SciELO - Scientific Electronic Library Online

vol.45 número3Ajuste de modelos de regresión aleatoria en evaluaciones genéticas de bovinos TropicarneColémbolos de suelos agrícolas en cultivos de alfalfa y de maíz adicionados con biosólidos en Aguascalientes, México índice de autoresíndice de materiabúsqueda de artículos
Home Pagelista alfabética de revistas  

Servicios Personalizados




Links relacionados

  • No hay artículos similaresSimilares en SciELO



versión On-line ISSN 2521-9766versión impresa ISSN 1405-3195

Agrociencia vol.45 no.3 Texcoco abr./may. 2011


Matemáticas aplicadas, estadística y computación


Model evaluation of quality attributes for hops (Humulus lupulus L.)


Evaluación de modelos de los atributos de calidad para los lúpulos (Humulus lupulus L.)


Martin Pavlovic1,2, Viljem Pavlovic3


1 International Hop Growers' Convention IHGC-Secretary General, 22, rue des Roses, F-67173 Brumath, France. ( * Author for correspondence.

2 Slovenian Institute of Hop Research and Brewing, Zalskega tabora 2, 3310 Zalec, Slovenia.

3 University of Maribor, Faculty of Agriculture and Life Sciences, Pivola 10, 2311 Hoce, Slovenia.


Received: July, 2010.
Approved: March, 2011.



Hops (Humulus lupulus L.) ate vital fot the brewing industry, as they contribute significantly to the organoleptic qualities of beet, including taste and flavor. Experimental hop breeding data from the Slovenian Institute of Hop Research and Brewing (IHPS) were used to create a model based on the multi-attribute decision modeling methodology. The model has 18 attributes hierarchically grouped within four main attributes: Biology, Chemistry, Morphology and Brewing value. Furthermore, utility functions in the model were defined by sets of elementary decision rules throughout the entire hierarchy for all aggregated attributes. The central part of the model contains 144 decision rules, which were specified according to the model users' previous breeding experiences. Four prospective hop hybrids and a reference hop variety with the target characteristics of plant resistance and brewing value were evaluated. Based on the breeding experiences and model results attributes' assessments were carried out. Decisions based on the model evaluation offered an additional tool for experts' final decisions in selecting appropriate materials for further breeding or the commercial use of hop plants.

Key words: hop industry, breeding, decision-making, model, IHPS.



Los lúpulos (Humulus lupulus L.) son fundamentales para la industria cervecera, ya que contribuyen significativamente a las cualidades organolépticas de la cerveza, como el gusto y el sabor. Los datos de mejoramiento experimental de lúpulos del Instituto Esloveno de Investigación de Lúpulos y Elaboración de Cerveza (IHPS) se usaron para crear un modelo basado en la metodología de modelado de decisiones de atributos múltiples. El modelo cuenta con 18 atributos agrupados jerárquicamente dentro de cuatro atributos principales: Biología, Química, Morfología y el Valor de elaboración de la cerveza. Además, las funciones de utilidad en el modelo se definieron por conjuntos de reglas de decisión primaria en la jerarquía completa de todos los atributos agregados. La parte central del modelo contiene 144 reglas de decisión, especificadas de acuerdo a las experiencias previas de mejoramiento de los usuarios del modelo. Se evaluaron cuatro posibles híbridos de lúpulo y una variedad de lúpulo de referencia con las características previstas de resistencia de las plantas y el valor de elaboración. Basándose en las experiencias de mejoramiento y los resultados del modelo, se realizaron las evaluaciones de los atributos. Las decisiones basadas en la evaluación del modelo proporcionaron una herramienta adicional para las decisiones finales de los expertos en la selección de materiales apropiados para mejoramiento futuro o el uso comercial de plantas de lúpulo.

Palabras clave: industria del lúpulo, mejoramiento, toma de decisiones, modelo, IHPS.



Brewing industry requires raw materials of a high quality and hops (Humulus lupulus L.) are vital, as they contribute significantly to the organoleptic qualities of beer, including taste and flavor. To remain competitive in the global hop industry, hop breeders must respond to the ever-changing needs of the brewing community by providing appropriate new varieties. The aim of hop breeding is to satisfy the needs throughout the whole chain of the hop industry. The target characteristics of new varieties include suitable brewing value (amount of alpha-acids, good flavor), high-quality crops and resistance to pests, as well as resistance to diseases. The main hop consumer is the brewing industry, which needs stable amounts of high-quality hops of desired varieties and origin, with constant quality (Pavlovic et al., 2008).

