1237_C06.fm Page 233 Wednesday, November 12, 2003 1 :02 PM
C. Software Programs to
Increase the Utility of
Predictive Microbiology
Information*
Mark Tamplin, Jozsef Baranyi, and Greg Paoli
CONTENTS
6.1 Introduction
6.2 Model Interfaces
6.2.1 Pathogen Modeling Program
6.2.2 Seafood Spoilage Predictor
6.2.3 Initiatives Leading to ComBase
6.3 Databases
6.4 Expert Systems
6.4.1 Types of Computer-Based Decision Support
6.4.2 Possible Applications of Expert Systems
6.4.3 Application to Qualitative Risk Assessment
6.5 Conclusions
References
6.1 INTRODUCTION
The advent of computer technology and associated advances in computational power
have made it possible to perform complex mathematical calculations that otherwise
would be too time-consuming for useful applications in predictive microbiology.
Computer software programs provide an interface between the underlying mathe-
matics and the user, allowing model inputs to be entered and estimates to be observed
through simplified graphical outputs. Examples of model software packages that
* Mention of brand or firm names does not constitute an endorsement by the U.S. Department of
Agriculture over others of a similar nature not mentioned.
2004 by Robin C. McKellar and Xuewen Lu
fi:":M ■
fllMl
1237_C06.fm Page 234 Wednesday, November 12, 2003 1:02 PM
have gained wide use in the food industry and research communities include the
Pathogen Modeling Program (PMP)* and the Food MicroModel (FMM).
Behind predictive software programs are the raw data upon which the models
are built. Access to these data has become important for validating the robustness
of models, for bringing transparency to microbial risk assessment, and for advancing
modeling techniques. Recent initiatives, such as the relational database, ComBase,
developed by the UK Institute of Food Research, Norwich, and US Department of
Agriculture-Agricultural Research Service are compiling tens of thousands of pre-
dictive microbiology data sets to describe the growth, survival, and inactivation of
microorganisms, and to accelerate model development and validation. Software
databases and spreadsheets have made it possible to organize large quantities of
data, and to search and retrieve specific items of information.
Along with advancements in databases and microbial modeling, a growing need
has developed for decision-support tools for navigating across large quantities of
data and retrieving specific information. Such information management systems have
been used in other scientific fields, but have not been adequately developed and
applied to predictive microbiology.
In this chapter, we present an overview of the role of commonly used software
applications in the field of predictive microbiology, and we provide examples of
software use in model interfaces, relational databases, and expert systems. Continued
development and application of software programs in this field will improve the
tools that are available to researchers and risk managers for enhancing the safety
and quality of the food supply.
6.2 MODEL INTERFACES
Software programs have markedly enhanced the use of microbial models by the
food industry, risk assessors, and food microbiologists. Well-designed interfaces with
intuitive features allow users to define parameter inputs and then easily observe
model outputs in graphic formats. This section describes more widely used predictive
microbiology software with demonstrated applications in food safety and quality.
6.2.1 Pathogen Modeling Program
The PMP is a free software package of microbial models that describes growth,
survival, inactivation, and toxin production under various conditions defined by the
user (Buchanan, 1993). The current version, 6.1, contains 37 models for 10 bacterial
pathogens that predict their growth and thermal and nonthermal inactivation. In
addition, the PMP contains dynamic temperature models for the growth of Clostrid-
ium perfringens and Clostridium botulinum. Such dynamic models are increasingly
sought by food industries that must meet performance standards for cooked-cooled
meat products. Depending on the specific model, environmental variable inputs
include atmosphere (aerobic or anaerobic), temperature (°F or °C), pH, water activity,
* U.S. Department of Agriculture- Agricultural Research Service Pathogen Modeling Program
(www.arserrc.gov/mfs/pathogen.htm).
