PDF hosted at the Radboud Repository of the Radboud University Nijmegen The following full text is a publisher's version. For additional information about this publication click this link. http://hdl.handle.net/2066/112320 Please be advised that this information was generated on 2021-04-06 and may be subject to change. http://hdl.handle.net/2066/112320 Analytrca C/zumcu Acta, 272 (1993) 41-51 Elsevler Science Pubhshers B V . Amsterdam 41 Expert systems in chromatography. Results of the ESCA project L Buydens Department of Analyttcal Chemistry, Kathoheke Unrverxte~t N&]megen, Toemoorveld, 6525 ED Ntpegen (Netherlands) P Schoenmakers Phrlps Research, P 0 Box 80000,5600 JA Emdhoven (Netherlands) F Marls and H Hmdnks Organon Intematwnal BC: Analytrcal R & D Laboratones, P 0 Box 20,534O BH Oss (Netherlands) (Recewed 9th July 1992) Abstract The final results of the ESCA project (Expert Systems for Chemical Analysis) are presented This IS one of the major projects m the field of expert systems for chromatography Expert systems have been developed that cover the unportant areas of method development m LC In the last part of the project attention was concentrated on two Lssues, the study of mtegratlon posslblhtles of the drfferent stand-alone systems and the Important aspect of vabdatlon and evaluation of the developed expert systems The mtegratlon studies and the results of the vahdatlon and evaluation are discussed Keywords Expert systems, Llquld chromatography, ESCA project The results of automation efforts by manufac- turers of chromatographlc instruments have led to an increased apphcablhty of chromatographlc mstruments for routme analysis The bottleneck of analysis IS situated mainly m the development of an optnnum method and m the interpretation of the results These processes usually require a lot of expertise and experience to solve the prob- lems that arise for each particular case Also, the quality control stage becomes an mcreasmgly nn- portant aspect to be automated As a result of automation, the numbers of analyses and results have grown so much that automatic quahty mom- tormg 1s necessary In view of the mcreasmg demands of good laboratory and management practice (GLP and GMP), this aspect will become even more unportant The mcorporatlon of ex- pertlse and expenence m instruments 1s therefore the next step to be taken Expert systems are software programs m which human expertise 1s unplemented Therefore, they seem to be the right approach for further au- tomation of instruments In other areas of chem- Istry they have already been demonstrated to be useful [l-3] In chromatography, a large amount of research has been carned out m a Jomt re- Correspondence to L Buydens, Department of Analytical Chemistry, Kathoheke Umversltelt NlJmegen, Toemooweld, 6525 ED Nqmegen (Netherlands) 0003-2670/93/$06 00 0 1993 - Elsevler Science Publishers B V All nghts reserved 42 L Buydens et al /Anal Chm Acta 272 (1993) 41-51 selection of application domaln t t ’ knowledge acquisition selection of tools knowledge implementation- validation Fig 1 Dlfferent steps III the development process of expert systems search proJect on the apphcablhty of expert sys- tems m chemical analysis (ESCA) [41 Partners from Industry and umversltles have cooperated to study the posslblhtles of the expert system ap- proach m LC In the first stage the development of LC methods for pharmaceutical compounds was selected as an apphcatlon area Within this area expert systems were developed that are rep- resentative of the whole area of method develop- ment selection of lmtlal method parameters, op- tlmlzatlon of selectlvlty and mstrumentatlon and finally vahdatlon of the method obtained As part of the project all aspects of the expert system building process have been investigated Aspects of integration and cooperation of expert systems have also been covered The project started m May 1987 and officially finished m May 1990 Durmg this period mterme- dlate results and fmdmgs have been commum- cated by means of presentations and so far about 25 papers have been published m mternatlonal Journals and numerous lectures and posters have been presented In this paper an overview of the most nnportant results 1s presented EXPERT SYSTEM DEVELOPMENT PROCESS The process of the development of expert sys- tems consists of several stages and has many different aspects It requires close cooperation between workers who have the necessary apphca- tlon knowledge and expertise and those response- ble for the lmplementatlon of the expert systems (“knowledge engmeers”) The different tasks are shown schematically m Fig 1 It shows clearly the sequence that has to be followed, the interaction between tasks and the loops that can occur The aim of the project was to study and demonstrate