During the last few decades there has been sustained interest in the ability to identify or assess hop varieties or potential breeding materials with chemical analysis of the essential oils and resin components, the composition of which is genetically controlled (Lemmens, 1998). However, expert decisions for further selection of hybrids require the synthesis of information from different kinds of data-subjective (assessments in the field) and objective (chemical analysis). To improve hop breeding results, various decision support methods are available for application. Javornik et al. (2005) report about molecular approaches, which can support breeders' decisions in early crossing combinations. Henning and Towsend (2005) use a similar approach, while De Keukleire (1999) suggest classical chemical analysis to choose the best new variety. Furthermore, the use of methods within the expert system in hop breeding has been discussed (Pavlovic et al., 2007). Recently, multi-criteria decision analysis (MCDA) has been recognized in hop hybrids assessment as a promising method. The MCDA approach is able to synthesize information about many different and often conflicting attributes into one unified assessment (Rozman et al., 2006) that can be numerical or discrete (Pazek et al., 2006). To date, MCDA has been widely used in agriculture, as described by Znidarsic et al. (2006), Scatasta et al. (2007) and Pazek and Rozman (2007).

The DEX method (Bohanec, 2008), which is employed in this study, uses qualitative variables and utility functions in the form of decision rules, and provides qualitative assessments of alternatives. Recently, DEX has been used in agronomy to assess the impact of cropping systems on soil quality (Bohanec et al., 2007) and the economic and ecological impact of using genetically modified crops (Bohanec et al., 2008).

This research belongs to an international project called Automation Agents in Decision Support Systems (Pavlovic et al., 2008). In this paper, a multi-attribute decision model based on DEX methodology for hop hybrids assessment applied to four promising hop hybrids (appropriate hybrids for further breeding) is presented also in a hop industry case study.


Material and Methods

Plant material

Within the hop breeding research program carried out at the Slovenian Institute of Hop Research and Brewing (IHPS), various hop hybrids appeared to have good prospects for further breeding according to the research objectives in terms of brewing and agricultural value (Cerenak, 2006).

In this research, the data from the four most promising Slovenian hop hybrids, A1/54, A2/104, A3/112, and A4/122, were compared with the reference variety, which had the desired characteristics for plant resistance and brewing value. The assessment was carried out with a qualitative multi-attribute model based on the DEX methodology.

DEX methodology

The assessment was carried out with a qualitative multi-attribute model based on the DEX methodology (Bohanec et al., 2000). We first developed the model and then applied it to assess the aforementioned hop hybrids. DEX combines traditional multi-attribute decision-making with some elements of Expert Systems and Machine Learning. The distinguishing characteristic of DEX is its capability to deal with qualitative variables. Instead of numerical variables, which typically constitute traditional quantitative MCDA models, DEX uses qualitative variables whose values are usually represented by words rather than numbers, such as UNACCEPTABLE, APPROPRIATE, HIGH, PROSPECTIVE, etc. Furthermore, DEX uses "if-then" decision rules to evaluate decision alternatives. The method is supported by the software program DEXi (Bohanec, 2008), which has been applied to various real-life decision problems (Znidarsic and Bohanec, 2007). DEX models are developed through the following three steps:

1) The problem is decomposed into less complex individual problems; the problems are hierarchically structured into a tree of attributes that represents the "skeleton" of the model. Terminal nodes of the tree, i.e., leaves or basic attributes, represent inputs to the model, and the root node represents the main output: an overall assessment of the evaluated alternatives (hop hybrids in our case). The internal nodes of the model are called aggregate attributes.

2) Each sub-problem is represented by a qualitative attribute with a defined value domain with a set of values. The value domain is discrete and typically consists of words. Usually, the domain is also ordered preferentially, so that consecutive words denote more and more desirable characteristics of the corresponding attributes.

