PII: S0098-1354(00)00590-1 Computers and Chemical Engineering 24 (2000) 2339 – 2350 Use of expert systems for the synthesis of downstream protein processes M. Elena Lienqueo *, Juan A. Asenjo Department of Chemical Engineering, Centre for Biochemical Engineering and Biotechnology, Uni6ersity of Chile, Beaucheff 861, Santiago, Chile Abstract This paper describes recent developments in rational process design and their application in biotechnology for large-scale downstream processes. For implementing an expert system, it is necessary to divide the separation process in two parts: recovery and purification, because the information and available heuristic knowledge are different in each part. In the case of the recovery process, the implementation of an expert system, called Prot – Ex, only use heuristic rules from literature and human experts. The sequences suggested were tested with a real example and it gave a satisfactory result. For the purification process two criteria have been defined for selecting the optimal sequences, the SSC criterion and the purity criterion. Both criteria were implemented in an expert system, Prot – Ex – Purification. This expert system was validated with experimental examples, in two cases and, the sequences suggested were adequate and valid. However, the sequences suggested by purity criterion have lesser steps than the sequences suggested by SSC criterion and the purity algorithm spends lesser time and computational resources than SSC algorithm, then the use of purity criterion is more recommendable for selection of purification sequences. Finally, we consider the proposed sequences found by the expert system a very good starting point for industrial process selection, a clear case of ‘expert amplification’. © 2000 Elsevier Science Ltd. All rights reserved. Keywords: Protein recovery; Protein purification; Downstream processing; Bioseparation; Expert system www.elsevier.com/locate/compchemeng 1. Introduction In the modern biotechnology industry, after success- ful fermentation or enzyme reactions the desired product must be separated and purified. The necessary steps to obtain these are commonly known as down- stream processing (DSP), which can account for up to 60% of the total cost, excluding the cost of the pur- chased raw materials (Lee, 1992). DSP is usually com- posed of a sequence of recovery and purification operations, whose final aim is to obtain the required protein at a prespecified level of purity. The recovery process is characterised by the objective of obtaining the product in solution from the production system. Purification takes the multiprotein solution and purifies the individual protein product to a high purity level, which is generally more than 90% pure. In addition, on a large scale, it is necessary to obtain the highest possible yield while minimizing the resources utilised and hence the cost (Asenjo, 1990). The main steps in a large-scale protein separation process are usually not more than eight, they are shown in Fig. 1. The principal unit operations used in DSP are shown in Table 1. At present, the bio-product markets are becoming very competitive. Therefore, it is important to choose the optimal flowsheet as early as possible, since once the process has been approved by regulatory agencies, its characteristics are fixed and are expensive to change. Thus, it becomes necessary to use rational design tools for this purpose. In order to achieve this we can recognise two different tasks: 1. Selection of separation sequence or flowsheet generations. 2. Design of individual unit operations. In this paper we will center our attention in the first task. Selection of separation sequences is a complex procedure in which a design evolves from a preliminary to a final stage in a trial and error fashion by repeatedly revising and refining the initial assumptions and restric- � II Pan American Workshop in Catalysis and Process Systems Engineering, September 2 – 3, 1999, Santa Fe, Argentina * Tel.: + 1-405-3255811; fax: + 1-405-3255813. E -mail address: bagajewicz@mailhost.ecn.uoknor.edu (M. Baga- jewicz). 