Path correlation and Bayesian analysis on popping expansion components in popcorn hybrids Received for publication: 17 June, 2019. Accepted for publication: 23 March, 2020 Doi: 10.15446/agron.colomb.v38n1.80462 1 Universidade Federal de Sergipe, Nossa Senhora da Glória, Sergipe (Brazil). 2 Universidade Estadual Paulista Júlio de Mesquita Filho, Jaboticabal, São Paulo (Brazil). 3 Universidade Estadual do Norte Fluminense, Campos dos Goytacazes, Rio de Janeiro (Brazil). * Corresponding author: gustavo.melhorista@gmail.com Agronomía Colombiana 38(1), 3-8, 2020 ABSTRACT RESUMEN Knowing the cause and effect among two or more traits can help to increase the selection accuracy of superior genotypes. The main objective of this study was to evaluate the cause and effect relationship between expansion volume and kernel size in pop- corn hybrids using path analysis and Bayesian network. A total of 41 popcorn hybrids were evaluated through a randomized complete block design (RCBD) with two replicates in the city of Jaboticabal, Brazil. The assessed traits were grain length (GL), grain thickness (GT), grain width (GW), caryopsis roundness index (CRI), mass of 50 grains (MG), and expansion volume (EV). Measurements were performed on individual grains, using three 50-grain samples from each plot. Pearson’s cor- relation coefficient, path analysis and Bayesian network were estimated. A negative correlation was detected among EV and the traits, except for GT. Path analysis indicated that MG has a direct and positive effect on EV and the negative correlation observed is mainly due to the indirect effects by GL and GT. Bayesian networks did not detect a direct association between kernel size and expansion volume while indicating that GT is the only trait that can affect popcorn f lake size. El conocimiento sobre la causa y el efecto entre dos o más rasgos puede ayudar a aumentar la precisión de la selección de genoti- pos superiores. El objetivo principal de este estudio fue evaluar la relación de causa y efecto entre el volumen de expansión y el tamaño del grano en los híbridos de palomitas de maíz utili- zando análisis de ruta y red bayesiana. Se evaluó un total de 41 híbridos de palomitas de maíz siguiendo un diseño de bloques completos al azar (DBCA) con dos repeticiones en la ciudad de Jaboticabal, Brasil. Los rasgos evaluados fueron la longitud del grano (LG), el grosor del grano (GG), el ancho del grano (AG), índice de redondez de la cariopsis (IRC), la masa de 50 granos (MG) y el volumen de expansión (VE). Las mediciones se realizaron en granos individuales, utilizando tres muestras de 50 granos de cada parcela. Se estimaron el coeficiente de correlación de Pearson, el análisis de ruta y la red bayesiana. Se detectó una correlación negativa entre VE y los rasgos, a excepción de GT. El análisis de ruta indicó que MG tiene un efecto directo y positivo sobre VE y la correlación negativa observada se debe principalmente a los efectos indirectos de LG y GG Las redes bayesianas no detectaron una asociación directa entre el tamaño del núcleo y el volumen de expansión, mientras que indicaban que GG es el único rasgo que puede afectar el tamaño de las hojuelas de palomitas de maíz. Key words: Zea mays L. var. everta, Bayesian networks, direct and indirect effect, correlation. Palabras clave: Zea mays L. var. everta, redes bayesianas, efecto directo e indirecto, correlación. Path correlation and Bayesian analysis on popping expansion components in popcorn hybrids Correlación de ruta y análisis Bayesiano sobre componentes de expansión en híbridos de maíz pipoca Gustavo Hugo Ferreira De Oliveira1*, Gabriela Pelegrini2, Tâmara Rebecca Albuquerque De Oliveira3, Maisa Nascimento Carvalho1, and Gustavo Vitti Môro2 Introduction The correlation between traits is of paramount importance for breeding programs, especially for the selection