Making Science Make Sense
| Home > Scientific Writing and Editing > Feature Article Sample | Home | About | Samples | Contact |
|
Automating the Discovery Process
Each decade for the past 40 years, automation has been pronounced the panacea that will make business functions from the factory floor to the administrative office more productive. Ironically, though, the lessons learned about implementing automation in one industry are not readily passed on to the next generation of adopters. Many of the lessons learned by manufacturing companies in the 1980s apply to discovery automation; the quotes introducing the sections of this article, for instance, were all published last decade in the Harvard Business Review. The challenge for both discovery laboratories and automation vendors is to apply these lessons to scientific methods and processes that are still rapidly evolving. Some observers predict a shakedown in the discovery automation industry, where unprecedented customer demand will need to be reconciled with the technology's relative immaturity. Whatever the outcome, research organizations will succeed or fail based on how well they have defined their automation needs as a business, rather than a merely technological, issue. Beyond the Need for Speed "The arena of competition is shifting from manufacturing to engineering, from process maintenance to process improvement." (2) The industrialization of discovery came as the result of a confluence of technologies and business pressures. Competition and heightened regulation have forced research organizations to reengineer R&D programs. Simultaneously, a dramatic increase in the number of molecular targets is fueling the need for faster alternatives to traditional discovery approaches. And these high-throughput processes are now enabled by more reliable and varied types of robotics equipment (dispensers, liquid handlers, shakers, weighers, etc.) along with sophisticated software programs for building integrated data management infrastructures. "Discovery automation has developed as the need has arisen," stated Troy S. Critchelow, communications manager at Advanced ChemTech (Louisville, KY). And the primary need was for speed. The definitive high-throughput techniques, combinatorial chemistry and high-throughput screening (HTS), play a perpetual game of one-upsmanship. More assays and testable targets mean more compounds to create, which in turn require more assays. This positive feedback cycle has made discovery automation remarkably different from efforts to automate in other industries. For the first time, the processes requiring support simply cannot be accomplished without automation. "We joke in the industry that, in the past, our primary competitor was not another automation vendor, but the lab technician who thought it was easier to do the work by hand," explained Bill Buote, senior software developer at Zymark Corporation (Hopkinton, MA). But manual methods are no longer an option for research organizations that have chosen to adopt high-throughput techniques. To implement combinatorial chemistry or HTS is to make a binding commitment to the technology and underlying infrastructure. According to Buote, this lack of an "easy out" has placed a whole new burden on automated systems. Where a manufacturing organization could let a robot fall into disuse, today's research organizations expect to get the most out of their investment by extending automation into new uses whenever possible. For vendors, this customer mindset, combined with staggering technological turnover, has created one of the most demanding customer markets in any industry. As Doug Cardwell, product manager for Core systems at Sagian Inc., a Beckman/Coulter Company (Indianapolis, IN), put it, "We need to be ready to move with a new instrument in less than six months, or it's obsolete." In addition, research organizations can no longer afford to trust the implementation and maintenance of automation to a dedicated, in-house technology "champion" -- those that do run the risk of creating isolated pockets of automation that fail to integrate effectively into the larger R&D process. Speed and numbers are still important, but organizations are now showing more concern for the process itself. "Companies are realizing that the quality of chemical compounds in a library is as important to the overall success of a research program as the quantity of compounds," reports Ed Long, marketing director at Argonaut Technologies (San Carlos, CA). "Focusing on issues of chemistry optimization leads to a far more efficient way of creating compound libraries than just churning out vast numbers of compounds by brute force." From Custom Systems to Standards "Every. . . system is governed by two kinds of software: the information system, which tells the machines what to build, and the control system, which tells the machines how to build. The question is, what information has to be known in real time?" (3) The rush to adopt high-throughput techniques has led many automation vendors to change their product strategies. The pioneers in laboratory automation, such as Zymark in the early '80s and Sagian this decade, introduced modular tools and built custom systems to handle specific customer needs. But according to Sagian's Cardwell, "The big custom systems were not giving customers the productivity gains they were expecting. There were too many workstations tied together, connected by inefficient software for scheduling assays and supported by unproven hardware that frequently broke down." The demand for easy to implement, scaleable systems has led these vendors and their counterparts in the automated synthesizer arena to create standard workstations driven, whenever possible, by open architectures and standard programming languages, such as Visual Basic for Applications. Yet even as vendors adopt standards, research organizations continue "to roll their own applications," says Zymark's Buote. One reason for this is that vendors have yet to settle on what the standards should be. The 96-well plate and its miniaturized counterparts have dictated many standard processes -- the generalized operations of adding and removing liquids, heating, cooling, and washing are all focused on this format, making these operations rather rugged. But other operations, such as how different systems communicate, what commands control certain operations, and how data is uploaded to or downloaded from readers, remain less well defined. Andy Zaayenga, president of TekCel Corporation (Martinsville, NJ) and the secretary of the Laboratory Robotics Interest Group, noted that the lack of standards puts research organizations at a disadvantage. "Standards are a must," he said. "Right now, each instrument interface, sometimes even within brands, is a custom project. The marketplace must drive the instrument manufacturers to adopt some common interface