SmithKline Beecham has become one of the world's largest pharmaceutical
companies as a result of its innovative research leading to major new
drugs for the treatment of bacterial infections, peptic ulcers,
arthritis, heart attacks, depression and the severe nausea induced by
drugs used to treat cancer. For more than twelve years, computational
chemistry has been used as an aid to the drug design process. Visual
computing is a vital part of molecular modelling, and throughout
SmithKline Beecham's research sites, Silicon Graphics workstations and
servers are used in this work.
"You may know something about the protein you're going to dock a molecule into," says Dr. Frank Blaney, Manager of Molecular Modelling and Quantum Chemistry at SmithKline Beecham, "but in less than one percent of cases is the full 3-dimensional structure known".
Blaney, and the other members of the UK computational chemistry team, based at SmithKline Beecham's four research sites to the north and south of London, use modelling techniques to develop hypotheses about those properties of drugs which are responsible for their biological activity.
The members of the team have a variety of backgrounds and expertise in such areas as quantum mechanics, computer graphics, protein modelling, small molecule modelling and Quantitative Structure-Activity Relationships (QSAR). QSAR involves a study of the relationship between the biological activity of a series of compounds and their properties, often using multivariate statistical techniques.
For all these computational methods, Silicon Graphics workstations and servers play a key role. Before the advent of Silicon Graphics, the modelling group originally used a number of DEC VAXes, linked to Sigmex or Evans and Sutherland graphics terminals.
An Easy Programming Environment
"One of the primary reasons SmithKline Beecham chose Silicon Graphics was because of the wide range of computational chemistry software already written for the systems," explains Dr Blaney. "But the continuing need for in-house program development also meant that we wanted a system which gave us a good graphics programming environment. Most members of the group were already proficient with programming in Fortran and some in C. These compilers, linked to the Silicon Graphics' Graphics Library, made the programming environment the easiest one to use for the generation of "real-time" graphics applications."
The IRIS Graphics Library(tm) is a programming interface used for
creating computer applications that allows users to visualise and
manipulate colour images. It is a real time interactive library of 440
software routines that are independent of any specific windowing system
or hardware platform. IRIS GL(tm) is especially useful for creating
models of molecular objects quickly and efficiently.
At SmithKline Beecham, one of the first development programs on the Silicon Graphics workstation was to write a comprehensive quantum mechanical suite, as part of a joint project, for MSI's (formerly Polygen) software package QUANTA. Polygen was one of the earliest computational chemistry software companies to use Silicon Graphics as its main platform. Many other programs have been developed within the group and for these tasks, the accessibility and functionality of Iris GL has proved to be extremely useful.
All of SmithKline Beecham's research sites in the UK share Silicon Graphics systems. These currently include five 4D/380 Silicon Graphics's POWER Series(tm) GTX computers, each of which has a throughput of 234 MIPS and features parallel RISC architecture that is fully compatible with other Silicon Graphics machines. This is where most of the CPU intensive work is focused. There are also two 4D/310VGX PowerVision(tm) graphics supercomputers and several smaller 4D/70, 4D/25 and 4D/35 systems. Last year alone, SmithKline Beecham bought 8 Silicon Graphics systems with a combined total of 43 processors.
The PowerVision systems are well suited to the mix of computational and graphics tasks required for molecular modelling and computational chemistry work. The systems can manipulate up to one million polygons per second.
Viewing in Stereo
Because of the complexity of the models generated by the computational chemistry team at SmithKline Beecham, they often use the Silicon Graphics' StereoView(tm) system to produce 3-dimensional stereoscopic computer models. Stereo viewing makes it easier to interpret and manipulate complex models from wire frame to fully rendered solid models.
"Stereo is absolutely essential where protein modelling is concerned," says Blaney. "As we dock potential drugs into the proteins, we can see the interactions clearly. On a flat screen it is very difficult to see these interactions in such complex 3-dimensional objects."
For the study of proteins in stereo 3-D, SmithKline
Beecham uses the Massachussets-based Molecular Simulations' software
packages, QUANTA(tm) and CHARMm(tm). Many other
packages however, are also used on the Silicon Graphics workstations.
For example, protein sequence analyses and homology modelling are
carried out using the Oxford Molecular packages, SERRATUS(tm), CAMELEON(tm) and
IDITIS(tm). Most small molecule modelling is done with the SYBYL(tm) software from Tripos Associates while
calculations on the electronic properties of molecules are carried out
with a mixture of public domain and commercial software including
SPARTAN(tm) from Wavefunction Inc. (based in
Irvine, Ca.) and GAUSSIAN 90(tm) from the group
at Carnegie-Mellon University in Pittsburgh.
Quantum mechanical calculations are often the methods of choice because, unlike molecular mechanics, they are not dependent on the availability of parameters. Such calculations however take a great deal of power. Using a single processor on the 4D/380, Gaussian 90 calculations often take 7-14 days of CPU time each. The increasing availability of parallel versions of these codes, which make full use of the multiprocessor environment of the 4D/380's, is expected to significantly shorten these times.
Other routine types of calculations such as molecular dynamics are also highly compute intensive and often take several days of CPU time. This demand for exceptional computational capability rather than just graphics performance led to a detailed evaluation of computing systems prior to the purchase decision, as Frank Blaney recalls.
"When we were looking at the POWER Series in October 1990," he says, "the computational chemistry group at SmithKline Beecham's US research centre compared them with IBM's RS/6000 workstation and the smaller Cray machines. On a price/performance ratio, Silicon Graphics came out on top."
For the future, the computational chemistry group at SmithKline Beecham hopes to purchase further Silicon Graphics systems, because of an increasing requirement from the medicinal chemists to carry out simpler modelling studies themselves. The introduction of the desktop IRIS Indigo(tm) RISC PC may well be the platform to achieve this at a reasonable cost.
"In our work, the calculations and graphics make full use of the Silicon Graphics systems," says Blaney. "With the IRIS Graphics Library, Silicon Graphics systems are easy to program, and the move over to X Windows will make available a greater range of software and take us closer to a common standard."
"Overall, we are extremely pleased with Silicon Graphics. The machines have proved a cost-effective solution for the full range of challenges we have faced."