of su- perior genotypes. Therefore, for traits that are difficult to measure or have low heritability, indirect selection uses an easily measurable trait with moderate to high herita- bility, thus allowing the breeder to obtain faster progress compared to direct selection using a difficult trait. Cruz et al. (2012) state that associations can generate indirect gains when the studied traits show a genetic correlation, especially on truncated selection of low heritability traits, which may increase selection efficiency. It is known that the popping expansion (PE) can be geneti- cally correlated with other grain characteristics such as ker- nel size, length, width and mass (Cabral et al., 2016; Ribeiro et al., 2016). However, these results vary since other authors 10.15446/agron.colomb.v38n1.80462 4 Agron. Colomb. 38(1) 2020 report a positive correlation or a lack of linear correlation between the main kernel characteristics of popcorn, as described by Carpentieri-Pípolo et al. (2002) and Daros et al. (2004). Soylu and Tekkanat (2006) reported a significant and not significant positive correlation between PE and kernel size and PE and mass of 1000 grains, respectively. In the literature, several studies on the association between traits show that the concomitant use of path analysis and Bayesian networks may generate accurate information for genetic improvement of popcorn (Felipe et al., 2015; Amaral et al., 2019). Path analysis is the unfolding of the correla- tion coefficient that informs not only the magnitude of the correlation values but also allows obtaining the cause and effect information of the correlation between two variables and how many other variables may inf luence this correla- tion (Cruz et al., 2012). Bayesian networks (BN) are an alternative modeling ap- proach, in which criteria are applied for selection models. In turn, it is important in the graphical representation of the causality relations between the system variables (Marques and Dutra, 2002), revealing information on the relationship of conditional dependence, which is represented by the joint distribution of a set of variables. In addition, it is an intui- tive way to understand the direct and indirect relationships between variables (Felipe et al., 2015). The Directed Acyclic Graph (DAG) is the graphical representation produced by this probabilistic graphical model that represents a set of random variables and their conditional dependencies, con- nected directly by arrows (Margaritis, 2003). The main advantage of the BN is that the modeling involves searching for a structure that is compatible with the joint distribution of the data, and the conditional dependence brought by the BN structure allows generating the most parsimonious representation of the dataset joint distribu- tion (Amaral et al., 2019). This feature makes this model interesting for the purpose of predicting and selecting variables for genetic improvement. It is known that the greatest expansion capacity is a desired trait in the commercial production of popcorn and that this capacity depends on factors, such as grain size, grain flake size and grain mass, among others, as well as chemical properties (Cabral et al., 2016). Furthermore, the evaluation and measurement methods are not clearly addressed in the literature. Thus, the objective of this work was to evaluate the cause and effect relationship between expansion capac- ity, weight, and grain size in popcorn hybrids using path analysis and Bayesian network. Materials and methods A total of 185 S3 inbred lines of popcorn were obtained from nine populations from different origins. The lines were separated by similar agronomic traits and within each population. The seeds of the lines were mixed in the same proportion to build up the population of origin, thus, obtaining nine synthetic populations. In the 