2004 by Robin C. McKellar and Xuewen Lu
1237_C06.fm Page 235 Wednesday, November 12, 2003 1:02 PM
ionizing radiation, varying concentrations of lactic acid, sodium chloride, nitrite,
and sodium pyrophosphate, or all of these. Model outputs for lag phase duration,
generation time, and time to a user-defined level of interest are displayed in either
hours or days. In addition, growth/inactivation curves are displayed in both graphical
and tabular formats, along with associated upper and lower confidence limits. New
features in the 6.1 version include links between specific models and associated
publications, separate printing of output tables and figures, and input of time-tem-
perature data sets in comma-separated-variable (csv) format. Future versions of the
PMP will include direct links to the data sets underlying individual models, optional
outputs for lag and no-lag scenarios, and operating the PMP on-line using
Microsoft.Net database technology.
Similar to studies found in the scientific literature, the majority of the PMP 6.1
models were developed in defined microbiological media. Consequently, the PMP
informs the user that there can be no guarantee that the predicted values will match
those that would occur in a specific food system. To make the PMP models more
useful to the food industry, as well as to food microbiologists and risk assessors,
more food-specific models are under development. Examples of PMP 6.1 models
include dynamic temperature for C. perfringens in cured beef and cured chicken
(Juneja et al., 2001; Juneja and Marks, 2002), and growth models for Salmonella
typhimurium on chicken breast meat (Oscar, 1999) with inputs for previous growth
at variable temperatures and NaCl concentrations.
For the majority of the PMP models, bacterial growth and survival are repre-
sented by the Gompertz model. Although this approach is normally satisfactory for
simple sigmoidal curves, the Gompertz model lacks the desired flexibility for mod-
eling dynamic conditions that are relevant to commercial food production, particu-
larly for thermally processed food. To meet this demand, future versions of the PMP
will incorporate the dynamic model described by Baranyi and Roberts (1994).
6.2.2 Seafood Spoilage Predictor
The Seafood Spoilage Predictor* (SSP), produced by the Danish Institute of Fish-
eries Research, is a predictive microbiology software package for the microbial
spoilage of fisheries products. It is more versatile than other similar programs by
being able to predict food spoilage with both fixed and fluctuating temperatures.
Also, the SSP has two model forms, relative rate of spoilage and microbial spoilage,
which estimate spoilage as a function of organoleptic change and change in microbial
levels, respectively.
6.2.3 Initiatives Leading to ComBase
The UK Institute of Food Research has produced curve-fitting softwares** that are
increasingly used by researchers to develop predictive models. The PC-based
Microfit and DMFit programs both use the model described by Baranyi and Roberts
(1994) to fit curves to time versus colony-forming unit bacterial growth data. Microfit
* Danish Institute of Fisheries Research (http://www.dfu.min.dk/micro/ssp/).
** Institute of Food Research (www.ifr.bbsrc.ac.uk).
2004 by Robin C. McKellar and Xuewen Lu
1237_C06.fm Page 236 Wednesday, November 12, 2003 1:02 PM
allows the user to compare the specific growth rates of different bacterial growth
curves and to measure statistical significance. The DMFit program is an Excel®-
based add-in feature that provides different types of curve-fitting programs and it
plots the fit and the data for each data set. The DMFit software is an advantageous
tool when the user anticipates incorporating the data sets into the ComBase relational
database (described in the next section) because of similar data input formats.
The Food MicroModel (FMM)* was a commercial package of models with a
graphical user interface, similar to the PMP (McClure et al., 1994); however, it has
been discontinued in its original form. The FMM received extensive use by the
food industry and offered users simple formats for inputting parameter values and
obtaining model output data. It contained microbiological broth and food-specific
models for pathogens and spoilage bacteria. Because of the commercial nature of
the FMM, the user had limited access to its underlying models and data sets.
Recently, the data behind the software have been merged with those that were
generated for the PMP as well as data from international collaborators and from
the published literature. The combined database (ComBase) is freely available at
http://wyndmoor.arserrc.gov/combase/. The intention is to combine data and pre-
dictions in a unified Web-based database and model package called ComBase-PMP.