0098-1354/00/$ - see front matter © 2000 Elsevier Science Ltd. All rights reserved. PII: S0098-1354(00)00590-1 M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 23502340 tions. For protein recovery processes, the flowsheet generations always have the same type of unit opera- tions (see Table 1). The problem that has to be solved is, e.g. choosing between alternative operations. For example, for cell disruption it is necessary to select between a homogenizer and a bed mill or for cell separation between cross-flow microfiltration and cen- trifugation. This selection is more or less done using heuristics, i.e. using rules of thumb to arrive at a rapid and reliable specification of equipment type (Leser & Asenjo, 1992). Another way to design is using appropri- ate mathematical correlations and models for optimising the task. On the other hand, for the protein purification pro- cesses the flowsheet consists of a chromatographic se- quence (see Table 1). This sequence has to be satisfied with maximum yield and a minimum number of steps (1, 2 or 3). This part of the process is undertaken in classical chemical process engineering, finding a rigorous solution using numerical methods like mathematical optimising techniques (Jennings, Teo, Wang & Yu, 1995) or using an expert systems approach (Eriksson, Sandahl, Brewer & Osterlund, 1991; Forslund, 1995). For the selection of an optimal sequence, purely mathematical techniques have limited use in biotechnology because of a lack of useful design equations and databases. The expert sys- tems approach is more attractive because it allows the use of empirical knowledge that is not rigorous in nature and is typical of that used by experts (Asenjo & Maugeri, 1992). In this paper, we show the implementa- tion and validation of an expert system for recovery operations and another one for purification processes. Fig. 1. General flowsheet for downstream processing of proteins. Table 1 Principal steps and unit operations used in downstream processing StepStages Unit operations Protein reco6ery Cell separation Centrifugation, membrane process (filtration, microfiltration), two phase aqueous partitioning Homogenisation, bead milling, chemical and enzymatic lysis and permeablizationCell disruption (only for intracellular proteins) Debris separation (only for Centrifugation, membrane processes (microfiltration, ultrafiltration), aqueous two-phase partitioningintracellular proteins) Concentration Ultrafiltration, precipitation Protein purification Pre-treatment or primary isolation Batch adsorption, ion exchange chromatography, affinity adsorption, hydrophobic interaction chromatography, aqueous two-phase partitioning Hydrophobic interaction chromatography (HIC), high resolution ion-exchangeHigh resolution purification chromatography, affinity chromatography, metal chelate chromatography (IMAC) dye-ligand chromatography HPLC Polishing of final product Gel filtration (GF) HPLC, ion-exchange chromatography M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 2350 2341 Table 2 Potential constraints ObservationsConstraints The product is only stable in a range of pHsRange of pH (e.g. 4–9) The product is stable only up to certainRange of temperature temperature Elimination of proteases, because they couldProteases degrade the protein product Inclusion bodies Inclusion bodies have to be solubilized and refolded for the production bulk industrial enzymes this is not the case. 1. How big is the flow rate? Bigger than 3 m3/h. Smaller than 3 m3/h. 3.2. Characterisation of starting material The second important point is the characterisation of the starting material. The process design will mainly depend on the physical, chemical and biochemical properties of the contaminating materials in the original broth and those of the protein that will constitute the final product. The properties of the starting material will be partially determined by: 1. Its fermentation source. Bacterial. Yeast. Mammalian cell. 2. How big is the cell concentration? 3. The type of cultivation medium used. Presence of albumin. Calf serum. Presence of proteases. Solid bodies like whole cells or cell debris. 4. Localization of the product. Intracellular. Extracellular. 5. Physicochemical properties of the product and the contaminant proteins in the final protein mixture in solution. Surface charge at different pHs and pI. Surface hydrophobicity. Molecular weight. Biospecificity toward certain ligands. 6. Stability of the final product is also of extreme importance as this will affect the types of operations that can be used as well as the conditioning and processing times that can be afforded. 3.3. Define possible separation steps and constraints It is necessary to have a database with all possible separation steps (Table 1) and to consider all potential constraints (Table 2). 