2013/2014 crop, the synthetic populations were sown in Jaboticabal, São Paulo, Brazil (2º15’17”S, 48º19’20”W, and 605 m a.s.l.) The climates in both places are classified as tropical Aw according to the Koppen classification, with 60 mm aver- age rainfall in the driest month and average temperature of 18°C in the coldest month of the year. The average tem- perature in the rainy season is 29°C. The soils are classified as Eutrophic Red Latosol and Dystrophic Red Latosol, respectively (Embrapa, 2006) and crossed in a complete diallel scheme by manual pollination. The progenies (diallel hybrids) were harvested, hand-milled, and part of the seeds was packed in a dry chamber, while another part was sown on the same site in the 2014/2015 crop. The randomized block experiment consisted of 41 diallel hybrids with two replicates, in plots of two 5 m lines, spaced 0.8 m between rows and 0.2 m between plants. Crop man- agement followed the requirements of the culture (Cruz, 2010). All ears in the plot were manually harvested at physi- ological maturity. After harvesting, all ears were threshed manually and kept in tagged paper bags in a dry chamber. From each replicate, three 50-grain samples were prepared and further used to evaluate the traits studied in this work. The work was divided into two steps. The first step con- sisted in estimating the traits for each 50-grain samples as follows: grain length (GL), defined as the distance between the tip and the base of the grain; grain width (GW), the distance from end to end of the widest part of kernel; grain thickness (GT), the distance between the two faces of the kernel, as proposed by Pordesimo et al. (1990), and caryopsis roundness index (CRI), the relationship between GL, GW and GT using the following equation proposed by Mohsenin (1970). CRI = (GL × GW × GT)1/3 (1) The traits GL, GW, and GT were measured with a digital caliper and expressed as millimeters. The second step consisted of estimating the traits for the complete 50-grain sample as follows: mass of 50 grains (MG) (considering a grain moisture content of 13%), measured using a precision scale and expressed as grams and popping expansion (PE), 5Ferreira De Oliveira, Pelegrini, Albuquerque De Oliveira, Nascimento Carvalho, and Vitti Môro: Path correlation and Bayesian analysis on popping expansion components in popcorn hybrids evaluated using a microwave (Brastemp®, 30 L, 0.8 KW/h, 800 W, 127 V, 30 x 53.9 x 42 cm, 16 kg). The grains were placed in paper bags and two repetitions per each genotype were used to subject the samples to the same popping time (2 min and 30 sec). PE was given by the ratio between f lake volume (ml) and its sample mass (g) and expressed in ml g-1. The f lake volume (VOL) was determined using a graduated 1000 ml cylinder, and homogenized by inverting the cylinder once, whereas f lake size (FS) was given by the ratio between VOL and the number of popped kernels in the sample. VOL and FS were used only in the Bayesian network analysis. The phenotypic (rf), genotypic (rg) and environmental (ra) correlations were estimated using the following expressions: 1 𝐶𝐶𝐶𝐶𝐶𝐶 = (𝐺𝐺𝐺𝐺.𝐺𝐺𝐺𝐺.𝐺𝐺𝐺𝐺)1/3 (1) 1 2 2 2 2 2 2 2 ( , ) ( , ) ( , ) ; ; . . . f g a fx fy gx gy ax ay Covf x y Covg x y Cova x y r r r σ σ σ σ σ σ = = = (2) 3 4 Pr 𝑋𝑋!,𝑋𝑋!,… ,𝑋𝑋! = Pr 𝑋𝑋!