6.3 DATABASES
Predictive microbiology software packages are based on two main pillars: databases
and mathematical models, and the scientific study of these belongs to both Bioin-
formatics and Biomathematics. Most chapters of this book deal with mathematical
models; however, the present chapter is primarily dedicated to the other two members
of this triangle (Figure 6.1).
The mutual dependence between mathematical models and databases is also
confirmed by the fact that the ultimate tests for predictions are comparisons with
observations. This can be done quickly and efficiently on large amounts of data only
if the data-recording format is strictly standardized and harmonized with the respec-
tive mathematical variables.
Predictive software
Biomathematics
(mathematical models)
Bioinformatics
(database)
FIGURE 6.1 Relationship between predictive software, biomathematics, and bioinformatics.
* Leatherhead Food Research Association, UK (www.leatherheadfood.com/lfl/index.htm).
2004 by Robin C. McKellar and Xuewen Lu
1237_C06.fm Page 237 Wednesday, November 12, 2003 1:02 PM
A database is a large collection of data organized especially for rapid search
and retrieval. As an example, a typical, everyday-used database is a personal address
book. Its computer form can be conceived as a table of rows and columns, where
the rows and columns are termed records and fields, respectively. In an address book,
typical fields are the name, address, and phone number. Fields belonging together
(in our example, belonging to one person) form a record.
When defining the fields, one must define the "resolution" of the database
according to the tasks for which it is used. A field, such as the address in our example,
could be further divided into town and street, thus increasing the resolution of the
record. However, this may cause problems when such an address as "Oak House,
Kingston" is to be recorded, with no street name. Therefore, if there is no special
reason for further resolution, it is unnecessary to subdivide the field. A valid reason
for further refinement could be if it was a frequent task to list those persons who
live on the same street.
Similar problems are encountered when building a database on microbial
responses to the environment. For the benefit of a fast data search, those elements
of information that are likely to be searched should be recorded in specified fields,
in a specified format. Therefore, the first step when building such a database is to
define the fields and their syntax, i.e., what values can be in a field. Typically, these
values are either numeric (for quantities) or alphabetic (for categories). It is common
to provide an interval for the values found in the numeric fields; similarly a list of
values can be used to define the content of a category field.
Because of the complexity of the food environment and the tedious nature of
food microbiology measurements, predictive food microbiology is typically a sci-
entific field where well-organized databases are in high demand. For example:
1. The inaccuracy of measurements can be compensated by numerous
measurements.
2. The variability of responses is a main focus of investigation, and quanti-
fication and modeling of variability require replicate measurements.
3. International data exchange cannot be done without defining compatible
database formats.
Accordingly, the first question when building a database is: What to record? In this
respect, the construction of a database is similar to that of a mathematical model;
both are, in some way, the art of omitting the unnecessary. One cannot record all
available information on microbiology experiments; some simplification and cate-
gorization is inevitable. The way these simplifications are carried out is sometimes
arguable, even subjective, but undoubtedly necessary, just as mathematical models
are necessarily simplified descriptions of nature, to understand and predict complex
phenomena. The distribution of information among database categories (fields) is
parallel to assigning mathematical variables to certain quantities; the relationship
between those fields, for fast search and interrogation, is parallel to relationships
and equations between mathematical variables for deriving conclusions and predic-
tions in a mathematical way.
2004 by Robin C. McKellar and Xuewen Lu
1237_C06.fm Page 238 Wednesday, November 12, 2003 1:02 PM
In general, producing software of microbiology data commonly involves the
following stages of development:
Level 1 — Raw data: these are as recorded by the person generating the data,
usually in spreadsheet format.
Level 2 — Database: these are sets of data, systematically categorized and
recorded, following a predefined structure (syntax). To create a database,
one needs computational skill, understanding of the data, and the ability to
make expert judgments when necessary.
Level 3 — Browser: a computer program to navigate in the database.
Level 4 — Simple predictions: these are usually interpolated values given by
mathematical equations, for example, the FMM or the PMP is at this level.
