PerspectivesAre you interested in submitting a Perspective Article? Be sure to read The Science Advisory Board's Editorial Guides for Perspective Articles. Click here. Does Technology Deliver The Goods? by J. R. Archer Discovery processes in both drug and agrochemical research have increasingly focused in the last decade on automated screening in conjunction with related technologies such as combinatorial chemistry, automated compound management and genomic based target development. Use of these techniques, at increasingly higher levels of throughput, potentially offers a more deterministic route to higher annual numbers of valuable molecules and products than the traditional intellectually based methods or computer based rational drug design. The technology used is generally novel and unfamiliar to the user community, which makes it both an interesting challenge and a major risk factor. The author's company has supplied robotic equipment into this area since 1992. From observation of many customer applications, it has become apparent both that high performance automated discovery technology is not, of itself, enough, and that the variation in achieved performance between nominally similar systems at different companies can be pronounced and strongly dependent on non-technical factors. The disparity between theoretical and achieved capacity rates in terms of data points and "hits" is substantial in many cases and frequently is down by at least two orders of magnitude. It is also clear that the failure to achieve the promised rates is becoming visible both internally and externally with the blame being laid easily, but unfairly, on the technology rather than on the true cause which lies in the surrounding hard and soft infrastructures. It is the contention of this paper that new HTS technologies, while both exciting and helpful, do not alone address the true problem and that the key issue to be addressed is how to put adequate current technologies into an operational environment where their theoretical and actual performances can be more closely aligned. FACTORS DRIVING AUTOMATED DISCOVERY The capital value that the stock markets place upon the major pharmaceutical companies in particular is primarily a function of the new product pipeline that can be demonstrated at all stages of the product development process. The extraordinarily high value placed on these stocks presumes a continuing and growing flow of "blockbuster" products with large and protectable markets. A well known analysis shows that every major pharmaceutical company needs to increase its average productivity from less than half a new chemical entity ("NCE") per year to around 2.2 NCEs per year merely to maintain its ranking and sustain the expected 10% annual earnings growth. HTS received the attention and investment that it has so far because it was arguably the only discovery path that could be readily scaled to give the necessary five-fold NCE yield improvement with least dependency on chance. This has been, and should be, the main reason for pursuing an HTS-based discovery process, provided that the technology can be made to deliver the promised results. To date, the second factor driving implementation has been the challenge and pioneering aspects of using HTS technology. Robotic technology has a high "fun" element which attracts the enthusiast who is intrigued by the multi-disciplinary skills involved. Most companies' first activities in this area rightly originated with such individuals who, for a relatively small investment in laboratory-scale solutions, were able to demonstrate basic process feasibility, albeit at modest throughputs. As the technology needs have grown and matured, however, these individuals have in some cases retained a central position. In other cases they have passed the responsibility on to others in their companies better suited to the more routine, detailed issues of setting up and running a high output facility. The way in which these pioneering enthusiasts still influence their companies' current needs is an important factor to the scale of performance being realised. While rarely discussed as such, the role of a few key personalities is probably the single most critical element in both the success and failure seen at individual companies to date. The final factor driving implementation is, perhaps unfortunately, the "fashion" aspect. Given the delay of several years between implementation and measurable commercial success, no major player can afford just to maintain a watching brief lest a competitor steals a lead. While the necessary multi-million dollar capital investments are non-trivial, they are a small proportion of a typical company's R&D budget. Purchasing a portfolio of potentially relevant technologies has been a typical strategy, which is arguably valid provided there is also a clear endgame of selecting and intensively using just the best of these. All too often, however, the "fashion" driver, combined with the analyst visibility factor described earlier, has resulted in large quantities of equipment being purchased without a clear exploitation path for it. Again this has often unfairly discredited the technology for non-technical reasons. SPEED, CAPACITY AND CAPABILITY A fully automated discovery process is a highly complex mix of multi-functional activities with diverse material flows, continuing assay evolution and introduction, and large process throughputs. The interdependency of these is correspondingly complex, with process bottlenecks difficult to visualise and anticipate. While perhaps not apparent to the HTS user community, it has strong abstract parallels with comparable processes in the manufacture of items such as semiconductors, personal computers and cars. These industries have long understood how to optimise the overall performance of a system, rather than just the individual process modules. In particular, these industries have recognised that the nominal peak speed of a process module can be irrelevant to total system output. However, much HTS technology is purchased against peak rather than net realisable output. The tale of the assay automation capable of 100,000 data points per day but which needed six weeks to prepare for that day and four weeks afterwards for data analysis is, unfortunately, not apocryphal. Machines which run all day, every day, albeit at lower nominal speed, have much higher value in a total system performance context. The simple analogy here is to consider the impact on daily commuting time of switching to a Formula One racing car. Despite a three-fold increase in peak speed, the effect on travel time will be nominal due to other infrastructure issues and constraints. This is a general principle. "Capability" has a very precise meaning in the context of industrial machinery. It takes account of what the machine can really achieve in the actual context in which it is being used. It embraces process repeatability and tolerances, reliability, peak and net throughput, and the necessary allowances for set-up and maintenance time. Ideally these will be specified to the machine supplier, verified on delivery and regularly checked in use over time. The capability then determines the suitability for use of that machine in the context of the overall system and for specific types of process. For a given configuration of interconnected machines, each of known capability, it then becomes possible to estimate the overall system "capacity". This is the best the system overall can do, under real world conditions, and will bear little obvious relation to the nominal