|Pa! ! !!! (3) 5 (2) where: Covf(x,y), Covg(x,y), and Cova(x,y) are, respectively, the phenotypic, genotypic and environmental correlations between the x and y variables; rf, rg and ra are the phenotyp- ic, genotypic, and environmental variances, respectively. The significance of coefficient b was evaluated by the F test and the correlation coefficient was analyzed by the t-test. A multicollinearity test was performed between the vari- ables and, according to this criterion, VOL and FS were removed from the database. Then, the rf matrix was used for unfolding the correlation coefficient into cause and effect studies of kernel traits on grain expansion capacity determined by the phenotypic path analysis (FPA). The correlation and path analyses were performed using the Genes software (Cruz, 2013). As a complement, a graphical analysis (Directed Acyclic Graph) was performed using Bayesian networks (BN). The BN is a graphical representation of a probability dis- tribution over a set of variables (Felipe et al., 2015). The Directed Acyclic Graph (DAG) represents the BN using nodes connected by arrows, and is used as an output to the modeling approach. In this case, it is used to illustrate the association between traits. This graph characterizes a joint probability of the data, which brings scale benefits due to the factorization (Aliferis et al., 2010). In a set of variables {X1, X2, …, Xp} with joint distribution Pr(X1, X2, …, Xp) and a DAG D that is compatible with this joint distribution (Pearl, 2000), the following factorization can be performed: 1 𝐶𝐶𝐶𝐶𝐶𝐶 = (𝐺𝐺𝐺𝐺.𝐺𝐺𝐺𝐺.𝐺𝐺𝐺𝐺)1/3 (1) 1 2 2 2 2 2 2 2 ( , ) ( , ) ( , ) ; ; . . . f g a fx fy gx gy ax ay Covf x y Covg x y Cova x y r r r σ σ σ σ σ σ = = = (2) 3 4 Pr 𝑋𝑋!,𝑋𝑋!,… ,𝑋𝑋! = Pr 𝑋𝑋!|Pa! ! !!! (3) 5 (3) where: Pai are the parents of Xi in D. The BN analysis involves searching for a structure that is compatible with the joint distribution of the data. The selected structure has already been used as a prediction tool, as described by Felipe et al. (2015). In this study, the BN was only used in the context of traits association. For the present work, the Hill-Climbing algorithm (“search and score” approach) was used to construct the BN from the means of each plot. The model was adjusted using the package “bnlearn” of the R software (Scutari, 2009). Results and discussion The majority of phenotypic (rf), genotypic (rg) and environ- mental (ra) correlation coefficients were significant (Tab. 1). Regarding the genetic correlation, six had a positive and significant rg (0.59, 0.62, 0.44, 0.84, 0.57, 0.33 to GL x GW, GL x MG, GW x GT, GW x MG, GT x MG and CRI x PE, respectively). Also, five had negative and significant rg (-0.72, -0.43, -0.49, -0.48, -0.39 to GW x CRI, GW x PE, GT x PE, CRI x MG and MG x PE, respectively). A positive correlation indicates that as one trait increases, the other trait also increases and that selection aiming at gain on one trait may ref lect on the other trait (Vencovsky and Barriga, 1992). In addition, five genetic correlations had higher val- ues than the environmental correlation, indicating that the observed linear association between these pairs of traits is little inf luenced by the environmental factors. TABLE 1. Phenotypic (rf), genotypic (rg) and environmental (ra) correla- tions between six grain traits of 41 diallel popcorn hybrids. Traits GW GT CRI MG PE GL rf 0.59 ** 0.21 0.19 0.58** -0.19 rg 0.59 ** 0.24 0.13 0.62** -0.20 ra 0.68 ** -0.14 0.46 0.10 0.01 GW rf 0.43 ** -0.63** 0.82** -0.41** rg 0.44** -0.72** 0.84** -0.43** ra 0.06 0.03 0.42** -0.04 GT rf -0.26 0.56** -0.48** rg -0.26 0.57** -0.49** ra -0.44** 0.36* -0.22 CRI rf -0.44** 0.29+ rg -0.48** 0.33* ra -0.34* 0.07 MG rf -0.38* rg -0.39** ra -0.15 *, **: significant at 5 and 1% by t-test; PE: popping expansion; GL: grain length; GW: grain width; GT: grain thickness; CRI: caryopsis roundness index; MG: mass of 50 grains. 6 Agron. Colomb. 