3.4. E6aluation of potential process integration Finally, the production process should try to inte- grate as much as possible, upstream, fermentation and downstream process. Asenjo and Leser (1996) have described possible areas and levels of process integra- tion. Process integration can involve either the whole process, specific parts of it or particular unit operations as shown in Fig. 2. Examples of process integration between fermentation and downstream and between downstream operations are shown in Table 3. 2. General aspects for designing a protein separation process The design of a protein recovery and purification process shares many characteristics with other engineer- ing design activities. To design a process or an opera- tion requires: 1. To satisfy of a number of constraints, such as purity, quality, process temperature and pH, desired yield, etc. 2. To know properties of the materials, for instance chemical and biochemical properties, thermody- namic and fluid properties of the process material. Using the information and knowledge of (1) and (2) to propose a sequence of equipment interconnected in a particular order. 3. Basic information for designing a separation process The design of a process to economically purify a protein, maintaining a high yield, yet obtaining a virtu- ally pure product while minimizing the cost, requires four main considerations: 1. Defining final product. 2. Characterisation of starting material. 3. Define possible separation steps and constraints. 4. Evaluation of possible process integration. 3.1. Defining final product It is necessary to define the final product and to have information on its used. For instance: 1. How the product is going to be used? Industrial. Diagnostic. Laboratory reagent. Therapeutic. Questions regarding the use and purity are vital (e.g. 80, 99 or 99.98% pure) as well as allowable ranges of specific impurity concentrations. With therapeutic proteins any impurities have to be minimized whereas M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 23502342 Fig. 2. The possible areas of process integration. Fig. 3. The basic architecture of an expert system. The knowledge showed in the previous sections can be used for defining expert system for the rational selection of downstream protein processes. 4. Expert systems Expert systems are programs that attempt to solve problems in a way similar to how a human expert would solve them. They incorporate ‘rules of thumb’ that experts in the field have developed through years of experience. The problems tackled are not necessarily solved in a procedural fashion, they are often vague, complex, and can contain incomplete or inexact infor- mation (Nebendahl, 1988). Expert systems contain three basic elements: a knowledge base, an inference engine, and a user inter- face. The architecture of an expert system is shown in Fig. 3. The knowledge base contains the information, which the program uses to reach decisions. The infer- ence engine is the program that manipulates the knowl- edge base to reach these decisions. Finally, the user interface allows the program and the user to communi- cate with each other in an effective manner (Harmon & King, 1985). There are software systems, called shells, used for the manipulation of heuristics, databases, and algebraic equations (like design equations). For in- stance, Nexpert Object™ and Clips. There are many examples of the use of expert systems in chemical and biochemical engineering (Siletti, 1989; Mulholland, Walker, Manis, Hinriks, Buydens, Blaffer & Schoen- maker, 1991; Jakus, 1992; Forslund, 1995). 5. Implementation of expert system The implementation of an expert system for separa- tion processes for proteins was divided in two different parts: 1. A first expert system, called Prot – Ex, for recovery process selection. 