Expert Systems for Liquid Chromatography - Analytical Chemistry

---

Focus

Expert Systems for Liquid Chromatography Researchers pursue different approaches in an effort to develop software that will aid methods development In the practice of liquid chromatography (LC), many factors affect the choice of mode of separation (reversed phase, ion exchange, etc.), packing material, mobile phase, and instrumental operating conditions. In addition, requirements regarding resolution, sensitivity, and run time vary for different sample types. The complex decisions involved in LC methods development are usually made by human experts, but this process is not always as efficient as it might be. Thus, in an effort to automate LC methods development, researchers at a number of companies and academic institutions are attempting to incorporate the experience of human LC experts into computerized expert systems. Expert systems, an outgrowth of research into artificial intelligence, are programs in which pertinent solutions to specific problems are arrived at by a humanlike reasoning process. Because there is no single "correct" approach to LC methods development, the systems currently under development often reflect widely varying philosophies on how an LC expert system might be configured. One view of expert systems is that they simply represent a new, valueadded way of communicating information. Tom Jupille of Pi Technologies (Orinda, Calif.) explains it this way: "From the point of view of the software writer, an expert system project is very much akin to writing a book or an article. Expert systems are a way of taking a particular person's expertise and making it available to non-expert users. "Somebody who is going to make use of a book on LC methods development," says Jupille, "has to begin at the beginning of the book and read through it, because later parts of the book will build on earlier parts. You can put the same information into an expert system format, and the infer-

ence engine and knowledge base will lead the user right away to the specific piece of information that is needed. It does this on an interactive basis. And it does it without necessarily forcing the user to comprehend something earlier in the program. This is a means of transferring information that becomes a much more practical tool." "People are struggling with different approaches to expert systems," explains Stan Deming of the University

" . . . there will be different kinds of expert system shells for different kinds of problems. I don't think we'll ever find one universal approach...W of Houston. "The direction I see is that there will be different kinds of expert system shells for different kinds of problems. I don't think we'll ever find one universal approach to all expert systems." Some of these different approaches will be discussed by Deming and others at a session on expert systems in chromatography to be held this month at the Eastern Analytical Symposium (EAS) in New York. DryLab The recently released DryLab programs (LC Resources Inc., San Jose, Calif.) are not expert systems, but some of the calculations they perform are similar to those that would be incorporated into an LC expert system. DryLab software, developed by Lloyd

1192 A · ANALYTICAL CHEMISTRY, VOL. 58, NO. 12, OCTOBER 1986

R. Snyder and John W. Dolan, is used for computerized simulation of LC runs. Instead of injecting a sample and waiting for the chromatogram, the DryLab user enters separation conditions into the program and obtains the same results he would have obtained from an experimental run. DryLab is based on a set of equations that predicts how various changes in chromatographic conditions affect resolution, plate number, column pressure, run time, solvent consumption per run, relative peak sensitivity, and other parameters. Theoretically, a chromatographer could look these equations up in a book and make the same calculations without any need to purchase software, but this is generally not done in practice because it is too inconvenient. What LC Resources has done is to organize this expertise, validate it, and package it for easy access by chemists. In addition to its functionality in helping to optimize chromatographic conditions, DryLab also has potential educational applications. As one academic chemist put it, "Students in the laboratory should have a simulator. They could then try 100 conditions and their effects on the chromatography without having to spend 100 hours on an instrument. If they could do that first and then go to the chromatograph, I think they would learn a lot more." Shell-based expert system Another LC program with potential educational applications is under development at Michigan State University (MSU). Adrian P. Wade of British Petroleum Research Centre, who is spending a sabbatical year with Stan Crouch at MSU, described this system late last month in a presentation at the Federation of Analytical Chemistry and Spectroscopy Societies meeting in St. Louis. In constructing an ex0003-2700/86/A358-1192$01.50/0 © 1986 American Chemical Society

Focus pert system for LC methods develop­ ment, Wade has been using the Teknowledge M.l expert system de­ velopment shell to encode the exper­ tise of chromatographer V. L. McGuffin as a set of rules and facts. An ex­ pert system development shell is a commercial program that provides the structure for development of an expert system. "We have a program that works," says Wade. "It gives us a number of options for sample pretreatment, mode of separation, and type of col­ umn packing, and solvent selection ca­ pability is currently being implement­ ed. The results are usually correct— that is, the selection our expert would make. But the system is still small and needs a lot more development. At present it contains about 150 rules and about 150 other knowledge en­ tries." The MSU system is primarily edu­ cational in nature, and commercializa­ tion is not anticipated. According to Wade, "If students can ask 'Why?' at any time, then obviously they will learn more from an interactive envi­ ronment rather than if they just sat and listened to a lecture or read a book on chromatography."