38(1) 2020 On the other hand, of the significant 11 rg values, six (GL x GW, GW x CRI, GW x PE, GT x PE, CRI x MG and MG x PE) had environmental correlations higher than the respective genetic correlations. According to Vencovsky and Barriga (1992), environmental correlations occur when the same variation of environmental conditions affects two distinct traits, in which negative and positive values indicate that the environment favored one trait to the detriment of another or harmed both due to the same causes of environmental variations. Several authors point out the importance of the correlations between traits of popcorn hybrids (Soylu and Tekkanat, 2006; Cabral et al., 2016) especially regarding the degree of genetic and environmental association between the main commercial attributes of the crop. This is important because genetic correlations are inheritable and, therefore, can be used in breeding programs in the direct or indirect selection of superior genotypes. In addition, information on the correlations between grain traits may increase the accuracy of genotype selection in breeding programs, especially regarding popping expansion. The highest significant rf estimates were calculated for GW and MG (0.82) and GW and CRI (-0.63), although the second one is a negative correlation. Among the rg estimates, the highest was also observed for GW and MG (0.84) and for GW and CRI (-0.72). The highest ra estimate was obtained between GL and GW (0.68) and between GT and CRI (-0.44) (Tab. 1). Regarding PE, rf and rg were positive and significant for CRI and negative for GL, GW, GT, and MG. A positive significant genotypic correlation (rg) 0.33 was observed between PE and CRI, and negative but not significant be- tween PE and GL. Non-significant correlations imply no linear correlation but not necessarily lack of association between two traits (Cruz and Regazzi, 1997). The signifi- cant genotypic correlations between PE x GT and PE x GW were calculated as -0.49 and -0.43, respectively. The three main grain dimensions suggest that smaller grains have higher popping expansion during popping. It is known that grain shape greatly inf luences its popping expansion (Lyerly, 1942). Cabral et al. (2016) calculated the genetic correlations between these three characteristics and PE and reported negative values for GL and GW, but positive for GT. Lyerly (1942) stated that round grains have higher PE than elongated grains. The MG trait is related to the overall kernel size of the sample. In this study, the genotypic correlation (rg) between PE and MG was significant and negative, -0.39. It is note- worthy that MG had positive and significant correlation coefficients with all-grain dimensions (GW, GT, and GL). Carpentieri-Pípolo et al. (2002) reported a negative cor- relation between popping expansion and ear total mass and grain mass per plant, important traits of crop grain yield. Daros et al. (2004) reported non-significant values for PE and grain yield. These results confirm the important effect of kernel size on PE. Cabral et al. (2016) suggested transforming these measurements into popcorn volume (PV) (Super-trait) ha-1 (PV, m3 ha-1) to reduce the effect of these correlations on the selection of productive parents with popping expansion above the national average. To split the correlation coefficients in path analysis, a multicollinearity analysis was performed between variables (Cruz et al., 2012). The obtained condition number (NC) was 81.41, classified as weak; therefore, path analysis can be performed without statistical problems. Coimbra et al. (2005) stated that in the presence of multicollinearity, inconsistent estimates of the regression coefficient could be observed, while the direct effects of the explanatory variables on the response variable would be overestimated, and the irrational values would not explain the studied biological phenomenon. It was observed that the highest values of positive direct effects on PE were for MG, GL and CRI and negative val- ues for GT and GW (Tab. 2). These effects corroborate the results of rg and rf shown in Table 1. TABLE 2. Estimates of the direct and indirect effects of grain traits on the popping expansion (PE) of popcorn. Effect GL GW GT CRI MG Direct on PE 0.03 -0.34 -0.39 0.012 0.11 Indirect via GL 0.018 0.006 0.005 0.02 Indirect via GW -0.20 -0.15 0.219 -0.29 Indirect via GT -0.08 -0.16 0.105 -0.22 Indirect via CRI 0.002 -0.007 -0.003 -0.005 Indirect via MG 0.65 0.09 0.064 -0.05 Total -0.19 -0.41 -0.48 0.29 -0.38 R2 0.28 Residual effect 0.84 PE: popping expansion; GL: grain length; GW: grain width; GT: grain thickness; CRI: caryopsis roundness index; MG: mass of 50 grains. The highest total effect on PE (-0.47) was observed for GT. However, of that value, -0.15 is represented by GW indi- rectly. The second-largest total effect (-0.41) was observed for GW while GT contributed with -0.16, the largest share. 