Table of Contents

Molecular modeling software is now available for a variety of computer platforms, opening the field to new users.

Molecular modeling can be a computationally intensive activity. Ab initio quantum mechanical calculations, many-thousand-step molecular dynamics calculations, and the rendering of high-quality molecular graphics all require substantial computing power to make virtual experiments viable.

As computing power has increased over the past 30 years, computational chemistry has taken full advantage of this extra power. More sophisticated algorithms, shorter time steps, and more accurate molecular modeling have contributed to making computational chemistry a more accurate and reliable tool.

UNIX has long been the platform of choice for serious molecular modeling work. Windows NT and Linux are emerging as serious competitors to the UNIX platform. As computing power increases even more and operating systems such as Windows NT and Linux have become available, software providers are rising to the challenge of making computational chemistry tools available for the personal computer (PC) and therefore to a wider spectrum of scientists.

The arrival of UNIX

Kenneth Thomson and Dennis Ritchie of Bell Laboratories wrote UNIX in the early 1970s. It was the first multiuser scalable operating system, and it rapidly became the standard platform in computer labs worldwide. (Scalable operating systems are those in which computational power increases with the number of computer processors.) AT&T made the original UNIX source code available, but UNIX has since been adapted into many proprietary versions, making the issue of UNIX standardization a persistently hot topic. The flexibility of UNIX means that it has been run on machines ranging from microcomputers to supercomputers; and it has been adopted by almost every large corporation, government lab, and academic group in the world.

In the 1970s, many computational chemistry algorithms were written as Fortran or C command line programs running under UNIX. The command line interfaces required operators to have skills in both computational science and chemistry.

As computing power increased and molecular graphics could be produced more rapidly, graphical user interfaces (GUIs) became popular. Not only did they allow users to interact intuitively with the molecular models, but they also made it less intimidating for nonprogrammers to use computational chemistry tools.

As graphics capabilities increased, the term “molecular modeling” gradually became synonymous with “computational chemistry”. Silicon Graphics Inc. (SGI) became a leading contender in the computational chemistry hardware market because of the high graphics capabilities of their machines. SGI’s graphics workstations used IRIX, the company’s proprietary version of UNIX. Other machines, such as IBM computers running AIX (IBM’s proprietary version of UNIX) and VAX machines running the VMS operating system, remained popular options. The built-in open operability of the UNIX platform made it simple for computational chemists to set algorithms running for hours or days on remote machines with sufficient power to carry out complex calculations.

Power to the PC

IBM launched the first PC in 1981, and it did not take long for Microsoft to pick up the trail. The prospect of desktop computing on compact, powerful machines drove the speedy development and acceptance of the PC.

By the end of the 1980s, technology had advanced so far that it was possible to build low-cost desktop computers as powerful as the mainframes of the 1970s. These advancements were still paltry compared with the computing power that was available on the powerful workstations of the day, most of which ran on UNIX.

DOS was the first operating system for PCs, and it grew in popularity throughout the 1980s and early 1990s. It was suitable for running many general applications, including early versions of Windows. Many computer software houses converted from mainframes to PCs and DOS for low-end applications. The low cost of PCs meant that by the 1990s, corporations, government labs, and academic groups could put PCs on practically every desk. However, the computing power required for molecular modeling—for the graphics, the quantum calculations, or even the molecular dynamics calculations—was simply not available on early models.

The mid-1990s saw the release of some nonexpert molecular modeling tools on the PC. The concept of low-end computational chemistry programs had not been viable previously because companies could not afford to buy UNIX workstations for occasional use by nonexperts. Programs such as ChemDraw and ISIS/Draw provided easy-to-use point-and-click modeling packages so that chemists could sketch molecules and perform simple energy minimizations on the PC. More powerful and sophisticated modeling and calculations, however, remained the preserve of UNIX workstations.

Client–server systems were the key to challenging the dominance of the UNIX platform. Network technology increased in sophistication throughout the 1990s and made it easier to connect increasingly powerful PCs to work in parallel.

More robust operating systems were introduced on the PC. With the release of Windows NT, Windows became a true operating system rather than an application running on MS-DOS. This robust environment made it a better option for the development of sophisticated applications running across networks.