the apphcatlon of expert systems m chromatography Therefore, it was felt that a single apphcatlon was too lnmted to demonstrate the objective For that reason, a number of (rela- tively) small domams were selected based on cn- terra of usefulness, difficulty and vanety These domains were derived from LC method develop- ment as shown m Fig 2 This process resulted originally m four expert systems method selection and retention optimization 1 selectivity optimaization .l. system optimization 1 method validation Fig 2 Stages m the chromatographlc method development L Buydens et al /Anal Chm Acta 272 (1993) 41-51 43 Knowledge acqulsltlon 1s the process of ex- tracting knowledge as complete as possible from the chromatography expert This knowledge should then be unplemented m a choosen tool by the knowledge engmeer, which m this instance was either a chemometrlclan or a software spe- cialist The latter reqmres more time to become familiar with the domain knowledge, but may reahse better systems m terms of completeness, consistency and appearance Chemometnaans, on the other hand, can acquire more chemical knowledge m a shorter period of time, but the quality of the resulting software can be inferior To implement expert systems m a computer, a software tool 1s required This tool can be a standard computer language such as Pascal or C However, the lmplementatlon of expert systems with classical languages requires considerable software engineering expenence and a large ef- fort m terms of manpower In recent years dedl- cated tools for developing expert systems have become available These tools are often referred to as “expert system shells” The available tools range from simple to very sophlstlcated One of the purposes of ESCA was also to evaluate the sultablhty of these tools Therefore, it was neces- sary to select some suitable tools to implement the apphcatlons This selection was based on the lmplementa- tlon of a small test knowledge base m different tools of varying size (large, mid-sized and small, see Table 1) The test knowledge base contained essential features of the final knowledge and was obtained from earlier work by De Smet et al [5] TABLE I List of tools evaluated Size Name Orlgm Runmng on Small Delfi 2 Medmm Goldworks a KESa Mylog Nexpert object a Large Sl KEE Knowledge Craft Netherlands PC USA PC USA PC France PC USA PC USA WorkstatIon USA WorkstatIon USA WorkstatIon TABLE 2 Summary of development environment of the expert systems Domam Shell Expertise centre a Know- ledge engl- neermg centre a Method selectlon and KES VUB VUB retention optimization Organon Optlmlzatlon systems Selectivity optlmlzatlon KES WB WB Phlhps NL System optlmlzatlon Nexpert Phdlps NL Phlhps object + Hamburg Pascal Method vahdatlon systems Repeatablhty system Gold- works Umcam UK KUN Ruggedness system Gold- works Umcam UK KUN a WB = Free Umverslty Brussels, Belgmm, KUN = Cathohc Umverslty Nljmegen, Netherlands, Phlhps NL = Research Lab, Emdhoven, Netherlands, Phdlps Hamburg = Research lab, Hamburg, Germany, Organon = AnalytIcal R&D Labs, Oss, Netherlands, Umcam UK = Umcam, Cambndge, UK In general, it was concluded that large tools such as KEE or Knowledge Craft were too comph- cated They also require large workstations or microcomputers to run on The small tools were clearly not adequate, e g , limited lmplementatlon posslblhtles, limited or no access to external databases and poor quality of the end user and knowledge engineering interface It was finally decided to use a selected group of mid-sized tools, which had the additional advantage of run- ning on normal PCs [6] Table 2 summarizes the eventual tools from which the systems were built Expert systems can only be expected to be useful m practice if they are reliable Conven- tional software products can be tested thoroughly through a number of standard procedures How- ever, for testing expert systems no standard pro- cedures exist With expert systems both the soft- ware and the knowledge base must be reliable and correct The fact that the knowledge base often contams a lot of heurlstlc knowledge poses specific problems to the testing phase [7,8] Con- sidering the increasing demands of GLP, this part m the development process cannot be overestl- d These were the tools selected 44 L Buydens et al /Anal Chun Acta 272 (1993) 41-51 mated In view of this, considerable effort was devoted to the vahdatlon and evaluation of expert systems during the last part of the ESCA project EXPERT SYSTEMS OF ESCA trast to DASH Compounds that are subjected to a purity check usually contam less then 5% of unknown lmpuntles Optmuzatlon 1s then usually not required LABEL was added to be able to study the integration of method selection systems with optlmlzatlon expert systems The expert systems that were