Selection of Bioprocess Simulation Software for Industrial Applications
Teri Shanklin,1 Keith Roper,2 P. K. Yegneswaran,2 Mark R. Marten1

Department of Chemical & Biochemical Engineering, University of Maryland Baltimore County, 1000 Hilltop Circle, Baltimore, Maryland 21250; telephone: 410-455-3439; fax: 410-455-1049; e-mail: 2 Merck & Company, West Point, Pennsylvania
Received 12 March 2000; accepted 29 June 2000

Abstract: Two commercially available, process-simulation software packages (Aspen Batch Plus v1.2, Aspen Technology, Inc., Cambridge, Massachusetts, and Intelligen SuperPro v3.0, INTELLIGEN, INC., Scotch Plains, Ner Jersey) are evaluated for use in modeling industrial, biotechnology processes. Software is quantitatively evaluated by Kepner-Tregoe Decision Analysis (Kepner and Tregoe, 1981). This evaluation shows that Aspen Batch Plus v1.2 (ABP) and Intelligen SuperPro v3.0 (ISP) can successfully perform specific simulation tasks but do not provide a complete model of all phenomena occurring within a biotechnology process. Software is best suited to provide a format for process management, using material and energy balances to answer scheduling questions, explore equipment change-outs, and calculate cost data. The ability of simulation software to accurately predict unit operation scale-up and optimize bioprocesses is limited. To realistically evaluate the software, a vaccine manufacturing process under development at Merck & Company is simulated. Case studies from the vaccine process are presented as examples of how ABP and ISP can be used to shed light on real-world processing issues. © 2001 John Wiley & Sons, Inc. Biotechnol
Bioeng 72: 483–489, 2001.

Keywords: process simulation software; process modeling; Aspen Batch Plus; Intelligen SuperPro

INTRODUCTION Process simulation software is a series of computer algorithms capable of mathematically modeling the performance of individual unit operations. Each model may describe several biological, chemical, and physical phenomena and may draw information from databases of material and equipment properties (Evans and Field, 1988; Gallier and Kisala, 1987). Aspen Batch Plus v1.2 (ABP) and Intelligen SuperPro v3.0 (ISP) are software packages designed specifically for the pharmaceutical and biotechnology industries and are intended to model all aspects of a process. Traditionally, process-simulation software has been used in the petroleum and chemical industries. Industrial develCorrespondence to: M. R. Marten Contract grant sponsor: Merck and Company, West Point, PA

opment of such software began in the late 1950’s, when software was often developed in-house by companies looking at custom applications (Biegler, 1989). Since this time, extensive distillation, extraction, and vapor–liquid equilibrium models have been developed. Because of this history, software is quite advanced at simulating chemical synthesis processes but considerably underdeveloped for the description of biotechnology operations (Potera, 1998). There are several challenges in the application of processsimulation software to biotechnology problems. First, unique unit operations such as fermentors, homogenizers, centrifuges, filters, and chromatography columns are used. These unit operations are not always well-characterized and work more on physical and biological phenomena than on the principles of phase equilibrium. Second, bioprocesses often involve underdefined raw materials and products such as cells and proteins for which physical properties or even structures may not be known. Additionally, substrates of unknown or varying composition such as soybean meal and yeast extract are often fed to the process. Third, the pharmaceutical industry often utilizes integration of batch and semi-continuous operations. This overlap between batch and semi-continuous steps can prove difficult to accurately simulate (Evans and Field, 1988). Finally, cleaning and sterilization of equipment and the sterilization of final products are necessary for the production of FDA-compliant pharmaceuticals. Simulation software must take into account the materials, equipment, and scheduling required for clean-inplace (CIP) and steam-in-place (SIP) operations. Development of process simulation software for the biotechnology industry has only taken place within the last 10 to 15 years. Because of its recent development there is limited published information on the subject. Those who have published on the subject consist mostly of software vendors discussing their own software (Galbe and Zacchi, 1992; Petrides, 1994; Petrides et al., 1995) or academics/ industrialists who have developed their own models for specific unit operations (Bulmer et al., 1996; Clarkson et al., 1996; Pascal et al., 1995; Zhou et al., 1997). This study is

© 2001 John Wiley & Sons, Inc.