Enter Linux

As Windows NT grew in popularity, it met an unexpected competitor in the form of Linux. Linux is a variant of the UNIX operating system and was originally developed by Linus Torvalds in 1991. Linux is free and is developed by a loosely knit team of programmers working all over the world. The system is distributed under an open source license agreement that stipulates that people can only download the Linux source if they agree to make their changes open source.

Torvalds wrote the first version of Linux in response to his frustrations with DOS. Linux works on almost every kind of computer in existence, but it is most widely used on PCs. It is likely to remain more robust than its closed-source competitors such as Windows NT because of the numerous developers working on it around the world. In this respect, it follows the pattern already demonstrated by Apache (Apache Software Foundation, Forest Hill, MD), the open source Web server software that is the most popular server software available today.

The arrival of Linux has prompted many companies, including SGI, to give up development of their proprietary UNIX versions and concentrate on Linux instead. Others, such as Solaris, have opted to make their versions of UNIX open source.

Figure 1 Step time for the same molecular dynamics calculation on four different platforms.

Before PCs could truly challenge UNIX workstations, they had to clear one final hurdle—the high quality of graphics required for molecular modeling. However, the real drive for improved PC graphics has come from the computer games market. The graphics performance required for modern computer games is at least as high as that required for serious molecular modeling. PC gamers demand good value for their money and expect quality to improve continuously. This has resulted in rapid development of relatively low-cost, high- performance graphics for a market that requires top-quality software in the $200 to $300 price range.

Today, most PCs can produce graphics of high enough quality to satisfy molecular modelers. PCs with good graphics cards can rival UNIX workstations. In fact, a modern PC laptop can produce sufficiently high-quality graphics to render large molecular structures, such as a 100 repeat-unit polymer chain, and manipulate them smoothly.

Speed trials

The following example demonstrates the performance differences between today’s popular computational chemistry platforms and PC alternatives.

A recent in-house performance study was run on four different platforms: Linux on a PC, Windows NT on a PC, IRIX on an SGI O2, and IRIX on an SGI Origin. O2 and Origin computers are commonly used for computational chemistry; Linux and Windows NT computers are potential alternatives. Molecular dynamics calculations simulate a series of experimental time steps. For the same molecular dynamics calculation, the amount of computational time taken to simulate a time step was measured for these four platforms. Figure 1 shows the results.

Although the Origin computer had the fastest step time, Linux and NT were not far behind, and both were clearly faster than the O2 computer. The results do not apply to all UNIX boxes for all applications and do not directly compare the same factors; however, they do indicate the emergence of a fairly level playing field. Once the generally lower price of a PC is considered, the comparison becomes striking. Table 1

Table 1. Performance characteristics and prices for four computer operating systems
System	CPU type	Processor clock speed, MHz	Typical computer price, U.S.$	Price x step time, s
Linux	P2	266	2,300	144
Windows
NT	P2	266	2,300	164
O2	R10K	180	17,000	1810
Origin	R10K	195	27,000	1200

gives some approximate prices and performance characteristics for the four systems. We have reached the point where the price– performance ratio is strongly in favor of PC platform Windows NT and Linux for many applications.

With the advent of Linux, parallel computing and clustering are popular options for achieving a large amount of computing power. The fastest computer available today is a bank of parallel PCs running Linux.

Making the transition

The price–performance comparison in Table 1 clearly points to PC-based operating systems as the platform of the future for computational chemistry. However, price–performance is not the only factor to consider.

Historically, almost all serious computational chemistry software has been written to run UNIX. Only nonexpert tools such as PC Spartan, ChemDraw, and ISIS/Draw have been available to PC users. For the present, however, the price– performance comparison between PCs and UNIX and the growing popularity of Windows NT and Linux mean that software vendors are responding to a demand for expert molecular modeling tools for the PC. As the software becomes available for PCs, the price–performance factor is bound to tip the balance in favor of PCs. For the present, however, most computational chemistry software will only be available for UNIX.

UNIX still offers advantages for computationally intensive calculations that cannot be rivaled by Windows. UNIX offers better scalability; for many large calculations, parallel computing on super workstations running UNIX is the only practical way of getting results in a reasonable time. UNIX also offers advantages in terms of its relative stability. Although Windows NT is stable enough for day-to-day tasks, it is not yet crash-proof enough to run large calculations that may take more than a month of computing time to complete. UNIX remains the best operating system for calculations of this scale.