7Ferreira De Oliveira, Pelegrini, Albuquerque De Oliveira, Nascimento Carvalho, and Vitti Môro: Path correlation and Bayesian analysis on popping expansion components in popcorn hybrids These results confirm that PE is strongly inf luenced by kernel size and that increasing kernel size may negatively affect PE. The total effect of MG on the PE was -0.38 and the direct positive effect was 0.11. In this case, the path analysis indicates a rather pronounced complex relation- ship between MG and PE, with the negative correlation between MG and PE due to the indirect effects of kernel size, especially GW and GT. However, when the three dimensions (GL, GW, and GT) are transformed into CRI, a positive total and direct effect on PE is observed. These results suggest that the CRI trait should be used for indirect selection aiming at genetic gains in EC. The residual effect was 0.84, indicating that a correlation in path analysis, to be statistically and genetically relevant, would need to be greater than the residual value of 0.84 (Souza, 2013). The associations between the evaluated traits using BN can be visualized using Directed Acyclic Graph (DAG) (Fig. 1). In this study, BN was used only to visualize the relation- ship between the traits of popcorn, in order to assess the dependence relationship of the studied variables. According to Yu et al. (2004), Bayesian networks can be used to demonstrate biological phenomena in experimen- tal data. Thus, grain dimensions such as width, length, thickness, and diameter are not directly related to PE and do not need to be used in prediction models (Fig. 2). It is noteworthy that among the dimensions measured in the grain, only GT can inf luence f lake size (FS). All the above considering that f lake volume depends exclusively on FS and that, based on these results, increasing kernel thickness may lead to genotypes with low EC values. DAG shows no direct or indirect relationship of GL and CRI with PE. It is also observed that GT and GW are directly associated with MG and grain size and can indirectly in- f luence the PE. These results corroborate the path analysis (Tab. 2) and those presented by Cabral et al. (2016) using correlation and path analysis. Although the analyses had few divergences, the con- comitant use of these strategies can increase the selection efficiency on traits aiming at increasing PE in popcorn breeding programs. Conclusions There is a significant negative genetic correlation between the popping expansion and grain width and thickness. The correlations between the mass of 50 grains with the popping expansion are due to the indirect effects of grain width and thickness. Grain thickness, width, length and diameter are not direct- ly associated with the popping expansion according to BN. We suggest using the Bayesian network, together with path analysis to aid in the selection and evaluation of traits in popcorn breeding programs. Literature cited Aliferis, C.F., A. Statnikov, I. Tsamardinos, S. Mani, and X.D. Kout- soukos. 2010. Local causal and Markov blanket induction for causal discovery and feature selection for classification part I: algorithms and empirical evaluation. J. Mach. Learn. Res. 11(7), 171-234. Amaral, C.B., G.H.F. Oliveira, and G.V. Môro. 2019. Bayesian network: a simplified approach for environmental similarity studies in maize. Crop Breed. Appl. Biot. 19(1), 70-76. Doi: 10.1590/1984-70332019v19n1a10 Cabral, P.D.S., A.T. de Amaral Junior, I.L. de J. Freitas, R.M. Ri- beiro, and T.R. da C. Silva. 