speed of any or all of the individual modules within it. Manufacturing industry has evolved sophisticated tools and techniques over thirty years to analyse, design and control these complex systems. Given the short remaining window for success, the HTS community will need to look to applying these methods very soon if it is to achieve its performance objectives. Module reliability is critical to system performance. Multi-machine systems need equipment which can routinely operate for 100+ hours unattended with mean time between failures in excess of 1,000 hours' use. The HTS paradigm is still, however, "manned automation" with operations attended by highly qualified staff at all times. Reliability also determines configuration strategy. Modern flexible manufacturing prefers multiple parallel units of modest capacity operating as decoupled cells rather than a single high performance unit for each process step. The "cellular" approach is much less affected by the failure of any one machine. In simple statistical terms, a cell-based system's overall reliability is the sum of its individual elements' reliability, while it is the product of these for single units operating serially. ORGANISATION AND CULTURE In most respects, other than name, a large integrated screening facility will have all the attributes of a modern factory. This is the complete antithesis of the user environment it serves and operates within. Currently, most discovery research organisations live in a quasi-academic, campus-based world in which it is necessary to provide an attractive working environment which encourages interchange of ideas within a non-hierarchical structure. It is also a world where life scientists predominate and engineers are often merely synonymous with facility maintenance. While the novel technology is appealing, the more prosaic matter of operating a routine and repetitive process, in which attention to detail is everything, is definitely an unattractive prospect. In simple terms, those who benefit most from a well run screening operation have little intellectual interest in being involved with it. The other organisational barrier is that most research organisations have several autonomous and distinct groupings, typically by therapeutic area in pharmaceuticals. The capital cost of a facility implies a single centralised screening resource, either one per site or one per company. Resolving the "loss of control" issues that this engenders can take many years and often has to be almost imposed from the top of the organisation to take effect. This again highlights how an otherwise sensible organisational arrangement in Research can adversely impact HTS effectiveness. Engineering as a discipline and profession is little understood in the life science research community. A critical factor in the successful systems we have observed has been the availability of professional graduate engineers to support the users. Their role has spanned definition of needs through project management to installation and subsequent support of the equipment and system. Such people exist in the manufacturing part of a user's company and can have a profound effect when "loaned" to the research function for this purpose. Headcount constraints in research organisations unfortunately mitigate against permanent recruitment of such people and related support technicians, so a successful installation may still struggle due to the subsequent lack of a technical supporting infrastructure. Note, however, that the electromechanical and software technology involved in complex robotic systems is not necessarily connected with the predominantly chemical engineering skills found in pharmaceutical or agrochemical production. As a general observation, the pharmaceutical industry is neither experienced nor highly competent in implementing high performance, robotic automation, principally because, up to now, it has never needed to be. SCALE AND COMPLEXITY Any self respecting discovery company will be planning for operational numbers in the range of 100,000 assays per day, 1 million-compound libraries, 1,000 combinatorial compounds per day, 4x109 cells per day, 1 million gene fragments, and so on. It is easy to assume that a simple lab protocol operating at the 10 per day level scales linearly to these numbers. Scaling is, however, a chaotic process in both the literal and scientific sense of the term. As an example, there is general anecdotal evidence that once compound libraries reach a threshold around 100,000 compounds, then the manual processing and paper based monitoring methods suddenly become completely dysfunctional. Typical symptoms include compound vials still waiting to be stored while also being required for a further order in the interim. Piecemeal implementation can also be problematic. This can be compounded when a module of new technology has to be installed alongside the current equipment in an existing facility without disrupting normal departmental service. This places additional expectations and pressures on the installation which are not consistent with a calm and careful programme of commissioning and training. RATE OF TECHNOLOGY CHANGE A frequent objection to any highly integrated screening facility is the perception that the technology is changing so rapidly it is not possible to commit to a fixed method long enough to automate it fully and effectively. In the limit this becomes a reason for total paralysis, but the reality is that technology is not changing as rapidly as any review of conference papers might suggest. An example is that, contrary to popular perception, most assays still use only 96-well microtitre plates rather than higher density formats. Interestingly, the right machinery can achieve 300,000 assays per day even with this "old" plate format. It is also interesting to find that the true leaders in the field often believe they are well behind others. Perception and reality are currently orthogonal axes in HTS. Much of the newest technology involves process miniaturisation in some form. This has benefits, particularly in reagent and compound savings, but will potentially suffer from exactly the same logistic and operational issues as constrain the performance of the current technology. Miniaturisation does not take away any of the problems of integrating and controlling multiple interdependent processes. The final misconception here is that only assay processes and methods which are very stable, well developed and long-lived can be fully automated. While this philosophy was true in the era of the "Model T" Ford, modern flexible manufacturing technology is all about coping with the uncertainty inherent in products and processes which can change weekly. Product stability for Intel or Dell lasts perhaps 3 months and their process automation has to be rapidly adaptable at short notice to new circumstances. It is perfectly normal elsewhere to design and commit to construction of highly automated facilities long before the detailed processes to be used within them have been fully defined. SQUARING THE CIRCLE Summarising the above it is the proposition of this paper that effective discovery automation is not limited by technology at all, but by factors which include:
In the limit The Drug Discovery Factory will provide the following benefits:
Richard Archer Chief Executive The Automation Partnership (Cambridge) Ltd Registered Office: York Way, Royston, Herts, SG8 5WY, UK http://www.automationpartnership.com ©1999, Society of Chemical Industry. Excerpted with kind permission. ### << Previous Next >> [ View All Perspectives ] |
|