Background The intent of this study is to determine the capability of off-the-shelf bioprocess simulation software to integrate all aspects of an industrial biotechnology process from unit operation modeling.unique in several ways: Only commercially available. fractional separations. and centrifugation for the removal of cell debris and protein. Aspen Batch Plus software does allow feed streams to be specified based on other feed streams. Once ratings were assigned the Kepner-Tregoe Decision Analysis (Kepner and Tregoe. Intelligen Super-Pro software has no means for tracking materials used during CIP and SIP operations other than treating these operations as an entirely new process. rather the exact masses of the salt and water must be entered into the software. software was rated on its ability to perform specific tasks in eight key categories covering the full range of the software’s capabilities. charge calculations require outside spreadsheet calculations. The final vaccine is madeup of 23 different polysaccharide antigens. 4. Therefore. Approximately 14 unit operations are involved including ultrafiltration.2 (ABP) and Intelligen SuperPro v3. and the software is being applied to an existing industrial process. This provided a large amount of process data from which to draw and allowed software unit operation models to be tested over a wide range of operating conditions. the software was applied to an existing process currently being developed at Merck & Company to produce a vaccine antigen. 1981) was used to convert these ratings into a flexible software selection tool which the process engineer can use to select a software package best catering to his or her needs.0 (ISP). When performing material balances in biotechnology processes. For example. 72. followed by purification of bacterial polysaccharide antigens. The software packages were eventually required to scale-up the process and predict manufacturing scale outputs. FEBRUARY 20. which could easily be incorporated into the software. Unit Operation Models Overall. Process data and analytical results from pilot-scale runs were used to set-up simulations using both software packages. integrated-process simulation. VOL. The results of these studies are provided as case studies in the Discussion section. software users throughout Merck were surveyed to determine the full utility of the software. These results were incorporated into the category ratings. However. These separation yields are simply constant yields. First. Because each of the 23 serotypes must be fermented and recovered separately. There is still room for improvement in both software packages in relation to performing charge calculations. These ratings were arrived at after many months of working intensely with the software to address processing issues and aid in scale-up of the vaccine process. each is produced by fermentation of a separate bacterial serotype that generates a unique immune response. the vaccine process is in essence made up of 23 individual processes each differing in their optimal processing conditions. In ISP. which cannot be set as a function of operating conditions or any process parameters. one must take into account the demands of cleanin-place (CIP) and sterilize-in-place (SIP) operations. flow rate and density of the stream. Additionally. Two types of models exist: reactions and separations. After working with the software for some time. the evaluation process began. A more detailed discussion of how the software faired in each category is provided below. Often. 2001 . further development is needed to allow feed streams to be charged based on vessel contents. if the user wishes to charge a 10 mmol/L buffer to a vessel. Performance in each category was rated on a scale of 1–5. eliminating some external calculations. material balance results are easily accessed by selecting the stream of interest in the process-flow diagram which opens a window detailing the composition. RESULTS AND DISCUSSION Evaluation of Software Table I represents the foundation of the evaluation performed in this study. To evaluate the software from an industrial perspective. Slightly more complex models include the centrifuge and 484 BIOTECHNOLOGY AND BIOENGINEERING. Only two software packages making such claims were on the market at the time of this study: Aspen Batch Plus v1. In ABP. Reactions require the user to enter the reaction stoichiometry while separations require the user to enter a separation yield for each stream component. the user cannot simply specify the buffer molarity. Processing consists of fermentation. software packages were selected based on their promise to provide complete. each category contains specific tasks the software was asked to perform. a simultaneous evaluation of two software packages is being made. Software packages designed to model specific unit operations were also found but did not fit the scope of our study. The ABP software has clean and sterile operations but these operations do not fully account for all equipment issues. reports must be generated which detail stream compositions. Material Balances Both software packages track material balances across an entire process. NO. A rating of “1” represents inability or very poor ability of the software to perform the task and “5” represents the completion of tasks in a manner that provides the user with quick and reliable results. offthe-shelf software is being examined. ABP and ISP unit-operation models are very simplistic. to managing equipment and expenses. The software is rated in eight catego- ries. The composition of intermediate and outletprocess streams is calculated assuming that the user has entered information about the feed streams and initialized all the unit operations.