ttThe question for those who make expert systems may not be, 'How does an expert solve the problem?' but 'How should an expert go about solving the problem using all the tools available to him?'*> Troubleshooting Another approach to expert systems is represented by HPLC Doctor ($130) from Pi Technologies. HPLC Doctor is an expert system developed by Ben Buglio of Hoffman-La Roche that is directed not at methods development, but at instrumental troubleshooting. According to Tom Jupille, the pro­ gram operates in an interactive fash­ ion: "Do you have spikes in your base­ line? Yes. OK, shut off the pumps. Are the spikes still there? Yes or no? The same sort of thing that shows up in

troubleshooting books and service manuals. It's like having a very experi­ enced LC jockey sitting in the next lab who you can go to and say, 'Hey, there's a problem here.' " Jupille says that Pi Technologies will eventually offer a broad range of expert systems. "We're interested in doing programs on post-column reac­ tion detector optimization, GC sta­ tionary phase selection, and solidphase extraction techniques. All of these will be done in conjunction with outside authors." Another program available from Pi Technologies at this point is Generic Expert, which, at $80, is one of the more inexpensive expert system development shells. Fuzzy sets Stan Deming's approach to expert systems has involved the use of fuzzy sets. "Most expert systems deal with very rigid logic—a lot of if, then, else rules," Deming explains. "But many times it's hard to put analytical ques­ tions and answers in that kind of con­ text. Our approach to expert systems is to work with fuzzy sets that are based on desirabilities. The answer you eventually get might say that the desirability of a certain goal is 0.83 on

For Quantitation in Chromatography there is only one choice!!! The EM SCIENCE/Hitachi* Model D-200G Chromato-lntegrator pro­ vides the chromatographer with unmatched capabilities for quantita­ tive analysis. From basic powerful single-channel version, the S E C O N D C H A N N E L — for use with a second detector as well as a very large increase in storage capacity. FLOPPY DISC — for long-term storage and recall of analytical raw or reduced data. P A N — a unique communication network for EM SCIENCE/ Hitachi HPLC Systems. RS232C — allows transfer of data to an external computer.

V.*

E X T E R N A L E V E N T S — through time-controlled contact closures external devices can be activa­ ted by the D-2000.

β

*?Z,«_

s e e B9ES

— B»

To complete your modular EM SCIENCE/Hitachi HPLC Systems, request additional information on: • Pumps • Fluorescence • Columns • UV Detectors Detectors • High Purity • Semi-Micro • Autosamplers Solvents/ Systems • Column Ovens Reagents

atpa

For additional information on the D-2000 or to arrange for a demonstration, call or write: f 5 ™ " " " ™ •' 111 Woodcrest Rd. K M Cherry Hill, NJ 08034-0395

cn •Reg. TM Hitachi, Ltd., Japan Circle 48 for literature.

EM

(609) 354-9200

E M SCIENCE

(800)222-0342

Λ Division ill ΓΜ Industries, In,

Associate of E. Merck, Darmstadt, Germany

Circle 49 for a demonstration.

1194 A · ANALYTICAL CHEMISTRY, VOL. 58, NO. 12, OCTOBER 1986

Focus a scale of zero to one. The algorithms are fairly straightforward—they turn out to be essentially summations of these desirabilities. The really hard part is coming up with meaningful desirabilities." The program Deming is presently working on is directed at methods development in analytical chemistry in general, but it will also be applicable to LC. Opening up new possibilities An extraordinarily innovative approach to expert systems is that espoused by Sterling A. Tomellini of the University of New Hampshire. According to Tomellini, expert systems should not simply mimic the way experts currently go about solving problems, because the advent of expert systems will change the way in which those problems are solved. "The question for those who make expert systems," says Tomellini, "may not be, 'How does an expert solve the problem?' but 'How should an expert go about solving the problem using all the tools available to him?' " For example, Tomellini has been conducting research on coupled-column gradient elution chromatography, in which two, three, or more short