2. A second one, Prot – Ex – Purification, for purifica- tion process selection. This division was carried out because both systems need different kind of information and the available heuristic knowledge is different, too (Asenjo, Herrera & Byrne, 1989). Prot – Ex uses only heuristic knowledge whereas Prot – Ex – Purification needs an important amount of quantitative data to make a good selection. In a consultation both expert systems are integrated and controlled by the main expert system. The main expert system: � Receives the information from the user, database or other sources. � Organizes and gives the information that each sub- expert system needs for suggesting a sequence. � Receives the sequences suggested by each sub-expert system � Gives the global sequences suggested to the user or other programme. Table 3 Examples of possible process integration Level of integration Action Cell immobilisation, cell recycling using membranes, use ofFermentation-downstream (the adequate design of the fermentation step can provide conditions to facilitate the downstream) expanded bed chromatography after fermentation Use of liquid-liquid separation, use of expanded bed for proteinInside downstream (recovery and purification) recovery, use of hydrophobic interaction chromatography after precipitation with ammonium sulphate M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 2350 2343 Fig. 4. Diagram of main expert system and relation with sub-expert systems Prot Ex and Pro Ex purification. The general diagram of the main expert system and sub-expert system with its relation is shown in Fig. 4. 5.1. Prot – Ex for reco6ery process The expert system for the recovery process was im- plemented using only heuristic knowledge. The heuristic rules were obtained from the literature and renowned industrial experts were consulted to clarify specific points for which knowledge is missing or there is am- biguous information. To do this, it was necessary to provide the experts with a questionnaire (Table 4). For recovery processes, it is only necessary to choose between two or three options. For example, for separa- tion of cells If is necessary to select between centrifuga- tion and cross-flow microfiltration. This selection is based on the following variables: � Thermal sensitivity of the product. � Shear liability of cells. � Size of the cells. � Process throughput. � Density difference between solid and liquid phase. � Cell concentration in the broth. Depending of the value of each variable it is possible to choose one or the other option. Considering the previous points, the expert system Prot – Ex was implemented. It has more than six hun- dred logical rules to emulate human reasoning. Its implementation was carried out in a commercial pro- gramme, Nexpert Object™ from Neuron Data. Consul- tations were carried out to validate the expert system Prot – Ex. The basic information used for the expert system is shown in Table 5. Table 6 displays a compari- son between a sequence suggested by Prot – Ex and a good process for recovery of somatotropin produced in Escherichia coli (Wheelwright, 1991). For this example, the recovery sequences suggested by Prot – Ex and the industrial process were very simi- lar. The difference with the published process is in step 1, where centrifugation was used instead of microfiltra- tion. For small cells such as E. coli microfiltration usually has clear economic advantage (Asenjo & Table 4 Typical questions presented to biotechnology experts for implementa- tion of the expert system for recovery processes (Leser, 1996) QuestionAspect Therapeutic What would your recommendation be if the product utilisation is therapeutic?product Are the common ‘broth conditioningCell separation processes’, like heating, freezing, coagulation, flocculation, enzymatic or other used to facilitate cell separation in recombinant DSP or not? Cell disruption Would you agree that the best suggestion of a mechanical method for breaking cells is ‘high pressure homogenizer’ for most real situations? Precipitation of What are the preferred methods for precipitation of nucleic acids?nucleic acid M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 23502344 Table 5 Basic information for designing a purification process of Somatotropin (Lienqueo et al., 1996) Defining final product Protein’s name Somatotropin 7.86 (Estimated)pI 22.0 KDa (Estimated)Molecular weight Surface hydrophobicity 0.93 M (Estimated) 25.0 mg/mlConcentration Titration curve pH 5.04.0 6.0 7.0 8.0 2.42 1.034.77 0.12Charge (coulomb/molecule)10−25 −0.03 Localization of the product Intracelullar, inclusion bodies IndustrialUtilization 4–9pH stability range 94%Final purity level Not knownSpecific affinity Characterisation of starting material E. coliFermentation source Patrick, 1990). Therefore, we considered that the se- quences suggested by Prot – Ex are adequate and it can be used as a starting point for the selection of a good industrial process. 