FACULDADE DE CIÊNCIAS AGRÁRIAS E VETERINÁRIAS 

UNIVERSIDADE ESTADUAL PAULISTA 

CÂMPUS DE JABOTICABAL 

 

 

 

 

 

 

POWER OF QTL MAPPING OF DIFFERENT GENOME-WIDE 

ASSOCIATION METHODS FOR TRAITS UNDER 

DIFFERENT GENETIC STRUCTURES: A SIMULATION 

STUDY 

 

 

 

 

 

Baltasar Fernandes Garcia Neto 

Zootecnista 

 

 

 

 

 

 

 

 

 

 

2018 



FACULDADE DE CIÊNCIAS AGRÁRIAS E VETERINÁRIAS 

UNIVERSIDADE ESTADUAL PAULISTA 

CÂMPUS DE JABOTICABAL 

 

 

 

 

 

POWER OF QTL MAPPING OF DIFFERENT GENOME-WIDE 

ASSOCIATION METHODS FOR TRAITS UNDER 

DIFFERENT GENETIC STRUCTURES: A SIMULATION 

STUDY  

 

 

 

Baltasar Fernandes Garcia Neto 

Advisor: Dr. Roberto Carvalheiro 

 

 

 

Dissertation presented to The Faculdade de 
Ciências Agrárias e Veterinárias Unesp, 
Câmpus de Jaboticabal, in partial fulfilment 
of requeriments for the degree of Mestre em 
Genética e Melhoramento Animal) (Master in 
Animal Breeding and Genetics) 

 

 

 

 

 

 

2018 



 

 

 

 

 

 

 

 

 

 

 

  
Garcia Neto, Baltasar Fernandes 

G216p  Power of qtl mapping of different genome-wide association 
methods for traits under different genetic structures: a simulation study 
/ Baltasar Fernandes Garcia Neto. – – Jaboticabal, 2018 

 v, 22 p. : il. ; 28 cm 

  
 Dissertação (mestrado) - Universidade Estadual Paulista, 

Faculdade de Ciências Agrárias e Veterinárias, 2018 

 Orientador: Roberto Carvalheiro 

Banca examinadora: Danisio Prado Munari, Ricardo Vieira Ventura 

 Bibliografia 

  
 1. GWAS. 2. Genetic Structure. 3. wssGBLUP. I. Título. II. 

Jaboticabal-Faculdade de Ciências Agrárias e Veterinárias. 
  

CDU 636.082 

  
Ficha catalográfica elaborada pela Seção Técnica de Aquisição e Tratamento da Informação – 

Diretoria Técnica de Biblioteca e Documentação - UNESP, Câmpus de Jaboticabal. 

 

 

 

 

 

 

 

 

 

 

 



 

 

 



 

 

DADOS CURRICULARES DO AUTOR 

 

 

Baltasar was born at Jaboticabal, Sao Paulo in October 7th, 1991. In 2010, he started 

the Animal Science undergraduate course at Universidade Estadual Paulista – 

UNESP in the same city. Since the beginning of the course, Baltasar was engaged in 

Aquaculture area, more specifically in shrimp and prawn farming. He participated in 

different post-graduate studies, and had a scientific research scholarship with Prof. 

Dr. Wagner Valenti, granted by the National Council for Scientific and Technological 

Development (CNPq). In 2013, he was granted with a scholarship (Science without 

Borders) provided by Coordination for the Improvement of Higher Education 

Personnel (CAPES) to study for one year in Ireland, at the National University of 

Galway – NUIG, where he was able to improve his English skills, attend some 

courses and learn more about a different culture. In 2015, he took his undergraduate 

degree in Animal Science at UNESP, defending a literature review research on 

heritability estimates in shrimp, under the supervision of Dr. Roberto Carvalheiro. In 

the same year, he spent 3 months at La Paz, Baja California Sur, Mexico doing his 

obligatory internship at CIBNOR research center, under the supervision of Prof. Dr. 

Ricardo Perez-Enriquez. This internship was focused on genetic molecular analysis, 

disease and pedigree identification in shrimp. In 2016, he started his M. Sc. at 

UNESP, in the animal breeding and genetics post-graduate program, with a 

scholarship from CAPES, under the supervision of Dr. Roberto Carvalheiro finishing 

it in 2018.  In the same year he started his PhD study at the same program and 

supervisor. 

 

 

 

 

 

 



DEDICATION 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I dedicate this dissertation to my 

family. My mother Silvana, my 

father Baltasar and my sister Ana 

Carolina. To my grandmother 

Maria and my fiancée Rachel.  

 

 

 



ACKNOWLEDGMENTS 

 

First of all, I thank to God for all blesses in my life. 

I would like to thank Dr. Roberto Carvalheiro for all the teachings and advices. I 

thank for all patience, kindness, and for being an example as animal scientist.  

I thank to Dra. Sandra Aidar de Queiroz, Dra. Lucia Galvão de Albuquerque and Dr. 

Danísio Prado Munari, animal breeding and statistics professors on my 

undergraduate and post-graduate period. Thank you for all the knowledge and 

incentive to study animal breeding. 

I would like to express my gratitude to all members of the internal and external 

examining committees: Dr. Danísio Prado Munari, Dr. Henrique Nunes de Oliveira 

and Dr. Ricardo Vieira Ventura for spending their valuable time and attention with 

this dissertation and for their priceless suggestions and contributions. 