developed m the ESCA project can be dlvlded mto two categories stand-alone systems and integrated systems A complete overview 1s given m Table 3 In this section the stand-alone systems are considered LIT 1s a small expert system that helps to select all important parameters of a literature method and that checks whether a hterature method can be treated by SLOPES Method selection always starts with the choice of the chromatographlc mode, be It GC or LC In LC a further refinement can be made by choos- mg, for example, the normal- or reversed-phase mode, and further a C, or a cyano phase In this way a decision tree can be built When the experunental results are not satls- factory, all three expert systems have an exten- sion by which adaptations of the method are suggested m order to obtain an acceptable reten- tion range of the compounds DASH (Drug Analysis System m HPLC) 1s the system that assists m the selection of LC starting condltlons for the purity check of pharmaceutical compounds Because of the complextty of the relationship between the structure (input) and suitable percentage of modifier (output), this sys- tem was developed only for heterocychc basic compounds As most of these compounds are new chemical entitles, there 1s no literature avall- able on LC analyses for these compounds [9l The next step m method development 1s selec- tivity optlmlzatlon Thus step typically mvolves the optlmlzatlon of the mobile phase composition m order to obtain an optimum dlstrlbutlon of the peaks over the chromatogram LABEL 1s an expert system that was devel- oped by one of the partners (VUB) before ESCA started It selects a method for the LC of drugs m pharmaceutical formulations (label clann analy- sis) [5] It was included m the project because It covers he sltuatlon that one sample must be analysed Ior different compounds This 1s m con- SLOPES (SeLectlWy OPtlmlzatlon Expert System) 1s an expert system which typically ad- dresses one of the nnportant aspects of selectlvlty optmuzatlon Imtlally attention was focused on the selection of an appropriate optlmlzatlon cn- terlon In the past, many optnnlzatlon crlterla have been put forward However, it should be recogmzed that a smgle criterion 1s not always the best one m all situations [lo] SLOPES will help the chromatographer to select the most approprr- ate criterion, which will then be used to Judge the quahty of the chromatogram [ill This selection of a criterion depends, for example, on the se- lected expernnental design and on the ObJectwe of the optumzatlon (e g , best spreadmg of peaks TABLE 3 OvervIew of the ESCA expert systems Method develoument Stand-alone Integrated stage 1 Imtlal method selection and retention optumzatlon 2 Selectivity optimization 3 System optmuzation 4 Vahdatlon a Names are explamed m the text expert systems a DASH, LABEL, LIT SLOPES SOS REPS RES expert systems (INT) a INTI DASH LABEL LIT + SLOPES INT II INT III SOS + SOS + REPS RES L Buydens et al /Anal Chun Acta 272 (1993) 41-51 45 or muumum analysis tune) Once the optnnum flow-rate It predicts also the reqmred analysis selectlvlty has been obtamed the mobde phase tune and the crltlcal resolution A result of a composltlon and stationary phase are kept con- consultation and the experunental venficatlon 1s stant for the next step, system optmuzatlon shown m Fig 3 SOS (System Optnmzatlon expert System [12]) The final step m method development 1s the can be used here to select a column with the vahdatlon of the method This means that the shortest analysis tune from a column set grven by quality of the results should be guaranteed to a the user In addltron, the user should also provide certain extent The nnportance of validation 1s a set of avadable detector cells and a hst of stdl increasing m view of mcreasmg GLP de- allowed time constants Fmally, some hnuts mands The level of vahdatlon depends mamly on should be given, such as the required mmmmm the intended use of the method A higher level of resolution between a relevant pair of peaks, maxi- vahdatlon IS required if, for example, the method mum pressure drop and maxunum flow-rate 1s to be used m a large number of laboratones Wlthm these constraints SOS recommends the over a long period of tnne Methods to be used column, Instrument parameters and optimum for regulatory analysis need the highest level of a Time (min) Fig 3 Optlmrzatlon mth the SOS expert system (a) Chromatogram before optlmlzatlon, (b) chromatogram obtamed with the condltlons as advtsed by SOS, by which an analysis tnne and a resolution of 6 2 mm and 16, respectwely, were predrcted 46 L Buydens et al /Anal Chm Acta 272 (1993) 41-51 vahdatlon Vahdatlon of a method requires the testing of its speclficlty, precision, accuracy and hnutatlons These so-called performance charac- teristics