Level 5 — Complex predictions (dynamic conditions, growth/no growth
boundary, probability of growth, Bayesian methods, expert systems): at the
moment, no predictive software package is available at level 5.
A new predictive microbiology initiative called ComBase is described by Baranyi
and Tamplin (2002). It aims at pooling data in a common database for predictive
microbiology purposes. The database will be a source of publicly available infor-
mation on bacterial responses to food environments.
One of the most important features of ComBase is that it can record not only a
single value in a field but also a "pointer" to a table, representing the variable
changing with time. A dynamic profile can be recorded this way, as a list of (time,
value) pairs, for either an environmental factor or a response variable. ComBase is
available via the Internet.
6.4 EXPERT SYSTEMS
The increased use of computer technology in food production, the widespread use
of the Internet in business, and the increasing ease and speed of development of
user-friendly software opens up a number of possibilities for software solutions in
supporting decisions that are based on predictive microbiology. This section
describes a number of ways in which decision-support software can contribute. We
focus on the employment of a particular class of software programs called expert
systems. These tools formally encapsulate knowledge and data and, given input
through a user-interface, generate conclusions and analyses to inform the user.
6.4.1 Types of Computer-Based Decision Support
The simplest way in which computers can support decisions is to provide for
intelligent storage and retrieval of data. This storage and retrieval can be easily
extended to add documentation retrieval. More advanced decision-support systems
can perform complex calculations given user input and then present results in
appropriate formats.
Beyond data storage, data retrieval and the implementation of calculations is a
class of tools that can bring a higher level of functionality and knowledge to decision
2004 by Robin C. McKellar and Xuewen Lu
1237_C06.fm Page 239 Wednesday, November 12, 2003 1:02 PM
support (Level 5 described above). These tools can bring more domain-specific
knowledge to the decision-support function by including more complex knowledge,
relationships among pieces of knowledge, and can impose a knowledge-based struc-
ture on the decision process.
Earlier in this chapter we discussed large predictive microbiology databases
(e.g., ComBase) and model interfaces (e.g., PMP) that are derived from these data.
Decision-support tools can be used to navigate across vast quantities of predictive
microbiology information, retrieve specific information, and then to build and
validate microbial models. Furthermore, decision-support tools can potentially
direct the user to alternative sources of information when appropriate models and
data are lacking.
An example of more complex decision support is the use of expert systems.
While a number of software tools whose functionality is derived from expert knowl-
edge may be called expert systems, the term is most appropriately applied to software
that provides advisory information based on a set of rules and algorithms for infer-
ence based on the relationships underlying these rules.
In these systems, the core knowledge is stored as a series of IF-THEN rules
that connect diverse evidence such as user input, data from databases, and the
formalized opinions of experts into a web of knowledge. Software tools are available
to facilitate the development of these rule-based systems, including tools to handle
the user interface and creation of an executable, stand-alone software product.
Recently, the capacity to create Internet-ready expert systems has been developed
and integrated into these development tools to greatly facilitate the dissemination
of expert knowledge in terms of both the development time and the ease of its
delivery to users.
There are various examples of decision-support systems which encapsulate
knowledge that is based in predictive microbiology. Examples include decision-
support tools to be applied in predicting food safety and shelf life (Wijtes et al.,
1998; Zwietering et al., 1992), and a step-wise system structured as a standard risk
assessment process to assist in decisions regarding microbiological food safety (van
Gerwen et al., 2000). Other examples of decision-support systems for microbial
processes are described by Voyer and McKellar (1993) and Schellekens et al. (1994).
Expert systems provide two main forms of decision support. They impose a
structure on the inference such that there is a consistent and deterministic pathway
between user input, data, and conclusions. This is useful where the decision domain
benefits from the assurance of consistency and where there is a desire to remove more
subjective elements of decision-making from the overall decision process. For orga-
nizations dealing with recurring and complex food safety issues, consistency and
reliability of decision-making may provide assurance to buyers, regulators, and the
organization itself that safety will not be compromised by unstructured or ill-informed
decision-making. In addition to concerns for subjective and variable decision-making,
the proper structuring of the decision (in terms of delineating and weighing the
necessary considerations and accessing the required knowledge to support the deci-
sion) may itself be the most difficult task for nonexpert users to carry out.