PCs are cheaper than UNIX workstations, but many organizations have already made large investments in UNIX-based hardware for running advanced computational chemistry programs. Switching to PC platforms may mean avoiding such expensive investments in the future, but the desire to use existing hardware may be difficult to resist.

A client–server architecture is an ideal means of adding low-cost PCs to a network while making full use of existing UNIX workstations. PC clients have sufficient power to produce high-quality molecular graphics and to perform simple dynamics calculations. Users then have the option to run computationally intensive calculations on remote UNIX, NT, or Linux servers. This allows organizations not only to take advantage of their existing investments in UNIX or PC work stations, but to add new PC clients relatively cheaply.

How users feel about making a move from UNIX work stations to PCs is an important factor affecting the rate of change to a new platform. Nonspecialists will certainly prefer to use PCs for computational chemistry projects.

Many expert users of computational chemistry are keen to start using PCs. Instead of owning a portion of a Cray, SGI Origin, or IBM SP3 in a remote machine room, they now can own a cluster of PCs in their own offices. This gives them more control over their resources and makes it easier to add extra computing power by adding low-cost PCs to the cluster.

There may be some resistance to moving to the PC among those who already use UNIX workstations. Not only does the move to a lower-end machine imply that the tools will not be as powerful as before, but there is the perceived threat of “dumbing down” the software to a “black box” set of calculations that give users less control over calculations.

In fact, these fears do not present real problems. If a truly expert molecular modeling environment is made available on the PC, it will be easier to use than its UNIX equivalent, making more functionality accessible to nonexperts. On the other hand, expert knowledge will still be needed to carry out effective and meaningful virtual experiments, just as expert chemical knowledge is still needed to produce good results in a well-designed chemistry lab. There will still be a need for expert computational chemists; in fact, their skills will be more in demand than ever as computational chemistry is deployed in corporation-wide communication systems and used as part of every research project.

Using Linux-based PCs may also make the transition from UNIX work stations to PCs more acceptable for traditional programming chemists. In a recent MatHub poll (www.mathub.com, user registration required, accessed July 25, 2000), Linux was cited as the firm favorite to be the future platform for computational chemistry. As a UNIX-related operating system, Linux looks likely to send out a strong challenge to UNIX and Windows NT as the computational chemistry platform of the future.

Supporting the spectrum of research

A move to PC-based molecular modeling will make computational chemistry more accessible in terms of the entry point for hardware and the psychological barriers of usability and familiarity. The challenge is to support the spectrum of users created by opening molecular modeling to a mainstream scientific market. This means having a molecular modeling package that supports expert and nonexpert users. A comprehensive molecular modeling environment that supports the spectrum of chemical research will have a major impact on the way in which chemical organizations operate.

Computational chemistry will become available to a wider audience. Moving to the PC does not mean a dumbing down of the software, and a well-designed package will provide inherent advantages in terms of usability. It will give experimentalists access to tools for sketching molecular models without a steep learning curve. Users already know their way around standard PC applications, and this knowledge can be transferred directly to computational chemistry tools designed to the same standards.

As experimentalists use computational chemistry more, they will routinely carry out “standard” tasks such as performing energy minimizations and simulating powder diffraction patterns. Computational chemists will then have more time to focus on their specialized work.

Hardware investment will no longer be a barrier to the use of computational chemistry. With a PC-based client and the option to run on UNIX, Linux, or Windows servers, organizations will be able to take advantage of their existing hardware resources without making a large investment in new hardware to add new clients. More users will have access to computational chemistry tools. Eventually, molecular modeling tools will be installed as standard applications on chemists’ PCs, much as an e-mail client, an Internet browser, and a word- processing package are now.

Experimentalists will speak the same language as computational chemists. With one molecular modeling package deployed corporation-wide, users will be able to sketch molecules, e-mail them to colleagues, and share the resulting files. Chemical communication will be supported and enabled throughout the organization, helping research projects proceed smoothly and efficiently. Providing experimentalists with simple molecular modeling tools will get them more involved in modeling and give them a better understanding of its advantages. They will realize the potential of computational chemistry and look for more opportunities to involve the computational chemistry team in research projects.