^The ultimate goal would be a closed-loop system... It's going to happen, but it will take a while." columns may be connected together in different ways to optimize separation problems (Anal. Chem. 1986,58, 904-6). "You have to have some data available on a computer to do that kind of experiment," says Tomellini. "You can't make an expert system at this time to solve that problem, because there isn't any knowledge available." "What Tomellini is saying—that as you have this capability you're going to change the way you do things—is right on target," adds Lloyd Snyder. "The DryLab programs are in the same category. Once you can run 100 experiments on the computer terminal, you discover a different perspective, and you not only learn how to solve the problem you're working on, you also learn a different approach to

any problem of that sort." Tomellini is still in the initial stages of writing his expert system, which is why his talk at the upcoming EAS symposium is entitled "The Foundations of an Expert System for HPLC." "What we're doing," says Tomellini, "is writing the mathematical routines and asking ourselves, 'How would we use these to solve the problem?' " The Varian project Other LC expert systems in the works include the system that has been under development at Varian for a number of years. The goal of the Varian expert system is to provide suggestions on appropriate columns, mobile phases, and detectors, based on the nature of the sample and detectivity requirements. The system would also help select the preferred type of chromatography (reversed phase, ion exchange, etc.) and sub-modes of chromatography (either ion suppression or ion pairing for reversed-phase chromatography, for example). At that point, the system would step to an optimization program. S. R. Abbott of Varian explains that "we have written the knowledge base to choose columns, solvents, and sol-

Automate ASTM D86, Gp 1-4; D850; and D1078

Precision ADA IV/Lab Management System automates distillation tests, manages data ADA IV automatic distillation analyzer, equipped with a powerful internal computer, can store twentyfour test programs for a wide range of samples. It performs, computes and prints hard copy results of distillation tests in full compliance with ASTM test methods D86 (groups 1 through 4), D850, D1078 and equivalent European standards. With built-in RS232 inter-

face, the ADA IV can be connected to the Lab Management System (LMS) or to a system mainframe computer in any location. With the optional Lab Management System (hardware and Precision software package), you can operate up to ten ADA IV analyzers to enhance the output of your laboratory. The more ADA'S you have, the more you can increase

P r e c i s i o n S c i e n t i f i c 3737 West Cortland St., Chicago, IL 60647 CIRCLE 168 ON READER SERVICE CARD 1196 A · ANALYTICAL CHEMISTRY, VOL. 58, NO. 12, OCTOBER 1986

yields by narrowing the specifications with timely, precise ASTM results. Call or write today for more information. U.S. Toll Free -

1-800-621-8820 Illinois 312-227-2660.

MEMBER SAMA SCIENTIFIC APP MAKERS ASSOC

Focus vent additives. Currently we're work­ ing on an optimization program devel­ oped by Leo de Galan of the Technische Hogeschool Delft, which takes these recommendations and then de­ cides whether to run isocratic or gradi­ ent elution chromatography. "The ultimate goal would be a closed-loop system," Abbott contin­ ues, "but we're not there yet. It's going to happen, but it will take a while. Also, it should be feasible for us to transport the system to a PC with ei­ ther an 80286 or 80386 processor, from the MicroVAX-II it currently resides on. Clearly, we can reach a lot more users with a PC-based program than a Micro VAX-based program." Optimization packages The expert systems currently under development will have to contend with the problem that many users have en­ countered in the use of LC solvent op­ timization programs: the fact that they don't always work. "About half the time our optimization package doesn't pick an optimum," explains Joseph Kosman of the Standard Oil Research and Development Center. "It sometimes interprets the data in­ correctly and then goes off into the

".. .if you put one up against the other, the topnotch analytical chemist is still going to do a better job." wrong part of the solvent diagram. It does work sometimes, but how well it works is dependent on the type of sample involved." This is not to say that optimization software has not been useful at Stan­ dard Oil and other laboratories, how­ ever. According to Kosman, the pri­ mary utility of the optimization pro­ gram used at Standard Oil is "not that it finds the optimum, but that it does enough separations that I can go back through them and pick out something that is close to what I would call the optimum. I've talked to a lot of people who use optimization packages in in­ dustrial settings, and they have told me the same thing: that the packages they were using didn't necessarily con­ verge on the optimum, but ran enough

CAN TEMPERATURE CONTROL ENHANCE LC PRECISION? If you are using anything other than the BAS 200 Problem Solver™, you may not know the answer. With the BAS 200 you would have an advanced temperature control system which can accommodate both electrochemical and UV detector cells, most column formats, and the injection valve. The result? Enhanced precision, lower baseline drift, improved symmetry and results beyond anything you could expect from a modular LC system. This unique approach to temperature control is only one of 20 great ideas incorporated into the hot, new Problem Solver™ from BAS.