2016. Relação de causa e efeito de caracteres quantitativos sobre a capacidade de expansão do grão em milho-pipoca. Rev. Ciênc. Agron. 47(1), 108-117. Doi: 10.5935/1806-6690.20160013 FS CRI GL GW MG GT PE VOL FIGURE 1. Bayesian network for the kernel traits of popcorn: grain width (GW), grain length (GL), caryopsis roundness index (CRI), flake volume (VOL), flake size (FS), grain thickness (GT), mass of 50 grains (MG), and popping expansion (PE). In the Bayesian network, the associations are arranged hi- erarchically, so it becomes easy to visualize the importance in terms of prediction and conditional dependence of the direct association between the evaluated traits (Felipe et al., 2015). Thus, DAG reveals that only MG and VOL are directly related to PE, indicating dependence on the mass of 50 grains and the popped volume of the popcorn. It is known that this relationship exists since PE is determined by the relationship between these two traits. 8 Agron. Colomb. 38(1) 2020 Carpentieri-Pípolo, V., H.W. Takahashi, R.M. Endo, M.R. Petek, and A.L. Seifert. 2002. Correlações entre caracteres quantitativos em milho pipoca. Hortic. Bras. 20(4), 551-554. Doi: 10.1590/ S0102-05362002000400008 Coimbra, J.L.M., G. Benin, E.A. Vieira, A.C. de Oliveira, F.I.F. Car- valho, A.F. Guidolin, and A.P. Soares. 2005. Conseqüências da multicolinearidade sobre a análise de trilha em canola. Ciênc. Rural 35(2), 347-352. Doi: 10.1590/S0103-84782005000200015 Cruz, C.D. 2013. Genes: a software package for analysis in experi- mental statistics and quantitative genetics. Acta Scient. Agron. 35(3), 271-276. Doi: 10.4025/actasciagron.v35i3.21251 Cruz, C.D. and A.J. Regazzi. 1997. Modelos biométricos aplicados ao melhoramento genético. UFV, Viçosa, Brazil. Cruz, C.D., A.J. Regazzi, and P.C.S. Carneiro. 2012. Modelos bio- métricos aplicados ao melhoramento genético. UFV, Viçosa, Brazil. Cr u z , J.C . (ed.). 2010. Cu lt ivo do mi l ho. Embrapa Mi l ho e Sorgo. (Sistema de produção 1). Sete Lagoas - MG. URL: ht t ps://w w w. s p o.c npt i a .embr apa .br/ l i s t a sptema?p_ p_ id=listaspportemaportlet_WAR _sistemasdeproducaolf6_ 1ga1ceportlet&p_p_lifecycle=0&p_p_state=normal&p_p_ m o d e = v i e w & p _ p _ c o l _ i d = c o l u m n - 2 & p _ p _ c o l _ cou nt=1& p_ r_p_ 619796851_tema Id=1712& _ listasppor- temapor t let _WA R _ sistemasdeproducaolf6 _1ga1cepor t l et_redirect=%2Ftemas-publicados (accessed 16 March 2020). Daros, M., A.T. do Amaral Júnior, M.G. Pereira, F.S. Santos, C.A. Scapim, S. de P. Freitas Júnior, R.F. Daher, and M.R. Ávila. 2004. Correlations among agronomic traits in two recurrent selection cycles in popcorn. Ciênc. Rural 34(5), 1389-1394. Doi: 10.1590/S0103-84782004000500010 Embrapa. 2006. Sistema brasileiro de classificação de solos. Em- brapa, Brasilia. Felipe, V.P., M.A. Silva, B.D. Valente, and G.J. Rosa. 2015. Using multiple regression, Bayesian networks and artificial neural networks for prediction of total egg production in European quails based on earlier expressed phenotypes. Poult. Scien. 94(4), 772-780. Doi: 10.3382/ps/pev031 Lyerly, P.J. 1942. Some genetic and morphologic characters affecting the popping expansion of popcorn. J. Amer. Soc. Agron. 34, 986-995. Doi: 10.2134/agronj1942.00021962003400110003x Margaritis, D. 2003. Learning Bayesian network model structure from data. PhD thesis, Carnegie Mellon University, Pitts- burgh, USA. Marques, R.L. and I. Dutra, 2002. Redes Bayesianas: o que são, para que servem, algoritmos e exemplos de aplicações.  