etc. COMMUNICATION TO THE EDITOR 485 . 1—Poor.) a 4 1 4 2 3 1 1 1 1 3 4 3 3 2 1 2 1 3 1 3 4 1 1 5 2 2 1 5 3 2 2 3 4 3 1 1 1 1 3 2 2 5 4 3 3 5 5 5 3 4 3 4 5 4 4 4 4 Software prices are for industrial licenses as of 2/15/99. 4—Very Good. heating. Tracking of CIP materials Tracking of SIP materials Tracks denaturation of product Absense of software bugs Unit-operations modeling Fermentor Centrifuge Ultrafiltration/microfiltration In-line filtration Chromatography Precipitation Reactors Drying Homogenizer Absence of software bugs Scale-up algorithms Results/reports format and exportability Provides process operational outline Process-flow diagram (PFD) Material/energy-balance reports Materials database Equipment database File transferability Energy balances Utilities of interest in software database Allows detailed utilities specification Heat-duty calculations Accounts for equipment-power requirements Absence of software bugs Scheduling Provides detailed process schedule Integration of continuous and batch operations CIP tracking SIP tracking Labor constraints Chart appearance Scheduling of multiple batches Exportability User interface/ease of use Time for process set-up Organization of information Materials of interest in software database Physical properties of interest in software database Software cost/requirements Price Software requirements Run time Overall robustness of software Planned software enhancements Resources dedicated to software development Software technical support Economic evaluation Raw materials costs Consumables costs (filters. 3—Good. Rating of software for Kepner-Tregoe Decision Analysis. SuperPro $3895/yr/copya Batch Plus $9868/yr/copya 5 2 5 2 2 4 3 3 3 1 2 1 1 1 3 1 1 2 2 2 2 2 4 3 4 3 3 2 3 4 5 2 3 5 3 3 2 1 4 4 1 3 2 3 4 2 3 3 2 5 4 2 5 4 5 4 4 Material balances Tracks material balance across entire process Performs charge calculations Stores commonly used solutions Calculates stream physical properties of interest User can enter measured stream physical properties Calculates stream flow rates Tracks vaporization losses due to purges. Rating Scale. 2—Fair. membranes. capital. 5—Excellent.Table I. etc) Labor costs Equipment costs Investment analysis (return on investment. etc.