3) To define the aggregation of values from the input to output attributes of the model, utility functions are defined for each aggregate attribute in the model. In the DEX method the utility functions are represented by decision rules, which are typically formulated by decision makers or domain experts.

Model for hop hybrids evaluation

The model for hop hybrids' evaluation takes into account four main factors: the Biology, Chemistry, Morphology and Brewing value of the hop hybrids. This is reflected in the hierarchical structure of the model, which consists of four corresponding sub-trees of attributes (Figure 1). Based on the breeding team's experience, sets of discrete values were defined for all 18 attributes in the model (Table 1).


The aggregate attribute Biology includes two basic attributes: (i) plant resistance, and (ii) plant outlook. Plant resistance is related to two main hop diseases, downy mildew on hops (Pseudoperonospora humuli) and powdery mildew on hops (Sphaerotheca humuli), and it was evaluated in a field collection of hop breeding material. Each plant was assessed based on the occurrence of infected tissue of the two main hop diseases from 0 to 2, where 0 indicates a resistant plant, 1 a moderately susceptible plant and 2 a susceptible plant (Seigner et al., 2005, Radisek et al., 2007). Since the reference variety was of low resistance against plant diseases, the mark BAD RESISTANT was given to samples with the same level of susceptibility within a tolerance of 15 %. Hybrids with a lower level of damaged plants were labeled RESISTANT, while those with a higher level of damaged plants were SUSCEPTIBLE.

The attribute plant outlook includes properties such as plant vigor, i.e., branching out, the technological ripeness, the length and the tightness of the hop cones. These properties were measured regularly according to a research plan methodology (Cerenak, 2006). The model enables three attribute values: BAD, SUITABLE and GOOD.


The aggregate attribute Chemistry consists of three attributes: (i) essential oils, (ii) ageing and (iii) bitterness. Bitterness is assessed through the relative content of alpha-acids, beta-acids in dry matter and the percentage of cohumulone in alpha-acids. For each attribute a set of numeric values was defined again based on chemical analysis, in which the following methods were used: humidity content, LCV (lead conductance value in hops), percentage of xanthohumol (Analytica EBC, 1998) and the HSI (ASBC, 1992). The attribute ageing, which determines the loss of alpha acid content defined by the HSI, depends on temperature. Thus, measurements were taken when the hops were stored at 4 °C and 20 °C. Individual hybrid assessment results were compared with the results of the reference hop variety. According to the DEX methodology, quantitative measures were converted into qualitative values. Hybrids were assessed as; WORSE if the attribute values were lower than 80 % of those from the reference variety; GOOD if they ranged between 80 % and 95 %; REFERENCE if they ranged between 95 % and 105 %; BETTER if they were higher than 105 %.


Attributes within the aggregate attribute Morphology: (i) spindle share, (ii) spindle length, (iii) weight of cones, (iv) density of cones and (v) seeds, illustrate the characteristics of hop cones (female blossoms) that represent the main plant product. These characteristics affect the hop production technology procedures and consequently hop yield quality and amount. Within the evaluation of these morphological characteristics, hop hybrids were evaluated after a harvest according to the approved research plan with the use of the domain system from 0 to 5 (Cerenak, 2006). In the model, marks 5 and 4 took a discrete value GOOD, mark 3 from the assessment was defined as ACCEPTABLE, while marks 2, 1 and 0 received the value BAD.

Brewing value

This attribute was determined with chemical analysis of wort and beer brewed with the selected hop hybrids and a reference variety. The results of the chemical analysis were the starting point for the assessment. After the hopping, all types of wort were analyzed with standard analysis protocols. In wort, the contents of the extract (%), alcohol (% v/v), P.I. value, color EBC, bitterness, alpha-acids (%), iso-alpha-acids (mg L-1), polyphenols (mg L-1) and anthocyanogens (mg L-1) were measured (Analytica EBC, 1998; MEBAK, 2002).