A second form of decision support provided by expert systems is to capture and
make available the expert knowledge itself. The result of the conversion of expert
2004 by Robin C. McKellar and Xuewen Lu
1237_C06.fm Page 240 Wednesday, November 12, 2003 1:02 PM
knowledge into a set of rules that can generate advice in the absence of direct expert
interaction could be a valuable asset for many institutions. Such tools may be
particularly useful when provided to smaller organizations that do not have ready
access to scientific expertise. The expertise that may be held by relatively few persons
can be made available to many users who might otherwise compete for the attention
of a few experts. In addition, time spent addressing repetitive enquiries can detract
from the expert's own efforts in maintaining and enhancing his knowledge base,
thus diminishing the overall generation of knowledge in an organization. With respect
to food safety, a key ingredient in the process is the knowledge and decision processes
that must be applied to assure a safe product. An expert system could provide
assurance of the supply and the quality of this knowledge ingredient.
6.4.2 Possible Applications of Expert Systems
A number of possibilities exist for the development of decision-support systems
based on rule-based expert systems. Examples (Paoli, 2001) include:
Virtual inspection: Simulating interaction with an inspector to allow estab-
lishments to self-assess their facility or a food processing operation. This
may be of interest to large companies and regulators who find their quality
control or inspection resources inadequate for the number of establishments
that they are required to assess.
Process deviation assessment: Tools which incorporate expert knowledge
regarding the best actions to take in case of process deviations, in terms of
assessing the seriousness of the deviation and in recommending corrective
actions for the implicated product and the process.
Problem-oriented education: Apart from providing expert advice, an expert
system could be used to assist in education regarding the importance of key
variables and to foster careful reasoning among operational decision-makers.
In-line real-time expert systems: Examples exist (though not in predictive
microbiology) of expert systems that receive real-time data and provide
continuous assessments of the status of systems, based on the combination
of these data and embedded knowledge-based rules that interpret the data
for display to operators. This could be applied to food production systems
where a complex set of variables requires monitoring combined with com-
plex reasoning to assure safety and quality.
6.4.3 Application to Qualitative Risk Assessment
Of international interest may be applications in the area of qualitative risk assess-
ment. While most of the recent attention in the field of microbiological risk assess-
ment has been paid to quantitative risk assessment (see Chapter 8), there is increasing
interest in the relative merits and drawbacks of a more qualitative form of risk
assessment. The merits of such an approach include the speed with which such an
analysis could be performed and the decreased reliance on quantitative data which
is often unavailable. Some drawbacks include the potential for qualitative risk
2004 by Robin C. McKellar and Xuewen Lu
1237_C06.fm Page 241 Wednesday, November 12, 2003 1:02 PM
assessment to become little more than a literature search with loose and unstructured
reasoning and untraceable conclusions.
Currently, there are no formal guidelines specifying what type of analysis would
constitute a properly performed qualitative risk assessment. Were such guidelines
to be developed, an expert system could be produced which encapsulated and
imposed an appropriate risk assessment process, as well as some of the requisite
knowledge that would be required to assure an adequate qualitative risk assessment.
Recent work by Crawford-Brown (2001) is an example of a decision-support tool
to assist in the development of a highly structured and well-documented qualitative
risk assessment. Further examples of this type of tool might be of great benefit to
those who would like to capture the potential value of systematic risk assessment
while avoiding the computational and data burden of quantitative approaches.