or convention so that selecting an instrument is driven by what is best for the science, and not by what interfaces can talk to each other." Lee Amon, marketing manager of biology products at MDL, agreed with Zaayenga's assessment of the problem. "MDL has developed software that helps scientists move data from readers to databases to analysis software without reformatting or reentering it, but the emergence of data-exchange standards would help everyone tremendously," he said. Most vendors have at least settled on standards within their own product lines. But these standards rarely apply to systems from other vendors. Cardwell pointed out that Sagian has worked with a number of partners to make various third-party stations compliant with its systems. In the synthesizer market, Advanced ChemTech and Argonaut Technologies are both pursuing alliances with information management providers to make their systems easier to implement. But true standards, where customers could conceivably choose "best of breed" instruments for different tasks, seem difficult to achieve. Zymark's Buote predicted that until vendors settle on standards, customers will continue to be tempted by custom applications. "Ask any consultant in manufacturing or financial services, and they will tell you to buy an existing product and adapt your process to the application," said Buote. "But in discovery, organizations still feel they are unique. And they are to some extent justified, because we as vendors have not provided systems that they see as defining their processes. Our collaboration would encourage customers, give them confidence, and prevent them from becoming 'roll your own' competitors." Viewing Discovery through the High-Throughput Lens ". . .[T]he implementers of automation give insufficient attention to the substrate for the electronic overlay -- that is, to what the combination of man and machine actually does." (4) The basic problems that confronted the early adopters of automation in other industries also exist for discovery laboratories. Analyzing the workflow to determine which systems to implement, finding the space for the instruments in the laboratory, preparing scientists and information systems to handle the flood of data generated once the robots start working -- these are perennial problems. But the nature of discovery automation also presents some unique challenges that can take both research organizations and automation vendors by surprise. In manufacturing, as in discovery, the goal of automation is to improve efficiency. But what it means to be efficient differs in these two disciplines. "The goal of the manufacturer is simply consistency, whereas the goal of the researcher is variety," said Advanced ChemTech's Critchelow. "So while automating a single reaction or assay -- like automating a single manufacturing operation -- is a simple task, it doesn't serve the ultimate goals of a discovery laboratory." The inherent complexity and variation of discovery automation requires both customers and vendors to envision clearly how the instruments will come together as part of an integrated research environment. One knee-jerk response is the tendency to overautomate. According to the LRIG's Zaayenga, many automation novices feel compelled to automate every assay or rely on workstations or overly complex modules to perform even the most simple processes. "Some complex assays are best handled by breaking them up into manageable parts that may even include manual steps," Zaayenga advised. "Each assay or synthesis should be evaluated to determine whether automation is indeed appropriate, which requires open communication between the customer and the automation vendor." In addition to reexamining the discovery process itself, research organizations also need to learn to look at the data differently. Bill Haag, owner of Haag Consulting in San Francisco, CA, specializes in helping his clients see the signal in the noise. As Haag explained, most scientists are trained to eliminate noise in search of the signal. But with systems that generate thousands of data points a day, the noise itself carries important information. "There is enormous structure to the noise on a 96-well plate, but to see it, you need to examine plates over time, rather than relying on averages and T-tests," Haag said. "Why use up your entire library chasing down false positives, when the answer may lie with the false negatives?" In the end, Haag said, the key to effective high-throughput discovery is experimental design. "We have the silicon CPU to drive both sides of the process: chemistry and screening. Now we need the people CPU to make the decisions about how to do the work. The decision-making process is the current bottleneck to the discovery process and research success." The Role of Information Systems: Making the Results of Automation Pay Off In a field as technologically complex as high-throughput discovery, good information systems are a necessity. For decisions to happen, data must flow from process to process, integrating with other types of information at key junctures. A good data flow not only speeds and optimizes physical processes, but also provides an archive that will support future data mining. The ultimate goal for many organizations is to have the molecular targets, synthetic reactions, reagent lists, validated structures, and assay results completely integrated, so that during analysis, data can be placed in the context of how it was generated. MDL software has been integrated with robotics and instruments by innovative companies throughout the industry. But more importantly, MDL software also powers Haag's "people CPU" -- the tasks that require human interaction and scientific judgement, such as reagent selection and synthesis planning for combinatorial chemistry. Clearly, automation will play a major role in the future of high-throughput discovery. Companies planning on implementing robotics in the research environment would do well to study the successes, as well as the false starts, that have taken place in other industries. Although many lessons can be learned from these examples, discovery research has a unique combination of needs that will also require new innovation. In the end, the vision is to combine the benefits seen in other areas, such as the manufacturing's gains in speed and consistency and the financial industry's new data mining opportunities, with the art of science as practiced in the laboratory. Notes 1. Foulkes, Fred K."People Make Robots Work." Harvard Business Review. Jan/Feb 1984, p. 94. 2. Jaikumar, Ramchandran. "Postindustrial Manufacturing." Harvard Business Review. Nov/Dec 1986 , p. 69. 3. "A CEO's Common Sense of CIM: An Interview with J. Tracy O'Rourke." Harvard Business Review. Jan/Feb 1989, p. 110. 4. Salerno, Lynn M. "What Happened to the Computer Revolution?" Harvard Business Review. Nov/Dec 1985, p. 129. |