5.2. Prot – Ex – Purification for selection of purification processes The selection of an efficient purification process is a critical step in downstream processes. These steps are not usually chosen in a rational manner, the common method being ‘trial and error’. Therefore, the use of some basic heuristic rules for separation processes has been proposed. These rules are: (Asenjo & Patrick, 1990; Prokopakis & Asenjo, 1990): 1. Choose the separation process based on the differ- ent physicochemical properties, such as: 1.1. Surface charge as a function of pH (titration curve) and pIs. 1.2. Surface hydrophobicity. 1.3. Molecular weight. 1.4. Biochemical properties such as biospecificity with different ligands. 1.5. Stability at different temperatures and pHs. 2. Eliminate those contaminant proteins and com- pounds that are found in greater percentage first. 3. Use a high-resolution step, as soon as possible. The chromatographic techniques ranked according to their efficiency are: 3.1. Affinity. 3.2. Ion exchange. 3.3. Hydrophobic interaction. 3.4. Gel filtration. 4. Do the most arduous purification step at the end of the process (final polishing). Considering those rules two selection criteria have been defined: Seperation Selection Coefficient (SSC) criterion and Purity criterion. The first one, SSC criterion, was developed by As- enjo and collaborators (Asenjo, 1990; Asenjo & Maugeri, 1992; Leser & Asenjo, 1992). The SSC crite- rion considers that the key parameter is SSC, which caracterizes the ability of the purification operation to separate two proteins. SSC value is defined as: SSC = DFhUi (1) Deviation factor (DF), this variable relates the differ- ence between a property of the target protein (for example, the dimensionless retention time in a specific chromatographic technique) and the same property of the contaminant. DF = �KDproduct − KDcontaminant� (2) Dimensionless retention time (KD), this variable rep- resents the behaviour of the proteins in a separation carried out by gel filtration, ion exchange or hydropho- bic interaction chromatography. Mathematical relation- ships for predicting KD have been derived using the physicochemical properties of proteins (Lienqueo, Leser & Asenjo, 1996). Table 6 Comparison between sequences suggested by Prot – Ex and industrial process for recovery of somatotropin (Lienqueo et al., 1996) Sequence suggested bySteps Industrial process Prot – Ex Centrifugation1 Crossflow microfiltration 2 High-pressureHigh-pressure homogenizationhomogenization 3 Centrifugation-pellet wash Disk centrifugation 4 SolubilizationSolubilization Renaturation5 Renaturation 6 Microfiltration Ultrafiltration 7 Concentration and diafiltration M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 2350 2345 Fig. 5. Algorithm of SSC criterion. Efficiency factor (h ) This parameter gives account of the unequal capability of the different separation pro- cesses to separate different proteins. Its value is con- stant for each type of separation and the chromatographic material used and have been mea- sured experimentally (Lienqueo et al, 1996). Concentration factor (Ui ) Its value represents the relative concentration of each contaminant. It will af- fect the selection criteria. In this way contaminants which are present in a high concentration have to be eliminated first. Ui = concentration of contaminanti concentration of all contaminants (3) The amount of contaminant eliminated after a chro- matographic step has been mathematically calculated M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 23502346 for different situations (Lienqueo, Salgado & Asenjo, 1999) Then the SSC criterion chooses as best chromato- graphic operation the one that has the highest SSC value (Leser, Lienqueo & Asenjo, 1996). After deter- mining which chromatographic technique gives the maximum SSC value, it is necessary to calculate the new concentration of all contaminant after the chro- matographic technique selected has been applied and construct a new database of contaminants concentra- tion. With this new database the system calculates the purity level and this value is compared with the level of purity required. The optimal sequence of steps is chosen until the required level of purity is reached. The al- gorithm used is shown in Fig. 5. Considering that the most important value is the final purity level and that now we had developed and al- gorithm