In a small experimental microbrewery, samples of the hop hybrids were used for brewing. The Brewing value of the beer samples brewed from those hop hybrids was based on a panel sensory assessment by 22 beer experts. Standard parameters such as the beer's taste, hop aroma quality, hop aroma intensity, bitterness quality and bitterness intensity of the beer samples brewed with the selected hop hybrids and the reference variety were examined according to DLG tests (MEBAK, 2002). The experimental beer tasting took place at the Research Institute of Brewing and Malting PLC, Prague, Czech Republic. The average mark from the sensory assessment panel was included in the model. The quality domain embraced three levels: BAD, GOOD and EXCELLENT. High priority was given to this attribute since it was determined to be an eliminating one. Namely, if the sensory assessments did not meet the brewing experts' minimal expectations, the hybrid was considered inappropriate for further hop breeding.

Utility functions

Utility functions are components of multi-attribute models that define the aggregation aspect of option evaluation. For each aggregate attribute y, whose descendants in the tree of attributes are x1 , x2 , ..., xn , the corresponding utility function f defines the mapping:

ƒ: X1 x X2 x... x XnY

where X1,..., Xn and Y denote value domains of the attributes x1, ..., xn and y.

In DEX (Bohanec, 2008), the value domains are discrete; therefore, the function f maps all the combinations of values X = X1 x X2 x ... x Xn into the values of Y The mapping is represented in a table, where each row gives the value of y for one combination of values x £ X. Rows are also called decision rules, because each row can be interpreted as an "if-then" rule of the form:

if x1=v1 and x2=v2 and ... and xn=vn then y=v

where v1 ∈ X1, ..., vn∈ and v ∈ Y.

The rules define the mapping of four attributes Biology, Chemistry, Morphology and Brewing value into the overall Hop hybrid assessment for one combination of values of the four attributes. Each decision rule defines the value of the Hop hybrid assessment for one combination of values of the former four attributes. Since Biology and Chemistry can take four different values, and Morphology and Brewing value can take three values (Table 1), there are 4x4x3x3=144 possible combinations; hence, there are 144 decision rules.

Table 2 shows the weights that were defined by the breeding and brewing experts for each corresponding attribute. The two main aggregate attributes, Chemistry and Brewing value, have the most important role in the hybrids assessment. Chemistry determines the type of hop variety (aroma, alpha or super-alpha type), while Brewing value includes an elimination characteristic; consequently, they were assigned relatively high weights, 35 % and 40 %.

Let us illustrate this method using the utility function that maps plant resistance (x1) and plant outlook (x2) to Biology (y). In this case, the whole table (Figure 2, bottom right) contains nine rules. Only three of these (rules 1, 8 and 9) were explicitly defined by the expert (this is indicated by the bold values in the Biology column). These three rules constitute the initial set S. In addition to these three rules, the user also indicated (Figure 2, bottom left) that the weights of the attributes plant resistance and plant outlook should be w1 = 86 % and w2 = 14 %.

The decision rules for the final hop hybrid assessment were defined according to the model users' previous breeding experiences. The rules were determined taking into account the minimum and maximum values of the four constituting attributes. For instance, the hop hybrid with the lowest defined values in all attributes was assigned the worst assessment result. On the other hand, the hop hybrid with the highest relative assessment result signified the hybrid with the highest optimal set of values, i.e., the best prospective hybrid for further breeding.


Results and Discussion

Evaluation of the hybrids

Among all the hop plants analyzed and eliminated stepwise through a selection procedure in hop breeding (Cerenak, 2006), the four hop hybrids A1/54, A2/104, A3/112 and A4/122 and the reference variety were involved in a comparative model assessment. The hop hybrids had been selected among sets of seedlings analyzed and assessed as significantly promising new hop varieties. Numerical data of analyses and measurements of hop cones as well as beer sensory estimation were used to describe hybrids' production and brewing quality parameters. They were analyzed, and the results were additionally discussed. The model enabled a final assessment of the hybrids based on the defined attributes and the decision rules within the defined utility functions.

All hop hybrids were described with fitting values, which were defined for each of the basic attributes. Furthermore, a list in which the hybrids were ranked in order of performance was created. This enabled a relative scoring system, in which hop hybrids or varieties could be compared against one another based on their performance. The level of importance was shown for each criterion within the assessment and pointed out differences among the hybrids analyzed. In addition, the model's results offered information for the next steps in the breeding process.