Researchers will have robust tools to support their chemical creativity. A usable, purpose-built molecular modeling package will make it easy for scientists to express and communicate their chemical ideas. New ideas can be tested immediately by running virtual experiments on the PC. Ideas that make it through this first stage can either be passed on to the computational chemistry group for further computational work or tested by experiment in the lab. The immediate feedback and flexibility offered by a desktop chemical environment allows scientists to express their thinking and try ideas out quickly, thus encouraging more original thinking. Virtual experiments also provide a safe environment in which to carry out dangerous or expensive experiments that could not easily be carried out in the lab. As one researcher at BP Amoco commented, “I’d rather delete a file than clean tar out from a reactor!” (1).

The first of the new breed

Figure 2 The Materials Studio interface.

In March 2000, Molecular Simulations announced the launch of Materials Studio, a new PC environment for materials science. Materials Studio is an expert molecular modeling tool designed specifically for the PC that follows accepted standards for graphical interface design, which makes it easy to use (Figure 2).

Materials Studio follows a client–server architecture. The GUI runs on PCs under Windows 95, 98, or NT. Calculations can be carried out locally on NT if desired, up to the computational limit of the client computer being used, but users also have the option of running computationally intensive calculations on remote servers. Materials Studio gives a choice of server platform—Windows NT, Linux, or UNIX—thereby covering most existing hardware resources and giving new users the opportunity to convert entirely to PC-based molecular modeling.

Materials Studio is a modular system. The core module, Materials Visualizer, is easy to use and includes everything needed to construct, optimize, and view models of molecules, polymers, and crystals. Scientists can start to use Materials Studio straightaway without going through a steep learning curve.

Three-dimensional graphical models, text, graphs, and tables can be exchanged easily between Materials Studio and standard Microsoft Office tools, which will help scientists to efficiently produce high-quality, informative reports. Molecular models and textual results can be copied and pasted to and from Microsoft Word and Excel.

Other Materials Studio modules released in version 1.0 include:

• Amorphous Cell, for predicting properties of bulk amorphous systems;
• Discover, a powerful simulation engine offering atomistic methods that can be applied to a wide range of molecules and materials;
• COMPASS, the state-of-the-art force field; and
• Reflex, for simulating X-ray, neutron, and electron powder diffraction patterns to determine the structure of crystalline materials.

The client–server architecture of Materials Studio means that hardware resources can be shared as well. This capability is relevant to academic circles, where server time is often booked on remote machines. Materials Studio is designed to make it easy to run remote server jobs; it fits into the natural workflow and allows all users to take maximum advantage of their hardware resources.

The future of computational chemistry

The concept of the paperless lab may never be a reality, but the computer-free lab is certainly a thing of the past. Electronic experimental data output, high-throughput materials and biochemicals screening, and laboratory information systems are rapidly becoming part of the researcher’s everyday life. Desktop productivity tools such as word processors and spreadsheets are used for writing reports and analyzing data. The addition of molecular modeling to this increasingly sophisticated environment is natural and almost inevitable.

The advent of PC molecular modeling tools such as Materials Studio paves the way for the future of computational chemistry. Materials Studio is the first in what is bound to be a succession of high-quality PC-based molecular modeling products. It is likely that in 10 years there will be few commercial computational chemistry tools that run only on UNIX workstations.

The future promises cheap computing power with easy-to-use modeling tools that will be deployed throughout organizations and far beyond the computational chemistry lab. Experimental chemists will interact with molecular models on their PCs and work closely with computational chemists to advance chemical and materials research using every available technological tool.

Bringing the latest scientific computing resources to every researcher will have a profound effect on the way researchers work. Chemical communication and the scope for individual creativity will be vastly improved and will open the doors to increased chemical innovation, the key to future scientific breakthroughs and economic growth. PC-based molecular modeling, whether under Windows or Linux, will be the winner in the great platform debate.

Reference

1. Knapman, K.; Warde, S. Computational Chemistry in the Chemicals Industry. Financial Times Business, January 2000, p 90.