to calculate the purity after a purification step, Fig. 6. Algorithm of purity criterion. M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 2350 2347 Table 7 Main variables used for SSC and purity criteria (Lienqueo et al., 1999) Variable Symbol Definition SSC criterion SSC = DFhUiSeparation SSC selection coefficient DF = �KDproduct−KDcontaminant�Deviation factor DF KD Relationships with physicochemicalDimensionless retention time properties of the proteins. (Lienqueo et al., 1999) Efficiency factor Values for each chromatographich techniques shown in Table 8 Peak width Values for each chromatographicS techniques shown in Table 8 UiConcentration Ui factor = concentration of contaminant i concentration of all contaminantsPurity criterion Pj PjPurity = concentration of the target protein Sconcentration of the proteins present one that has the least number of total steps. Consider- ing these criteria an expert system was implemented, Prot – Ex – Purification, and its implementation was car- ried out in the same shell than Prot – Ex. For testing these criteria, consultations were carried out to validate the expert system Prot – Ex – Purification using both criteria. Table 9 and Table 10 show the basic information used for purification of bovine serum albu- min (BSA) and b-glucanase, respectively and Table 11 and Table 12 show a comparison between sequences suggested by Prot – Ex – Purification and experimental sequence for both examples. 5.3. Comparison between SSC and Purity criteria In both examples the sequences suggested are valid and adequate. Nevertheless, in the first case, the se- quence suggested by the purity criterion has one fewer step than that suggested by the SSC criterion, that mean less capital cost. This situation takes place be- cause the SSC criterion considers the contaminant that gives the highest SSC value for one protein only. The purity criterion chooses the optimum chromatographic step considering all the contaminant proteins present. Hence, this situation can occur when there are several contaminants present in similar quantities as was the case in this example. On other hand, the purity criterion spends less time and computational resources than the SSC criterion, since the purity algorithm gives the optimal sequences in less time than the SSC algorithm. 5.4. Comparison between sequences suggested and experimental sequences In both examples the sequences suggested result in experimentally valid ones and are a good starting point for an industrial process. There are differences between the estimated and experimental final purity level, which are less than 20%. These differences could originate in different causes: � Hydrophobic interaction between the protein, this effect was not considered in the development of correlations for estimated dimensionless retention time for hydrophobic interaction chromatography (HIC) (Table 8). � Asymmetric chromatographic peaks, the peak shape could be different to a triangle shape. The differences between estimated and experimental purity level could be minimised: � Improving the prediction of dimensionless retention time for HIC. � Using a Gaussian approximation in the representa- tion of the chromatographic peak instead of triangle simplifications. Table 8 Expressions and parameters used for SSC criteria and purity criteria (Lienqueo et al., 1999) Efficiency factor h Peak width SChromatographic techniques Anion exchange 1.00 0.15 Cation exchange 0.151.00 0.86Hydrophobic interaction 0.22 Gel filtration 0.66 0.46 we implemented the purity criterion as a possible selec- tion criterion. This criterion compares the final purity level obtained after a particular chromatographic tech- niques has been applied. The purity concept is defined as: Purity = concentration of the target protein Sconcentration of the proteins present (4) After determining which chromatographic technique gives the highest purity level, the system chooses this as the technique to use at this step. A sequence of steps is chosen until the required level of purity is reached. Finally, the system creates a list with the defined se- quence of operations. The algorithm is shown in Fig. 6 The main parameters, variables and correlations used for SSC and purity criteria are summarized in Tables 7 and 8. Both criteria consider that the cost of each chromatographic option is equal, except for affinity chromatography. Therefore, the optimal system