Results/Reports Both ABP and ISP software provide the user with materialand energy-balance reports. The ISP program uses a graphical interface to input a process-flow description. Several heat-transfer models are available ranging from a simple model using overall heat-transfer coefficients to a complex model incorporating fouling coefficients and jacket dimensions. Neither package accounts for constraints such as available labor or time limitations between cleaning and processing. Next. The software is able to account for operational.e. Both packages include material databases from which the user can select materials that will be used in the process. utility flow rates. while ISP sends reports to a text file. One vast improvement in unit-operation modeling would be to allow the user to enter their own model based on operating parameters the software already tracks. Energy Balances Simulation software can perform heat-duty calculations to determine the amount of heating or cooling required for the Table II.. The rating results were converted to a percentage of the total possible score for each category and these percentages are shown in Table II.8 0. 1981) is provided in Table II in the hope that it can be customized by the reader and used to make an educated decision about purchasing commercial bioprocess simulators. NO. which could very well be used as a starting point for a batch record. This weight is User Interface The ABP’s user interface is a text recipe similar to an operational procedure. Rating Category Material balances Unit operations modeling Results/reports format and exportability Energy balances Scheduling User interface/ease of use Software cost/requirements Economic evaluations Total Score for Scale-Up of Merck’s Vaccine Process SuperPro 45% 44% 53% 56% 35% 75% 83% 84% Batch Plus 58% 33% 63% 68% 58% 60% 60% 88% Weight 1 1 1 0. 4. Selecting a Software Package An example of a Kepner-Tregoe Decision Analysis (Kepner and Tregoe. and by allowing the results of several simulation scenarios to appear on the same report. but ABP provides more detailed scheduling information and better accounts for CIP and SIP scheduling. ABP calculates the heat duty of operations and uses this information to perform heat-transfer calculations. 72. These models allow ABP to predict heattransfer times. all decision outcomes being considered (i. The ISP software provides a very useful process-flow diagram. and exit-utility temperatures. the ABP text recipe provides a starting point for the development of batch records. completion of a given operation. ABP and ISP) are rated in key categories. Kepner-Tregoe Decision Analysis is a method for making informed decisions in a quantitative manner. As the software stands. The ISP-user interface is much more intuitive than that of ABP. FEBRUARY 20. and overhead costs.7 SuperPro 45% 44% 53% 50% 32% 60% 66% 59% 58% Score Batch Plus 58% 33% 63% 61% 52% 48% 48% 62% 60% 486 BIOTECHNOLOGY AND BIOENGINEERING.9 0.8 0. Evaluation of process simulation software using Kepner-Tregoe Decision Analysis. Both packages could better communicate results through the increased use of graphs. First. However. The ABP program sends these reports to either Microsoft Excel or Microsoft Word . The ABP program can be used to generate a process recipe in MSWord. the user is often left to manually export the results for plotting. Scheduling Both ABP and ISP perform process scheduling and provide the user with a scheduling Gantt chart. each category is assigned a weight depicting the category’s importance in the overall decision. Economics Economic analysis is a strength of both software packages. This rating has been performed and recorded in Table I. VOL. The process-flow diagram produced by ABP is generated in Visio and becomes unreadable if too many pieces of equipment are used. The user can point and click anywhere on the process-flow diagram (PFD) and gain information about that operation or process stream.kinetic reactor models in ISP and the in-line filtration model in ABP. which is less convenient. These material databases include preloaded components and their physical properties. 2001 . allowing the engineer to determine where cost reduction would be most significant. However. The ISP software calculates heat duties for most operations and tracks utility usage. equipment.9 0.

Case Studies Here we present case studies from the application of ABP and ISP to Merck’s vaccine-antigen process. The estimation of flocculent density represents the ability of simulation software to estimate some parameters. These case studies are intended to provide examples of how the software performs in industrial situations and have been selected not only to highlight areas where the software performs well but also areas where the software was unable to perform the desired simulation. These variableprocess parameters include alcohol concentration. Figure 1 compares the actual pilotscale process data to the ISP predictions of pellet weights. In reality. The resulting density was assumed to be constant over serotype and used for all simulations. The software would then use this value irrespective of the processing conditions such as cell debris concentration in the feed stream. which are otherwise unmeasurable. This model was used to simulate a pilot-scale. the presence of such a large quantity of flocculents may increase particle–particle interactions within the centrifuge and result in hindered settling (Steinour.truly subjective and should be modified to fit the decisionmaker’s situation. The results shown in Figure 1 indicate that when more complex unit-operation models are incorporated. serotype-dependent parameters included size of cell flocculents. the ABP software had a slightly higher overall score than ISP but it is unlikely this will be the case for all software users. The quantification of vapor emissions is critical to determining the environmental impact of a proposed process. they can provide a good estimate of process outputs over a wide range of operating conditions. These predictions are compared to the actual process losses in Figure 2. For our purposes. clarification centrifuge. represents unique processing conditions. feed stream viscosity. A head-tohead comparison with ABP could not be performed because the ABP centrifuge model does not incorporate a predictive model that takes into account process conditions. Comparison of actual pilot-scale centrifuge pellet weight and that predicted by SuperPro simulation. For example. Key personnel from all aspects of process development including both research and manufacturing were polled and the resulting weights have been entered into Table II. Hindered settling would produce a smaller pellet and is not accounted for in the ISP model. or residence time. The current version of ISP does not perform vapor-loss calculations. a value of 0. nitrogen purge rate. a variable for which a measured value was not available. This is one possible explanation for the discrepancy between actual and predicted pellet weight for serotype F. In our case. Case Study 1: Centrifugation The centrifugation model in ISP contains a predictive steady-state Stokes’ settling model. A discrepancy in the material balance around this operation was noticed and thought to be the result of organics being stripped from the process stream during the explosion-proof purge. changes in the centrifuge speed. Case Study 2: Vapor Emissions The ABP software package has the ability to perform vapor–liquid equilibrium calculations and predict vapor losses. We used ABP to model the vaporization loss of alcohol and water from a centrifugation step. We used ABP to predict the material that would be lost to the vapor-phase during processing. The next step is to multiply each category’s rating by its weight to arrive at a score for that category. Serotypedependent parameters were measured from the actual process and entered into the ABP simulation. Each bacterial serotype is produced individually and therefore. and nitrogen pressure. An explosion-proof purge (nitrogen purge) is invoked during this step due to the presence of a large quantity of alcohol.8 would be entered for the separation factor for cell debris. and centrifugeresidence time. differing in its processing conditions. The serotype F run had a large quantity of small flocculents that the ISP model predicts will completely settle during centrifugation. Serotype D has been marked with an asterisk to indicate that actual and simulation results were set equal to estimate flocculent density. current and would-be software users at Merck & Company were surveyed to identify which categories they felt held the most importance in our goal of scaling-up a vaccine-manufacturing process. The ISP program was used to take these variations into account and calculate the resulting pellet weight and composition. if pilot-scale data demonstrated that 80% of the cell debris entering the centrifuge was removed. Figure 1. All category scores were then summed to arrive at an overall score for both software packages. duration of the nitrogen purge. The ISP model predicted the centrifuge-pellet weights generated during the processing of seven different bacterial serotypes. Each serotype represents a unique bacterial fermentation and subsequent recovery. For centrifugation. Figure 1 shows a good correlation between the actual pellet weight and that predicted by ISP. The data in Figure 2 represent the total mass-balance COMMUNICATION TO THE EDITOR 487 . 1944).

