploitation of grid computing to solve problems; Coveney is an international leader in this area, having won awards (including a 2005 Supercomputing Conference HPC Challenge Analytics Award and the 2006 International Supercomputing Conference Award in the Life Sciences) for achievements of the grid-enabled application SPICE (Simulated Pore Interactive Computing Environment) in advancing the understanding DNA translocation across membrane-bound protein channel pores in biological cells [http://www.realitygrid.org/Spice].

Funding for simulation-based sciences at UCL comes from the Engineering and Physical Sciences Research Council [EPSRC, http://www.epsrc.ac.uk], the Biotechnology and Biomedical Sciences Research Council [BBSRC, http://www.bbsrc.ac.uk], the Technology Strategy Board [TSB, http://www.innovateuk.org], the European Union (EU) and the US NSF. As noted above, one particularly large EU project under UCL leadership is EU FP7: Virtual Physiological Human, with a budget of around €300M over 7 years. At UCL, the impression was that the current funding environment for simulation-based research, particularly in the biological and biomedical sciences, was positive.

In training and education, UCL researchers find that few starting PhD students are adequately prepared for research and careers in computational sciences, and have little or no knowledge of good software engineering practice, etc. UCL participates in a Boston University NSF-funded IGERT grant supporting interdisciplinary training of graduate students in computational science as one of the overseas centre in IGERT participants can choose to work for a 3-month period during their PhD degrees. Problems cited by Coveney were the difficulty of developing and maintaining software beyond the lifetime of project funding (one attempt to do this is the materials database at http://db.foxd.org) and the ongoing challenges endemic to grid computing of security, confidentiality, privacy, authentication and authorization balanced against speed and efficiency.

In his presentation, Richard Catlow focused on the UK HPC Materials Chemistry Consortium (HPC-MCC). The HPC-MCC consists of 23 faculty members across the UK and their group members, for a total of almost 100 participants. HPC-MCC members employ the latest developments in HPC technologies, in a wide-ranging program of development, optimization, and applications studies aimed at modeling and predicting the structures, properties and reactivities of functional materials, including catalysts, ceramics, minerals and molecular materials. The techniques include large-scale forcefield-based simulations (molecular dynamics and Monte Carlo methods) as well as electronic structure techniques employing density functional theory, Hartree-Fock and hybrid techniques. Computational chemists and chemical engineers find joining the HPC-MCC is a quicker and more efficient route to gaining access to large-scale HPC resources than trying to access those resources directly. This is similar to the process available to materials/chemical/biological simulators in the US through the Department of Energy (DOE) nanoscience centers. E.g., the users of the Nanomaterials Theory Institute (NTI) in the Center for Nanophase Materials Sciences (CNMS) at Oak Ridge National Laboratory (ORNL) have access to the large allocations the NTI-CNMS has on the DOE’s capacity and capability HPC platforms. NTI researchers work with users to convert their problems into codes that will run efficiently on the HPC platforms, allowing users to get access quickly to large levels of state-of-the-art computational capabilities. Likewise, new members of the HPC-MCC can draw on the combined experience and expertise of HPC-MCC members in order to use HPC facilities efficiently. Several examples of HPC-MCC members research accomplishments were given, including a study of chlorine adsorption on Ag(111) surfaces (de Leeuw and Nelson 2004) and a study of the active sites in titanium silicate (To et al. 2007). Information about the HPC-MCC, its meetings and its membership is available at its website, http://www.dfrl.ucl.ac.uk/mcc.

Mike Gillam highlighted the Daresbury Laboratory-based Collaborative Computational Projects (CCPs, http://www.ccp.ac.uk) beginning in the 1970’s, the national HPC consortia (including the UK Car-Parrinello consortium, the Materials Chemistry consortium covered in detail by Richard Catlow, and the Mineral Physics consortium), and the national-level HPC resources from the 1990 onwards, culminating in HECToR (http://www.hector.ac.uk, scheduled to reach 250 Tflop by October, 2009). Focusing specifically on UCL, he noted that UCL has more than 30 research groups engaged in materials modeling (in the departments of Physics and Astronomy, Earth Sciences, Chemistry, Mechanical Engineering, etc, and the London Centre for Nanotechnology, http://www.london-nano.com), coordinated under the Materials Simulation Laboratory led by Gillam. UCL researchers account for ~30% of national computational resource use, and have strong involvement in international grid projects (particularly through the activities of Peter Coveney). Several internationally used codes (such as GUESS, Gaussians used for embedded system studies (Sushko et al. 2000) and CONQUEST, http://www.conquest.ucl.ac.uk, a linear scaling density-functional-theory-based electronic structure code) were developed at UCL. Some of the technical issues he outlined that impact current and next-generation HCP algorithms and hardware are development time—during the time taken to develop, implement, validate, optimize the algorithms, computer power may have increased by a factor 1000—and the trend of increasing numbers of processors and number of cores rather than more powerful processors—this is useful for increasing the size of problems that can be addressed, but does not impact the time-scale problem, which in many ways is the most fundamental and will require new theory. As educational and training challenges, he cited formal education in computational science as being the only way to redress the trends of inefficient HPC codes and the growing “black-box” mentality to simulation codes. At present, he offered the opinion that these issues are not well recognized in the UK funding agencies.