may be tested by vahdatlon procedures Many of these procedures make use of complex mathematics and rely on statistical designs, such as the Plackett-Burman design (e g , for precl- slon testing) Expert systems can be of great help here m guiding the (mexpenenced) user m the set-up of such advanced designs, m the calcula- tion and m the interpretation of the results In the ESCA project, precision testing was chosen as the most challenging Item m method vahdatlon to demonstrate the apphcablhty of expert systems Precision testing mvolves, m fact, three separate parts repeatability, reproduclblhty and rugged- ness In the repeatability test the analysis 1s re- peated a number of times under Identical condl- tlons Thrs 1s m contrast to the reproduclblhty test, where different conditions, e g , different instruments m different laboratories, are applied Finally, m a ruggedness test the effect of small changes m the operating condltlons, 1 e , temper- ature, flow-rate and mobile phase composltlon, on precision 1s tested Repeatability and rugged- ness testing were the SubJect of different expert systems REPS (Repeatablhty testing System [13,14]) is an expert system m which Goldworks software is combined with the Lotus l-2-3 spreadsheet package The expert system 1s used to select test procedures for repeatability Based on the usage reqmrements, an experimental design IS set up The spreadsheet can run the algorithms and cal- culates the variances for peak areas and heights and retention times The expert system 1s able to interpret the results and to perform a diagnosis based on how the above parameters vary to- gether An example of a rule can illustrate how a diagnosis proposal IS reached For instance, if the variance of retention time and the variance m peak areas are large, and the variance of peak heights 1s small, then it 1s concluded that the problem of repeatablhty 1s unpreclslon of the flow-rate RES (Ruggedness Expert System [15,161) IS a modular expert system m which the Goldworks software 1s combined with the procedural lan- guage C It 1s intended to assist m the proper set-up of a complete ruggedness test This m- volves heurlstlcal (experience-based) knowledge to select the proper factors (and appropriate iev- els) to which the method should be rugged Sta- tlstlcal knowledge 1s necessary to choose a proper design based on the selected factors and the intended usage of the method The experimental results are interpreted If applicable system suit- ability criteria are provided, factors that cause problems are Identified INTEGRATION STUDIES [17,18] The stand-alone expert systems described above all tackle a specific sub-problem of the method development process These systems are implemented m different shells and run on dlffer- ent hardware Ideally, the chromatographer should be able to consult the system that 1s needed in a specific situation as part of a complete method development expert system Also, from the vlewpomt of the knowledge engineers it was seen as a challengmg task to Integrate stand-alone systems of different orlgms Because of the three knowledge engineer centres involved m the pro- ject, It was decided to study three partial mtegra- tlons (see Figs 4-6) There are two important aspects to this First, the analytical experts had to reahze that to pro- duce meaningful integrations It was necessary to fill knowledge gaps between the different stand- alone systems, so that additional knowledge ac- qulsltlon was inevitable This resulted m conad- erable extensions of the exlstmg systems and m the addition of new expert systems, such as LA- BEL and LIT Second, the knowledge engineers had the dlfflcult task of hnkmg sub-systems of different origins mto an acceptable architecture ZNT Z [17] The structure of the architecture of INT I IS given in Fig 4 In this scheme the supervisor 1s the essential part, having the strate- gic knowledge to route the end user to the dlffer- ent expert systems INT I is a typical example for which relatively much addltlonal knowledge was necessary for the integration and Integration was necessary m order to obtain a sultable system As L Buydens et al /Anal Chm Acta 272 (1993) 41-51 Fig 4 Structure of the Integrated system 1 (INT I) an extension, it IS felt that the integration with the SOS system 1s necessary because using SOS strongly influences the selection of the optlmlza- tion criterion ZNTZZ [19] Figure 5 shows the structure of the second integrated system Two of the five subsys- tems (REPS and SOS) are the orlgmal stand-alone systems whereas the other modules were built to add flexlblhty to the system This architecture allows the user to consult the system m three different situations It can be used to assess the repeatability of a new method or to check the repeatability of