6.5 CONCLUSIONS
On the basis of the trends seen for the past ten years it can be said that the demand
for predictive microbiology software programs will expand with increasing applica-
tion of microbial models to food systems. This includes the use of predictive models
for research and for the development of Hazard Analysis and Critical Control Points
systems, product formulation, and risk assessment. Concomitantly, continued
advances in database and decision-support software, as well as Internet technologies,
will provide risk managers with better tools for seeking relevant information and
making informed decisions. However, to sustain this effort, multiinstitutional collab-
orations will be critical for managing the input, organization, quality control, and
dissemination of large quantities of predictive microbiology data. Ultimately, these
advances will provide researchers, students, and educators with greater access to
information for improving the safety and quality of the food supply.
REFERENCES
Baranyi, J. and Roberts, T. A. 1994. A dynamic approach to predicting bacterial growth in
food. Int. J. Food Microbiol. 23:277-294.
Baranyi, J. and Tamplin, M.L. 2002. ComBase: a combined database on microbial responses
to food environments. 1st International Conference on Microbial Risk Assessment:
Foodborne Hazards. College Park, MD, p. 23.
Buchanan, R.L. 1993. Developing and distributing user- friendly application software. /. Ind.
Microbiol. 12:251-255.
Crawford-Brown, D. 2001. Expert system for waterborne pathogens. Presented at workshop,
"Decision-Support Tools for Microbial Risk Assessment" at the Society for Risk
Analysis Annual Meeting, Seattle, WA.
Juneja, V.K., Novak, J.S., Marks, H.M., and Gombas, D.E. 2001. Growth of Clostridium
perfringens from spore inocula in cooked cured beef: development of a predictive
model. Innovative Food Sci. Emerging Technol. 2:289-301.
Juneja, V.K. and Marks, H.M. 2002. Predictive model for growth of Clostridium perfringens
during cooling of cooked cured chicken. Food Microbiol. 19:313-3297.
2004 by Robin C. McKellar and Xuewen Lu
1237_C06.fm Page 242 Wednesday, November 12, 2003 1:02 PM
McClure, P.J., Blackburn, C. de W., Cole, M.B., Curtis, P.S., Jones, J.E., Legan, J.D., Ogden,
I.D., Peck, M.W., Roberts, T.A., Sutherland, J.P, and Walker, S.J. 1994. Modelling
the growth, survival and death of microorganisms in foods: the UK Food Micromodel
approach. Int. J. Food Microbiol. 23:265-275.
Oscar, T.P. 1999. Response surface models for effects of temperature, pH, and previous growth
pH on growth kinetics of Salmonella typhimurium in brain heart infusion broth.
J.FoodProt. 62:106-111.
Paoli, G. 2001. Prospects for expert system support in the application of predictive microbi-
ology. Presented at workshop, "Decision-Support Tools for Microbial Risk Assess-
ment" at the Society for Risk Analysis Annual Meeting, Seattle, WA.
Schellekens, M., Martens, T., Roberts, T.A., Mackey, B.M., Nicolai, M.B., Van Impe, J.F.,
and De Baerdemaeker, J. 1994. Computer-aided microbial safety design of food
processes. Int. J. Food Microbiol. 24:1-9.
van Gerwen, S.J.C., te Giffel, M.C., Van't Riet, K., Beumer, R.R, and Zwietering, M.H. 2000.
Stepwise quantitative risk assessment as a tool for characterization of microbiological
safety. J. Appl. Microbiol. 88:938-951.
Voyer, R. and McKellar, R.C. 1993. MKES tools: a microbial kinetics expert system for
developing and assessing food production systems. J. Ind. Microbiol. 12:256-262.
Wijtes, T., Van't Riet, K., in't Veld , J.H.J., and Zwietering, M.H. 1998. A decision support
system for the prediction of microbial food safety and food quality. Int. J. Food
Microbiol. 42:79-90.
Zwietering, M.H., Wijtzes, T., Wit, J.C. de, and Van't Riet, K. 1992. A decision support system
for prediction of the microbial spoilage in foods. J. Food Prot. 55:973-979.
2004 by Robin C. McKellar and Xuewen Lu
;