protein is considered a contaminant and must be reduced to minimum levels during processing. 72. 4. discrepancy recorded for this step. Additionally. In most instances. In most cases we see the expected result— actual losses are greater than predicted losses because actual losses reflect both vaporization and transfer losses. Figure 4.Figure 2. To simulate scale-up. initializing models meant entering a reaction or separation yield for each component entering the unit operation. Simulated clearance studies showing current process-protein concentrations and simulated maximum-allowable concentrations. operating conditions that differed from the pilot-scale runs were updated to reflect the Case Study 5: Schedule Optimization Scheduling issues can also be investigated with simulation software. including material loss during transfer. To assess how well the software performs scale-up. the simulation was used to determine the optimum operating conditions that would minimize vapor losses while still providing protection against explosions. The ISP simulation is slightly closer to the manufacturing scale results after step 3 due to the more sophisticated centrifugation model in ISP which allows the degree of separation to change as a function of centrifuge speed. equipment specifications were changed to reflect the larger manufacturing scale equipment. VOL. For example. outputs are shown in Figure 3. The virtual-clearance study represents an ideal case. Then. NO. In this case study we investigate using simulation software to perform a “virtual” clearance study. The scaled-up simulations were then used to predict the manufacturingscale yields. a chromatography model might predict that 70% of inlet protein will be removed regardless of whether the feed stream contains trace quantities of protein or protein levels which would saturate the column. FEBRUARY 20. In the Merck vaccine process. Most ABP and ISP unit-operation models are much simpler and as a result neither package is able to fully predict scale-up. virtual clearance studies do provide a first guess as to the outer limits of the process and can be performed in either ABP or ISP. actual manufacturing procedures. These updated operating conditions include flow rates and centrifuge speeds changed due to pumping and equipment constraints. 488 BIOTECHNOLOGY AND BIOENGINEERING. Case Study 3: Scale-Up Yield Prediction Case studies 1 and 2 describe applications of the software’s more developed unit-operation models. Figure 4 compares the actual protein concentration after each step to the maximum allowable protein concentration as determined by ISP. predicted manufacturing yields are compared to actual process yields in Figure 3. Figure 3 shows that the manufacturing simulations do not do well at predicting the actual manufacturing yields. software unit operation models were initialized with pilot-scale data. Batch Plus-predicted vapor losses compared to actual process losses due to an explosion-proof purge during centrifugation. 2001 . Actual process-step yields for the manufacturing and pilotscale compared to SuperPro and Batch Plus predictions of manufacturingscale yields. Each serotype represents a unique bacterial fermentation and subsequent recovery. Manufacturing-simulation outputs fall close to the pilot-scale results. The results in Figure 2 are very useful in confirming that a material-balance discrepancy of the order of magnitude seen in the batch record could be caused by vapor emissions. because most software unitoperation models utilize a constant-removal rate and are not effected by the inlet-contaminant concentration. Also. differing in its processing conditions. We used ABP to investigate methods for reducing Figure 3. This is because the majority of unit-operation models simply use the yields set from pilot-scale data and are not a function of the equipment and operating parameters changed to reflect the manufacturing scale. However. Case Study 4: Virtual Clearance Studies Clearance studies determine the contaminant-removal rate required at each process step to achieve a final product that has contaminant levels below an allowable limit.