a previously validated method Also, the posslblhty of usmg the system as a trouble-shooting tool turned out to be a valuable feature ZNT ZZZ Another posslblhty to link stand-alone expert systems 1s shown m Fig 6 In the rugged- ness expert system (RES) the system optimization system (SOS) 1s incorporated as an extra module The scheduler has knowledge on when to activate which module The different modules take care of the different tasks m the ruggedness test and the SOS module has been added to provide solu- tions for problems that have been detected by the diagnosis module SOS can be used m a number of situations Pnmanly, SOS can help to improve 41 a method when the resolution has fallen below a critical level during the ruggedness testing SOS can then propose new condltlons based on the requirement for higher resolution Both systems had to be adapted slightly and/or extended m order to make a sensible integration These studies show that integration often re- sults m complex structures, 1 e , they are less user friendly This endorsed m fact the orlgmal dea- slon to build limited stand-alone systems On the other hand, it was shown that integration 1s use- ful m situations where the chromatographer often has to switch between systems Slmllar conclu- sions have also been reported elsewhere [20] VALIDATION AND EVALUATION OF THE ESCA EX- PERT SYSTEMS Considerable attention was paid to the testing of the expert systems [21,22] It 1s important to note that systems were tested with special empha- SIS on their performance rather than on appear- ance aspects such as a nice user interface The latter should, however, be of sufficient quality to make an understandable system Two mam stages have been dlstmgulshed, the vahdatlon and the evaluation stage The vahdatlon process involved checking the software and testing the knowledge base by the responsible expert The procedure that was fol- lowed involved the selection of a number of test cases by the expert The expert solved the test cases manually, while the expert system was also Requirements Editor subsystem Method Editor subsystem System Optimization subsystem Repeatability Testing subsystem Perform User Action subsystem Fig 5 Structure of the Integrated system 2 (INT II) I Common Datastructure Fig 6 Structure of the integrated system 3 (INT III) consulted The test cases were selected wlthm the scope of the systems and to make use of as much of the knowledge base as possible Whenever differences between the expert systems solution and that of the expert were seen, the cause of the discrepancy was identified This led to the addl- tlon of mlssmg knowledge or to the correction of L Buydens et al /Anal Chrm Acta 272 (1993) 41-51 exlstmg knowledge To decide on the proper per- formance of the systems, a set of pass/fall cnte- na were defined by the knowledge engineer and the expert prior to testing The systems were improved untd agreement was reached between the expert and the expert system After vahdatlon the systems were subjected to the second stage, the external evaluation The evaluation phase consisted of testing the expert system in practical situations, to evaluate the system’s performance m dally practice Gen- erally, these tests were performed by external evaluators, 1 e , chromatographers not involved m budding the system Unbiased problem cases were put to the expert system All inputs and outputs of the systems were registered and, whenever possible and appropriate, verified with expen- ments A list of performance criteria were ldentl- fled by the knowledge engineers and the experts for each system These criteria took mto account aspects of the man/machine interface, the con- sistency of the system and its hmltatlons A hst of criteria IS given m Table 4 The evaluations were carried out by different persons, ranging from experts m method develop- ment to students with little or no experience Summarlzmg the evaluations of the three mte- grated systems the followmg conclusions can be drawn The user friendliness expressed among others m a good user interface, clear screen text and easy help functions was Judged to be good m TABLE 4 Example of evaluation criteria Man-machine interface (user Interface) Choice of phrases Explanation Consistency testing System limits Operation (mouse, keyboard, file Input, etc 1 Usabdlty/ease of use Accuracy (correct answer, quality of advice) Reproducibility (repeatability, same mput same output) Robustness of software (does the system lock up or fall over) Ruggedness (small changes m input small changes m output, similar cases, similar answers) Conflict (two rules with the same input give a different output) Missing rules (input leads to no realistic output) Are essential parts missing? Are there examples of strange answers m extreme cases, e g , incomplete input, nonsense output? Techmcal content do the systems do a useful Job7 L Buydem et al /Anal Chm Acta 272 (1993) 41-51 49 INT III and INT III Some of these aspects are closely related to the quality of the shell KEiS (INT I) belongs to the older generation of shells m which the above features can be improved Case study pH optlmrzatwn With respect to the knowledge, a great variety exists between the systems The knowledge col- lected m INT I IS most complex and generally of heuristic nature Although thrs system was re- stricted to basic pharmaceutrcal compounds, there 1s still a lot of chemical and analytical knowledge to add The knowledge m the other systems 1s better defined and proved to be complete m a broad field of apphcatlons The evaluators found the item “factor choice m the ruggedness module very flexible On the other hand, they asked for more flemblhty m the experunental design In the followmg example an mterestmg apph- cation of expert systems 1s shown m which algo- rithmic-based knowledge 1s combined with heuristic knowledge The complex@ of some steps m the method development process til be demonstrated INT I deals wth method selection and selec- tlvlty optunlzatlon In this system three modules are present for the method selection and subse- quent retention optmuzation After an experi- ment it has to be decided whether the selectivity has to be optmuzed The expert system ade- quately helps to select a method for the selectlv- ity optimization, ~12, sequential or simultaneous approach, and which parameters have preferably 1400 1 a 1200 j 400- 7 1400- 1200- 1000- 600- 600- 400 3 0 “_ 5 10 1s 20 Tlma (mtn 1 Fig 7 Optmuzatlon with SLOPES (a) Chromatogram m one of the expenments, (b) result obtamed after optlmuatlon wth SLOPES 50 L Buydens et al /Anal Chm Acta 272 (1993) 41-51 to be optlmlzed, VU, percentage of modifier, mixture design, temperature, pH The next step 1s to carry out the optnnlzatlon It was chosen to implement only the software tools to carry out pH optlmlzatlon m a simultaneous approach The pH optlmlzatlon was selected because this 1s a relatively new area For other types of optlmlza- tlon one can use commercially available software tools Before the experiments for the pH optlmlza- tlon can be carried out, the parameter space has to be defined, an expernnental design has to be selected and a criterion has to be chosen for the calculation of the optimum In these three mod- ules heuristic knowledge for the selection of pa- rameter space for aads and for bases, chemomet- nc knowledge for the selection of an appropriate design and algorlthmlc knowledge for the calcula- tion of the optimum 1s used In Fig 7, two chromatograms are shown, one before and one after pH optlmlzatlon The opti- mlzatlon was done by means of SLOPES The mam hmltatlon of the system IS to describe accu- rately the retention behavlour of each solute over the parameter space selected (retention surface) It 1s well known that the relationship between retention and pH 1s an S-shaped curve The cal- culation of the retention surface through the measuring points cannot be done by a quadratic function However, to keep the number of meas- urements small it was decided to fit a quadratic function through the data points and to study the pH variation over only a small pH range of 3 units By this means a reasonably accurate pH optlmlzatlon could be achieved Main results of the evaluation Expert systems can provide very powerful as- sistance during method development because of the heuristic knowledge (expertise and expen- ence of a specialist) that 1s implemented m these systems However, during the evaluation phase of all the expert systems it became clear that the attitude towards expert systems 1s strongly de- pendent on the expertise level of the evaluator The accesslblhty of the specialist’s expertise was clearly appreciated by inexperienced users The mtroductlon of the expert systems resulted for those users m a considerable amount of time saving m the development process Experienced users could appreciate the quality of advice given by the systems They are more mterested, how- ever, m comparing the expertise m the systems with then own experience When the strategy implemented m the system did not agree with then own expertise, it resulted m dlssatlsfactlon wrth the expert system, because their own expen- ence, probably better adapted to their specific atuatlon, IS not considered by the system This 1s especially the case when the knowledge domam of the expert system 1s strongly susceptible to mdlvldual opmlons This gives rrse to a second aspect, that at present expert systems are not flexible enough (Minor) changes that could result m a better performing expert system for a partlc- ular situation may be lmposslble to make