is the M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 23502348 6. Conclusions It is possible to use artificial intelligence tools, as expert systems, for selection of optimal protein separa- tion processes. For implementing an expert system, it is necessary to divide the separation process in two parts, because the information and available heuristic knowl- edge are different in each case. The first part for the recovery of proteins uses only heuristic rules from human experts and literature, the expert system was called Prot – Ex. The second part of a purification pro- cess is an hybrid expert system, which uses heuristic rules and mathematical correlations and models for selecting the optimal protein purification sequences, expert system called Prot – Ex – Purification. Both expert systems, implemented in the shell Nexpert Object™, were integrated by a main expert system. Considering the cases studied, like the recovery of somatotropin, the purification of BSA and the purifica- tion of b-glucanase, the sequences proposed by the expert system are a good starting point for a practical industrial process. Nevertheless, there are differences between estimated and experimental purity level, these could be minimized by improving correlations for pre- dicting retention time and approximation of the peak shape. In the specific case of Prot – Ex – Purification, the sequences suggested by the purity criterion have less Table 9 Basic information for desining a purification process of BSA from artificial mixture (Lienqueo et al., 1999) Defining final product Protein’s name Bovine serum albumin (BSA) 4.9p I 67.0 KDaMolecular weight Surface hydrophobicity 0.86 M Concentration 2.00 mg/ml Titration curve 5.0 6.0pH 7.04.0 8.0 −0.14 −1.16Charge −1.681.03 −2.5 (Coulomb/molecule) 10−25 Localization of the Extracellular product IndustrialUtilization pH stability range 4–9 Final purity level 94% Not knownSpecific affinity Characterisation of starting material Artificial mixture (BSA, ovalbumin, thaumatin, soybean trypsin inhibitor)Fermentation source Hydrophobicitymw (Kda)Contaminant Titration curve (coulomb/molecule)10−25Initial concentration (M)(mg/ml) pH 4.0 pH 5.0 pH 6.0 pH 7.0 pH 8.0 −2.3643.8 0.54 1.40 −0.76 −1.65Ovalbumin −2.202.0 SBTI 2.0 1.22 −0.76 −1.54 −2.17 −2.1324.5 0.90 0.89 1.94 1.90 1.98 1.87Thaumatin 2.0 0.9122.2 Table 10 Comparison between sequences suggested by Prot – Ex – Purification and experimental process for purification of BSA from a mixture of four proteins (Lienqueo et al., 1999) Sequence suggested by Prot – Ex – PurificationStep Experimental validation process Purity criterionSSC criterion Chromatography step Purity Chromatography step PurityPurity Chromatography step 64%Cation exchange pH 6.0 Anion exchange pH 7.0 64%33% Anion exchange pH 7.01 Hydrophobic interaction Hydrophobic interaction 80%49% Hydrophobic interaction2 95% 97%3 Anion exchange pH 7.0 M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 2350 2349 Table 11 Basic information for designing a purification process of b-1,3-glucanase (Lienqueo et al., 1999) Defining final product Enzyme name b-1,3-glucanase (bgl IIa) 4.1pI 31 KdaMolecular weight 0.00 MSurface hydrophobicity 0.62 mg/mlConcentration Titration curve 5.0 6.0 7.0 8.04.0pH 1.46 −0.62Charge −1.02 −2.33 −2.52 (Coulomb/molecule) 10−25 ExtracellularLocalization of the product Utilization Industrial pH stability range 4–9 70%Final purity level Not knownSpecific affinity Characterisation of starting material B.subtilus ToC46Fermentation source Contaminant mw (Kda) HydrophobicityInitial concentration pH 4.0 Titration curve (coulomb/molecule)10−25 (mg/ml) (M) pH 5.0 pH 6.0 pH 7.0 pH 8.0 41.0 1.5 1.461 0.262.74 −0.87 −1.65 −2.04 2 2.74 32.9 1.5 0.00 −2.70 −3.51 −3.51 3 35.50.25 0.2 −0.55 −0.22 −0.73 −1.82 62.5 0.0 −1.060.42 −1.174 −2.79 −3.32 40.6 0.0 −0.555 −0.220.09 −0.73 −1.82 69.6 0.0 −0.550.09 −0.226 −0.73 −1.82 0.257 40.6 0.0 1.46 −0.47 −1.06 −1.04 69.6 0.0 1.460.25 −0.478 −1.06 −104 Table 12 Comparison between sequences suggested by Prot – Ex – Purification and experimental process for purification of b-glucanase (Lienqueo et al., 1999) Sequence suggested by Prot – Ex – PurificationStep Experimental validation process Purity criterionSSC criterion Purity Chromatography step Purity Chromatography stepChromatography step Purity 32% Hydrophobic interaction 32%Hydrophobic interaction Hydrophobic interaction1 35% 70% Anion exchange