Based on the breeding experiences and the model results, assessments of overall as well as individual attributes were carried out. A3/112 and A4/122 reached the overall level of the reference variety and were thus assessed as appropriate for further breeding or commercial use. In contrast, A1/54 and A2/104 did not meet expectations in their attributes in relation to the reference variety. A2/104 was overall assessed as WORSE, while A1/54 was assessed as NON-PROSPECTIVE. Therefore, they were considered hybrids with less breeding potential in terms of agricultural and brewing values. All evaluation results for the model attributes are shown in detail in Table 3 and additionally discussed.

A3/112 and A4/122

The model results showed that A3/112 was assessed as the hybrid with the highest breeding potential. In Biology, this hybrid was assessed as EXCELLENT, in Chemistry as REFERENCE, in Morphology as ACCEPTABLE and in Brewing value as GOOD. In relation to the reference variety, this hybrid had a comparative advantage in Biology but also a comparative disadvantage in Brewing value. A comparison of the alpha-acids value, for example, for A4/112 and the reference variety marked them as equal. Furthermore, A3/112 was relatively superior in ageing (HSI),plant resistance and lower cohumulone attributes, which have the most important economic effects in terms of the hop price. By all quality parameters analyzed in the attribute Brewing value, A3/112 was equivalent to the reference variety. In hop aroma intensity and quality, A3/112 was assessed as inferior, but otherwise in bitterness intensity as equal to the reference variety (Table 3, Figure 3).

A4/122 was well balanced in its properties. In no attribute did this hybrid receive either outstanding (+) or disappointing (—) marks. Furthermore, compared with A3/112, A4/122 was equally ranked in the attributes Chemistry, Morphology and Brewing value. However, A3/112 was better ranked in the attribute Biology. The hybrid A4/122 and the reference variety were in the broad spectrum equally ranked, but in some of the attributes both were behind the characteristics of A3/112 (Table 3).

A2/104 and A1/54

In terms of model results, A2/104 took third place among the four hybrids analyzed. This hybrid was assessed as equal to the reference variety in Chemistry and Morphology. A2/104 had higher alpha-acid content than the reference variety; however, Brewing value was assessed with the model's lowest mark, BAD (Table 3). Furthermore, this hybrid had relatively the highest (non-desirable) content of polyphenols as well as the lowest sensory assessment value in hop aroma intensity and bitterness intensity.

Hybrid A1/54 was assessed as GOOD in terms of the attribute Chemistry; however, even in that attribute, it ranked behind the other three hybrids analyzed. A1/54 was also ranked the lowest with BAD in Brewing value, the most important attribute. This hybrid required the highest quantity of hops needed within the brewing process. The sensory beer taste results ranked A1/54 at the bottom. Among all the hybrids, A1/54 scored the lowest marks in hop aroma and bitterness quality, which also resulted in the lowest mark BAD obtained for the Brewing value. That placed this hybrid in fourth place among the hybrids analyzed (Table 3).



Despite minor deficiencies such as the use of qualitative data only, we found that this approach fulfilled most of the breeders' expectations and revealed considerable advantages in comparison with other approaches. The multi-attribute model can therefore be regarded as a useful supplementary tool for hop hybrids assessment.

This method cannot entirely replace experts in breeding and brewing but can serve as an additional instrument in decision-making, since the decisions based on model testing offered much faster results that validated the application of the model for further research. In the future, data about new upcoming hybrids will be added and assessed in comparison with experts' decisions. Furthermore, we also expect to upgrade the model by adding new attributes in response to new goals in hop breeding programs.


Literature Cited

ASBC, 1992. American Society of Brewing Chemists. Methods of Analysis. 8th ed. H — 6,12. The Society, St. Paul, MN.         [ Links ]

Analytica EBC, 1998. European Brewery Convention. Grundwerk: Section 7. EBC Analysis Committee, Hans Carl Getraenke Fachverlag, Nuernberg, Germany. 33 p.         [ Links ]

Bohanec, M., and B. Zupan. 2004. A function-decomposition method for development of hierarchical multi-attribute decision models. Decis. Support Syst. 36: 215-233.         [ Links ]

Bohanec, M., J. Cortet, B. Griffiths, M. Znidarsic, M. Debeljak, S. Caul, J. Thompson, and P.H. Krogh. 2007. A qualitative multi-attribute model for assessing the impact of cropping systems on soil quality. Pedobiologia 51(3): 207—218.         [ Links ]