explore equipment/facility change. 1994. Bio/technol 6:200–203. Bioproc Eng 15:331–337. The new rational manager. and it was discovered that a section of piping shared between several unit operations was causing significant constraints in the processing time. Ind Eng Chem 36:618. Based on this quantitative evaluation it was found that the Aspen Batch Plus and Intelligen SuperPro software packages are well suited to perform basic material and energy balances. Tregoe BB. outs. Kepner CH. However. Process optimization by simulation. Chem Eng Prog 83(8):60–66. 1987. 1981. Titchener-Hooker NJ. Potera C. Dagot C. and perform economic analyses. Appl Biochem Biotechnol 34:93–104. The ISP scheduling program does not look at piping constraints and therefore. Gallier PW. The major advantage of simulation software is that all aspects of pro- Table III. Field RP. As seen in Table III. 1996. Kisala TP. Zacchi G. Bulmer M. Galbe M. New Jersey: Kepner-Tregoe Inc. Software users are encouraged to use a Kepner-Tregoe decision analysis when determining if a specific simulation software package will meet their needs. process time could be reduced by 20% by eliminating this piping constraint. Table III shows the current manufacturing-cycle time for a specific unit operation. Zhou YH. Computerbased simulation of the recovery of intracellular enzymes and its pilotscale verification. Pingaud H. Calandranis J. 840. Simulation of ethanol production processes based on enzymatic hydrolysis of lignocellulosic materials using Aspen Plus. CONCLUSIONS Commercially available process-simulation software was rated on its ability to aid in the development and management of an industrial biotechnology process. 1997. Biotechnol Bioeng 48:529–541. 29. Computer simulation systems. Biotechnol Bioeng 46:202–217. Bioprocess simulation: A new tool for process development. Evans LB. An advanced computing environment for modeling and design of integrated biochemical processes. 1995. cess management are incorporated into one format and can be investigated simultaneously. Chem Eng Progr 85(10): 50–61. Steinour HH. Computer-aided process analysis and economic evaluation for biosynthetic human insulin production—A case study. answer scheduling questions. Pascal F. Pons MN. Bioproc Eng 14:81–89. could not be used for this type of analysis. Pilot-scale verification of a computer-based simulation for the centrifugal recovery of biological particles. Scenario Current process Removal of equipment constraint Removal of piping constraint Process time 325 h 322 h 261 h COMMUNICATION TO THE EDITOR 489 . Corriou JP. Titchener-Hooker NJ. Sapidou E. 1992. 1998. 1988. Titchener-Hooker NJ. References Biegler LT. Aspen Batch Plus analysis of process scheduling constraints. Chemical process simulation. The ABP program was used to model the effects of purchasing an additional piece of equipment to reduce equipment constraints. This program was able to investigate piping usage. 1996. Comp Chem Eng 18:S621–S625. Bulmer M.processing times for a particular part of the process. Modeling of an industrial alcohol fermentation and simulation of the plant by a process simulator. Bioproc Eng 16:367–374. Petrides DP. 1989. Holwill ILH. 1944. This did not significantly reduce the cycle time. Dunnill P. Gen Eng News 18(16):10. the software was found to lack rigorous. Petrides D. 1995. BioPro Designer. Clarkson AI. Clarkson AI. A study of the use of computer simulations for the design of integrated downstream processes. predictive unit-operation models limiting their ability to accurately predict unit operation scale-up or to optimize operating conditions. Engasser JM.

Sign up to vote on this title
UsefulNot useful