Co- ppe Sistemas, Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil. Mohsenin, N.N. 1970. Physical properties of plant and animal materials: structure, physical characteristics and mechanical properties. Gordon and Breach, New York, USA. Doi: 10.1002/ food.19870310724 Pearl, J. 2000. Causality: models, reasoning and inference. Cam- bridge Universit y Press, New York, USA. Doi: 10.1017/ S0266466603004109 Pordesimo, L.O., R.C. Anantheswaran, A.M. Fleiscaman, Y.E. Lin, and M.A. Hanna. 1990. Physical properties as indicators of popping characteristics of microwave popcorn. J. Food Sci. 55(5), 1352-1355. Doi: 10.1111/j.1365-2621.1990.tb03934.x Ribeiro, R.M., A.T. do Amaral Júnior, G.F. Pena, M. Vivas, R.N. Kurosawa, and L.S.A. Gonçalves. 2016. Effect of recurrent selection on the variability of the UENF-14 popcorn popu- lation.  Crop Breed. Appl. Biotechnol. 16(2), 123-131. Doi: 10.1590/1984-70332016v16n2a19 Scutari, M. 2009. Learning Bayesian networks with the bnlearn R package. CRC Press, Florida, USA. Doi: 10.18637/jss.v035.i03 Souza, T.V. 2003. Aspectos estatísticos da análise de trilha (Path analysis) aplicada em experimentos agrícolas. MSc thesis, Universidade Federal de Lavras, Lavras, Brazil. Soylu, S. and A. Tekkanat. 2006. Interactions amongst kernel prop- erties and expansion volume in various popcorn genotypes. J. Food Eng. 80(1), 336-341. Doi: 10.1016/j.jfoodeng.2006.06.001 Vencovsky, R. and P. Barriga. 1992. Genética biométrica no fito- melhoramento. Sociedade Brasileira de Genética, Ribeirão Preto, Brazil. Yu, J., V.A. Smith, P.P. Wang, A.J. Hartemink, and E.D. Jarvis. 2004. Advances to Bayesian network inference for generating causal networks from observational biological data. Bioinformatics 20(18), 3594-3603. Doi: 10.1093/bioinformatics/bth448 https://www.spo.cnptia.embrapa.br/listasptema?p_p_id=listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet&p_p_lifecycle=0&p_p_state=normal&p_p_mode=view&p_p_col_id=column-2&p_p_col_count=1&p_r_p_619796851_temaId=1712&_listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet_redirect=%2Ftemas-publicados https://www.spo.cnptia.embrapa.br/listasptema?p_p_id=listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet&p_p_lifecycle=0&p_p_state=normal&p_p_mode=view&p_p_col_id=column-2&p_p_col_count=1&p_r_p_619796851_temaId=1712&_listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet_redirect=%2Ftemas-publicados https://www.spo.cnptia.embrapa.br/listasptema?p_p_id=listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet&p_p_lifecycle=0&p_p_state=normal&p_p_mode=view&p_p_col_id=column-2&p_p_col_count=1&p_r_p_619796851_temaId=1712&_listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet_redirect=%2Ftemas-publicados https://www.spo.cnptia.embrapa.br/listasptema?p_p_id=listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet&p_p_lifecycle=0&p_p_state=normal&p_p_mode=view&p_p_col_id=column-2&p_p_col_count=1&p_r_p_619796851_temaId=1712&_listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet_redirect=%2Ftemas-publicados https://www.spo.cnptia.embrapa.br/listasptema?p_p_id=listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet&p_p_lifecycle=0&p_p_state=normal&p_p_mode=view&p_p_col_id=column-2&p_p_col_count=1&p_r_p_619796851_temaId=1712&_listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet_redirect=%2Ftemas-publicados https://www.spo.cnptia.embrapa.br/listasptema?p_p_id=listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet&p_p_lifecycle=0&p_p_state=normal&p_p_mode=view&p_p_col_id=column-2&p_p_col_count=1&p_r_p_619796851_temaId=1712&_listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet_redirect=%2Ftemas-publicados https://www.spo.cnptia.embrapa.br/listasptema?p_p_id=listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet&p_p_lifecycle=0&p_p_state=normal&p_p_mode=view&p_p_col_id=column-2&p_p_col_count=1&p_r_p_619796851_temaId=1712&_listaspportemaportlet_WAR_sistemasdeproducaolf6_1ga1ceportlet_redirect=%2Ftemas-publicados