pH 6.5 70% Anion exchange pH 6.52 60–70%Anion exchange pH 6.5 steps than the sequences suggested by SSC criterion, then the Purity criterion is more recommendable. On other hand, the use of expert systems for the synthesis of downstream protein processes is a clear case of ‘expert amplification’. Heuristic knowledge from experts was complemented with databases and design equations. To implement such a solution more globally there is a lack of databases of protein properties. With ad- vances in the area of proteomics and protein engineer- ing in a few years there could be sufficient information for such an approach. An effort should be made to generate such data. Acknowledgements Financial assistance from Fundación Andes and Vicerrectoria Académica, Universidad de Chile (Beca PG/080/97). This project was also supported by Proyecto Catedra Presidencial en Ciencias. M.E. Lienqueo, J.A. Asenjo / Computers and Chemical Engineering 24 (2000) 2339 – 23502350 References Asenjo, J. A., Herrera, L., & Byrne, B. (1989). Development of an expert system for selection and synthesis of protein purification processes. Journal of Biotechnology, 11, 275 – 298. Asenjo, J. A. (1990). Selection of operations in separation processes. In J. A. Asenjo, Separation processes in biotechnology (pp. 3 – 16). New York: Marcel Dekker. Asenjo, J. A., & Patrick, I. (1990). Large-scale protein purification. In E. L. V. Harris, & S. Angal, Protein purification applications: a practical approach (pp. 1 – 28). Oxford: IRL Press. Asenjo, J. A., & Maugeri, F. (1992). An expert system for selection and synthesis of protein purification processes. In P. Todd, S. K. Sikdar, & M. Bier, Frontiers in bioprocessing II (pp. 358 – 379). Washington: American Chemical Society. Asenjo, J. A., & Leser, E. W. (1996). Process integration in biotech- nology. In M. Verrall, Downstream processing of natural products (pp. 123 – 137). Chichester: Wiley. Eriksson, H., Sandahl, K., Brewer, J., & Osterlund, B. (1991). Reac- tive planning for chromatography: chemometrics and intelligent laboratory systems. Laboratory Information Management, 13, 185 – 194. Forslund, G. (1995). Designing for flexibility: a case study. Expert Systems, 12 (1), 27 – 37. Harmon, P., & King, D. (1985). Expert systems: artificial intelligence in business (p. 283pp). New York: Wiley. Jakus, W. (1992). Artificial intelligence in chemistry. Collect Czech Chemistry Communication, 57, 2414 – 2442. Jennings, L. S., Teo, K. L., Wang, F. Y., & Yu, Q. (1995). Optimal protein separation. Computers & Chemical Engineering, 19 (5), 567 – 573. Lee, J. M. (1992). Biochemical engineering (p. 320). Englewood Cliffs, NJ: Prentice-Hall. Leser, E. W., & Asenjo, J. A. (1992). Rational design of purification processes for recombinant protein. Journal of Chromatography, 584, 43 – 57. Leser, E. W. (1996) Prot – Ex: an expert system for selecting the sequences of processes for downstream purification of proteins, PhD thesis, University of Reading, England. Leser, E. W., Lienqueo, M. E., & Asenjo, J. A. (1996). Implementa- tion in an expert system of selection rationale for purification processes for recombinant proteins. Annals of the New York Acadamy of Sciences, 782, 441 – 455. Lienqueo, M. E., Leser, E. W., & Asenjo, J. A. (1996). An expert system for selection and synthesis of multistep protein separation processes. Computers & Chemical Engineering, 20, 189 – 194. Lienqueo, M. E., Salgado, J. C., & Asenjo, J. A. (1999). An expert system for selection of protein purification processes: experimental validation. Journal of Chemical Technology & Biotechnology, 74, 293 – 299. Mulholland, M., Walker, N., Manis, F., Hinriks, H., Buydens, L., Blaffer, T., & Schoenmakers, P. (1991). Expert system for re- peatability testing of high-performance liquid chromatographic methods. Journal of Chromatography, 550, 257 – 266. Nebendahl, D. (1988). Expert systems: introduction to the technology and applications (p. 211). New York: Wiley. Prokopakis, G. J., & Asenjo, J. A. (1990). Synthesis of downstream processes. In J. A. Asenjo, Separation processes in biotechnology (pp. 571 – 601). New York: Marcel Dekker. Siletti, C. A. (1989). Computer aided design of protein reco6ery pro - cesses, PhD thesis, (pp. 379). Massachussets Institute of Technol- ogy, Cambridge, USA. Wheelwright, S. M. (1991). Protein purification: design and scale up of downstream processing (p. 228). München: Carl Hanser. .