Bohanec, M. 2008. DEXi: Program for multi-attribute decision making, User's manual, Version 3.00. IJS Report DP-9989, Jozef Stefan Institute, Ljubljana. (Accessed: July 2008).         [ Links ]

Bohanec, M., A. Messean, S. Scatasta, F. Angevin, B. Griffiths, P.H. Krogh, M. Znidarsic, and S. Dzeroski. 2008. A qualitative multi-attribute model for economic and ecological assessment of genetically modified crops. Ecol. Model. 215: 247-261.         [ Links ]

Cerenak, A. 2006. Hop breeding. Final project report. Slovenian institute of hop research and brewing. Contract Nr. 231105-000090. 59 p.         [ Links ]

De Keukleire, D. 1999. Fundamentals of beer and hop chemistry. Quimica Nova 23(1): 108-112.         [ Links ]

Javornik, B., J. Jakse, N. Stajner, P. Kozjak, and A. Cerenak. 2005. Molecular genetic hop (Humulus Lupulus L.) research in Slovenia. Acta Hort. (ISHS) 668: 31-34.         [ Links ]

Lemmens, G. W. C. 1998. The breeding and parentage of hop varieties. Brew. Dig. 73(5): 16-26.         [ Links ]

MEBAK. 2002. Brautechnische Analysenmethoden, Band II. Methodensammlung der Mitteleuropaischen Brautechnischen Analysenkommission Weihenstephan. 332 p.         [ Links ]

Pavlovic, M., F. N. Koumboulis, M. P. Tzamtzi, and C. Rozman. 2008. Role of automation agents in agribusiness decision support systems. Agrociencia 42(8): 913-923.         [ Links ]

Pavlovic, V., A. Cerenak, M. Pavlovic, and C. Rozman. 2007. Use of numerical methods in hop breeding — concept of an expert system. Hmeljarski bilten 14: 11-17.         [ Links ]

Pazek, K., C. Rozman, A. Borec, J. Turk, D. Majkovic, M. Bavec, and F. Bavec. 2006. The Use of multi criteria models for decision support on organic farms. Biol. Agric. Hort. 24 (1): 73-89.         [ Links ]

Pazek, K., and C. Rozman. 2007. The Decision support system for supplementary activities on organic farms. Agricultura 5(1): 15-20.         [ Links ]

Radisek, S., A. Cerenak, and B. Javornik. 2007. Hop disease resistance breeding: Procedures and techniques in selection of seedlings. Hmeljarski bilten 14: 5-10.         [ Links ]

Rozman, C., and K. Pazek. 2005. Application of computer supported multi-criteria decision models in agriculture. Agric. Conspec. Sci. 70(4): 127-134.         [ Links ]

Rozman, C., K. Pazek, M. Bavec, F. Bavec, J. Turk, and D. Majkovic. 2006. The Multi-criteria analysis of spelt food processing alternatives on small organic farms. J. Sustainable Agric. 28(2): 159-179.         [ Links ]

Scatasta, S., J. Wesseler, M. Demont, M. Bohanec, S. Dzeroski, and M. Znidarsic. 2007. Multi-Attribute modeling of economic and ecological impacts of agricultural innovations on cropping systems. J. Systemics, Cybernetics and Informatics 4(2): 52-59.         [ Links ]

Seigner, E., A. Lutz, H. Radic-Miehle, S. Seefelder, and F. G. Felsenstein. 2005. Breeding for powdery mildew resistance in hop (Humulus lupulus L.). Strategies at the Hop Research Center, Huell, Germany. Acta Hort. (ISHS) 668: 19-31.         [ Links ]

Znidarsic, M., M. Bohanec, and B. Zupan. 2006. proDEX - a DSS tool for environmental decision-making. Environ. Model. Softw. 21(10): 1514-1516.         [ Links ]

Znidarsic, M., and M. Bohanec. 2007. Automatic revision of qualitative multi-attribute decision models. Fund. Computing Decis. Sci. 32(4): 315-326.         [ Links ]

Creative Commons License Todo el contenido de esta revista, excepto dónde está identificado, está bajo una Licencia Creative Commons