2701 Kent Ave West Lafayette Indiana 47906 Β iο a n a I y t c a I I n c S y s t e m s ,

Telex: 276141 (317) 463-4527

CIRCLE 17 ON READER SERVICE CARD 1198 A · ANALYTICAL CHEMISTRY, VOL. 58, NO. 12, OCTOBER 1986

separations that they could pick one out as a usable optimum. The main advantage is that it runs the instru­ ment by itself and performs a lot of separations." Ed Kikta, manager of analytical ser­ vices at FMC Corporation, affirms that the commercial optimization pro­ grams fail on occasion. "We've seen them diverge instead of converge," says Kikta. "That can happen very easily. However, there are also cases where they come up with the same conclusions a good methods develop­ ment chemist would come up with, and we've seen them come up with better conclusions. Still, if you put one up against the other, the top-notch analytical chemist is still going to do a better job." Where will it all lead? "One idea we have for an expert sys­ tem," explains Joseph Kosman, "is one where a client could come into the laboratory with a sample and be prompted by a computer to answer questions about it. Based on those re­ sponses the computer would choose a column and starting solvents. This in­ formation would be fed to an LC opti­ mization system, which would opti-

ι80° C

S8°C 5S°C

Focus

" The problem is that a computer is just not smart enough to pick up the little cues that a chromatographer sees the significance of immediately." mize and run the separation. Then the information would be fed into an expert system, which would evaluate it. If you couple that with a robot, you could automate sample preparation, sample introduction, running of the sample, and obtaining the final results. "So you could think about the possibility of a lab without a chromatographer," Kosman continued. "You replace the expert with a computer that has the knowledge base." "I would disagree totally with that," Lloyd Snyder responds. "Human beings are good at one kind of activity and computers are good at another. The problem is that a computer is just

not smart enough to pick up the little cues that a chromatographer sees the significance of immediately." And Adrian Wade adds that "even if the difficulties of getting 25 years of expertise out of an expert and into a computer can be overcome, the expert will not and should not be done away with. Expert systems will simply make the expert more productive, leaving the expert more time to push forward the frontiers of science." But in at least one laboratory, a system very close to what Kosman proposes is currently being designed. The Analytical Director, a system under development in T. L. Isenhour's group at Utah State University, is an integration of laboratory robotics, analytical instrumentation, and computer software that Isenhour claims will design, test, and implement its own analytical procedures (J. Chem. Inf. Cornput. Sci. 1985, 25, 292-95). "We hope to demonstrate that a computerized artificial intelligence robotics system can function as an expert in analytical chemistry whose abilities continue to expand with experience," says Isenhour. "Such a system would be able to transfer or even pool its knowledge with other labora-

WUtimately, one could imagine an entire integrated laboratory that would select or develop its own procedures for solving analytical problems.'' tory systems, such as a network of computerized robot analysts. Ultimately, one could imagine an entire integrated laboratory that would select or develop its own procedures for solving analytical problems." How does Isenhour feel about the controversy about whether the expert can ever be replaced by the expert system? "I think the jury is still out," he says. "Can we replace him right now? No. On the other hand, my hobby is flying, in which automated systems can now land DC-10s and 757s in zero visibility. These systems put the planes on the runway and apply the brakes. And the pilot is just watching the dials while it happens." S.A.B.

UNVEILING THE WORLD'S MOST SENSITIVE ELECTROCHEMICAL DETECTOR Advancing The Art Of Chromatography Twenty-five years experience in the development and manufacture of electrochemical instrumentation gives EGfirG Princeton Applied Research the know-how to develop an Electrochemical Detector for a broad range of measurements. This new Model 400 is able to detect and measure extraordinarily low signals permitting truly usable operation at currents down to 10 picoamps.

A Masterpiece Of Convenience... The Model 400 proves that electrochemical detectors do not have to be finicky. Its unique cell design plus remote autoffset, computer control, menu-driven display operation and pre-electrolysis control make the Model 400 convenient and easy to use.

... And Of Versatility The EGS*G Princeton Applied Research Model 400 provides a unique combination of: • DC Mode • Dual Potential Mode • Pulse Mode • Oxidative and Reductive Detection • Cyclic Voltammetry • A Complete Selection of Cell Components (including a metal-free cell system) • Computer Interface (IEEE and RS232) and Pre-electrolysis Control, to name a few.

An Exhibition Of Quality and Of Service Contact our LC Applications Group for a demonstration or for more information on how the Model 400 Electrochemical Detector advances the art of chromatography. Call (609) 895-0100 today! Model 400 Electrochemical Détecte*

Π EE&G

PRINCETON APPLIED

RESEARCH

Electrochemical Instruments Division CN 5206 • Princeton NJ 08543-5206 • (609) 895-0100 • TELEX: 843409 CIRCLE 47 ON READER SERVICE CARD

1200 A · ANALYTICAL CHEMISTRY, VOL. 58, NO. 12, OCTOBER 1986