Sally Price described her research on the computational prediction of organic crystal structures and polymorphism and the closely related control and prediction of the organic solid state (CPOSS) project (http://www.cposs.org.uk) that she leads. The goal of her work, reviewed recently (Price 2008), is to develop a computational method to predict the crystal structure of a molecule without experimental input, and to predict all (practically important) polymorphs and properties, as aid to experimental discovery and understanding crystallization. One of the important practical applications of polymorph prediction is in the pharmaceutical industry, where variations in crystal polymorph can alter dramatically the bio-availability of a drug. Her methodology is to search systematically for plausible crystal structures (considering 3000-100000 possible crystal structures), using quantum mechanics to predict molecular structure and represent the charge distribution within the molecule, high-quality forcefield models to minimize the lattice energy of each crystal structure, and analyze the structures with the lowest energies (calculate properties and store in database to compare with experiment). The prediction of progesterone crystal structures was given as an example of the methodology and its ability to understand puzzling experimental results. The results are placed in an on-line searchable database. Although the work is scientifically interesting and industrially valuable, challenges include the use of 12 substantial codes (all published, licensed and documented), of which only 2 are in-house, and problems of continuity and stability in funding, computational infrastructure and personnel. Right now the project depends significantly on the goodwill of several contributors, some of whom are retired and contribute out of personal interest.

The final speaker, Maziar Nekovee, is a research fellow at British Telecom (BT) Research and in the Centre for Computational Sciences at UCL, working with Peter Coveney and others. Nekovee is interested in ad hoc wi-fi networks, created on the fly by, e.g., laptops interacting with hot spots and each other, and vehicles interacting with each other and traffic control systems. Given that a single wi-fi device is already a complex stochastic state-machine, which is difficult to model, putting these devices together to form networks is creating extremely complex engineering systems that are very large scale (over a million BT hubs, many millions of wi-fi-enabled smart phones, are highly distributed (unlike cellular systems), and may show unpredictable behavior. Hence, simulation-based planning and evaluation is becoming essential, to address issues such as security (e.g. in response to cyber attacks), predictability (reliable performance), mobility (pedestrians, buses, cars, trains) and scalability (interference/radio spectrum). One recent application, the simulation of worm propagation in ad hoc wi-fi networks (Noekovee 2007), was described. Nekovee enumerated some of the computational challenges of these kinds of simulations, including efficient updating of large-scale mobile wireless communication graphs, coupled simulations of vehicular traffic and high-fidelity wireless communications, and parallel simulation and synchronization of discrete event wireless simulations. Many of these challenges have analogues in molecular/materials simulation. E.g., borrowing from the ideas of neighbor lists in molecular dynamics simulation has led to advances in updating large-scale mobile wireless communication graphs; and the coupling of vehicular traffic and wi-fi communications is a multi-timescale problem, with the typical timestep for traffic simulations being 1 second, while the typical timescale of wi-fi wireless simulations is about 1 microsecond. Another challenge is for the combination of algorithmic efficiency and HPC capability to reach the point where simulations run faster than real time, allowing the possibility of control; right now, even for modest-sized systems (640 node wi-fi network), simulations run between one and two orders of magnitude slower than real time.

Researchers at UCL are engaged in a wide variety of simulation-based scientific and engineering research projects. UCL’s high level of simulation-based research activity, its institutional commitments to provide, now and into the future, state-of-the-art capacity computational capabilities to its users with zero cost at the point of use, and its leadership in applications of grid computing, make it an exemplar of simulation-based engineering and science (SBES).

Clearly, UCL’s commitment to support SBES research is paying dividends in recruitment of faculty in this area. One example is Richard Catlow, who left the prestigious Royal Institution (RI) in 2007 after 18 years to move permanently to UCL (http://education.guardian.co.uk/higher/news/story/0,,2063727,00.html), taking a very large group (30–40 scientists) with him.

The WTEC team was left with the impression that UCL is aggressively positioning itself as a leader in SBES within the UK and internationally, and that the connection to the medical applications of SBES is a significant driving force in the strategic planning at UCL.