by the user Only the knowledge engineer who 1s aware of the structure of the knowledge base 1s able to make such changes without unexpected conse- quences A third conclusion of the evaluation 1s that the mam attention should be devoted to the mtegra- tlon of expert systems with laboratory mstru- ments When this 1s reahzed, expert systems can be used to obtain rapidly accurate advice while the user remains free to choose his or her own approach Concluszons The ESCA project can be considered as a pioneer project for the apphcatlon of expert sys- tem technology m analytical chemistry Method development m LC was selected as the apphca- tlon expertise area The expertise that has been considered m the project covers the important areas of method development m LC It was only possible to cover this large area because many recogmzed experts participated m the project It would be almost lmposslble to fmd a single expert to cover all the different aspects of method development Different expert systems resulted from the pr@ Ject Most of these are still m the research phase, but a few have been further developed for com- mercialization (system optimization system, ruggedness system) From the results of vahda- L Buydens et al /Anal Chrm Acta 272 (1993) 41-51 tlon and evaluation it can be concluded that expert systems are potentially very useful for method development m LC The benefits of the systems are that method development can be done more consistently and more efficiently and that better optimized and validated methods are produced Even when the systems were not yet complete this conclusion became clear Expert systems still have to fmd their way mto the chromatographlc laboratory Users will have to accept computer programs that assist during tasks such as method development This requires, as stated above, expert systems that are flexible and easy to mtegrate It should be possible to add new knowledge or to adapt the system according to changes m the apphcatlon environment Re- search on this subject remains necessary This research proJect was partly funded by the EEC, as ESPRIT project P1570 ESCA T Blaf- fert, A Cleland, T Hamolr, G Kateman, J A van Leeuwen, D L M Massart, M Mulholland, H PlnJns, B G M Vandengmste, N Walker and all other temporary participants are acknowl- edged for their contributions which made this final report possible REFERENCES B A Hohne and T H Pierce (Eds 1, Expert System Apph- catlons m Chemistry (ACS Symposmm Series, No 4081, Amencan Chemical Society, Washmghton, DC, 1989 J W Dolan, L R Snyder and MA Quarry, Chro- matographla, 24 (1987) 261 S S Wdhams, Trends Anal Chem ,9 (1990) 63 D Goulder, T Blaffert, A Blokland, L Buydens, A Chhabra, A Cleland, N Dunand, H Hmdnks, G Kate- man, H van Leeuwen, D Massart, M Mulholland, G 10 11 12 13 14 15 51 Musch, P Nalsh, A Peeters, G Postma, P Schoenmakers, M de Smet, B Vandegmste and J Vmk, Chro- matographla, 26 (1988) 237 M De Smet, A Peeters, L Buydens and D L Massart, J Chromatogr ,457 (1988) 25 J A van Leeeuwen, B G M Vandegmste, G J Postma and G Kateman, Chemom Intell Lab Syst , 6 (1989) 239 P Pohtakls and S M Weiss, ArtIf Intel], 22 (1984) 23 R Wehrens, L Buydens and G Kateman, Chemom In- tell Lab Syst , 12 (1991) 57 H Hmdnks, F Mans, J Vmk, A Peeters, M De Smet, D L Massart and L Buydens, J Chromatogr ,485 (1989) 255 P J Schoenmakers, The Optlmlzatlon of Chromatographlc Selectlvlty, a Guide to Method Development, Elsevler, Amsterdam, 1986 A Peeters, L Buydens, D L Massart and P J Schoen- makers, Chromatographla, 26 (1988) 101 P J Schoenmakers, N Dunand, A Cleland, G Musch and T Blaffert, Chromatographla, 26 (1988) 37 M Mulholland, J A van Leeuwen and B G M Vandegm- ste, Anal Chum Acta, 223 (1989) 183 M Mullholland, N Dunand, A Cleland, J van Leeuwen and B Vandegmste, J Chromatogr , 485 (1989) 283 J A van Leeuwen, L M C Buydens, B G M Vandegmste, G Kateman, P J Schoenmakers and M Mulholland, Chemom Intell Lab Syst , 10 (1991) 337 16 J A van Ixeuwen, L M C Buydens, B G M Vandegmste, G Kateman, P J Schoenmakers and M Mulholland, Chemom Intell Lab Syst , 11 (1991) 37 17 18 19 20 21 22 P Contl, H Puyns, N Vandendnesche, M Desmet, T Hamolr, F Mans, H Hmdnks, P Schoenmakers and D L Massart, Chemom Intel1 Lab Syst , 11 (1991) 27 L Buydens, J Van Leeuwen M Mulholland, B Vandegm- ste and G Kateman, Trends Anal Chem , 9 (1990) 58 M Mulholland, N Walker, F Mans, H Hmdrlks, L Buydens and T Blaffert, J Chromatogr , 550 (1991) 257 F A Settle and M A Pleva, Chemom Intell Lab Syst , 11 (1991) 13 J A van Leeuwen, L Buydens, B G M Vandegmste, G Kateman, A Cleland, M Mulholland, C Jansen, F A Mans, PH Hoogkamer and J H M van den Berg, Chemom Intel1 Lab Syst , 11 (1991) 161 F Mans, H Hmdrlks, J Vmk, A Peeters, N Vandendrle- sche and D L Massart, J Chromatogr ,506 (1990) 211