I wish to thank MSc Thaise Pinto de Melo for all help and patience. I learned so 

much with you. Thank you for your valuable cooperation.  

I thank to Dr. Haroldo Henrique de Rezende Neves in name of GenSys Consultores 

Associados for all the support and effort in developing this dissertation. 

I am really grateful to the Genetics and Animal Breeding post graduate program and 

Animal Science undergraduate course at FCAV- UNESP Jaboticabal. Thank you for 

all knowledge presented to me since 2010.  

I thank the financial support from Coordenação de Aperfeiçoamento de Pessoal de 

nível Superior (CAPES).  

I wish to thank my friends from the shrimp farming sector at CAUNESP. In special to 

Rafael, Caio and Roberto. Thank you for the friendship and advices all these years.  

I thank to my friends of the animal science department: Grazy, Henrique, Giovana, 

Laiza, Thaise, Daiane, Angel, Helsi, André and Andrés. 

I thank to my friends: Leonardo (Jamal), Guilherme (Guizim), Kayo (Diabético), 

Jhonatan (Jhow), Vitor (Vitão), José (Zé-Mayer), Camilotti, João, Matheus (Bolacha). 

Thank you for all memorable moments and friendship. 

To my fiancée Rachel for all loving support and for being my partner in the difficult 

and good times. 



i 
 

 

CONTENTS 

 

ABSTRACT ............................................................................................................... ii 

RESUMO .................................................................................................................. iv 

Literature Review ..................................................................................................... 1 

Introduction .............................................................................................................. 5 

Material and Methods ............................................................................................... 7 

Simulated scenarios .......................................................................................... 7 

Population structure .......................................................................................... 7 

Genome simulation ........................................................................................... 8 

GWAS analyses ................................................................................................. 9 

Comparison criteria ......................................................................................... 11 

Results and Discussion ......................................................................................... 11 

Conclusions ............................................................................................................ 19 

References .............................................................................................................. 20 

 

 

 

 

 

 

 

 

 

 

 

 

 



ii 
 

 
POWER OF QTL MAPPING OF DIFFERENT GENOME-WIDE ASSOCIATION 

METHODS FOR TRAITS UNDER DIFFERENT GENETIC STRUCTURES: A 

SIMULATION STUDY 

 

ABSTRACT 

 

The complexity of the traits that can present different genetic structures, such as 

polygenic or affected by genes of major effect, in addition to different heritabilities, 

among other factors, make the detection of QTLs challenging. Several methods have 

been employed with the purpose of performing genome wide association studies 

(GWAS), aiming the mapping of QTL. The single-step weighted GBLUP 

(wssGBLUP) method, for example, is an alternative to GWAS, which allows the 

simultaneous use of genotypic, pedigree and phenotypic information, even from non-

genotyped animals. Bayesian methods are also used to perform GWAS, starting 

from the basic premise that the observed variance can vary at each locus with a 

specific priori distribution. The objective of the present study was to evaluate, 

through simulation, which methods, among the evaluated ones, more assist in the 

identification of QTLs for polygenic and major gene affected traits, presenting 

different heritabilities. We used the following methods: wssGBLUP, with or without 

additional phenotypic information from non-genotyped animals and two different 

weights for markers, where w1 represented the same weight (w1=1) and w2 the 

weight calculated according to the previous iteration process (w1); Bayes C, 

assuming two values for π (π = 0.99 and π = 0.999), where π is the proportion of 

SNPs not included in the model, and Bayesian LASSO. The results showed that for 

polygenic scenarios the detection power is lower and the additional use of 

phenotypes from non-genotyped animals may help in the detection, yet with low 

intensity. For scenarios with major effect, there was greater power in the detection of 

QTL by all different methods with slighter superior performance for the Bayes C 

method. However, the inclusion of additional phenotypic information caused bias in 

the estimates and harmed the performance of the wssGBLUP in the presence of 

major QTL. The increase in heritability for both structures improved the performance 

of the methods and the power of mapping. The most suitable method for the 



iii 
 

detection of QTL is dependent on the genetic structure and the heritability of the trait, 

and there is not a superior method for all scenarios. 

Keywords: GWAS; Genetic Structure; wssGBLUP; Bayesian Methods. 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 



iv 
 

PODER DE MAPEAR QTL DE DIFERENTES MÉTODOS DE ASSOCIAÇÃO 

GENÔMICA AMPLA PARA CARACTERÍSTICAS COM DIFERENTES 

ESTRUTURAS GENÉTICAS: ESTUDO DE SIMULAÇÃO 

 
 

RESUMO 

 

A complexidade das características que podem apresentar diferentes estruturas de 

ação gênica como, por exemplo, poligênicas ou afetadas por genes de efeito maior, 

aliado a diferentes herdabilidades, entre outros fatores, tornam a detecção de QTLs 

desafiadora. Diversos métodos têm sido empregados com o intuito de realizar 

estudos de associação ampla do genoma (GWAS), objetivando o mapeamento de 

QTL. A metodologia weighted single-step GBLUP (wssGBLUP), por exemplo, é uma 

alternativa para a realização de GWAS, que permite o uso simultâneo de 

informações genotípicas, de pedigree e fenotípicas, mesmo de animais não 

genotipados. Métodos Bayesianos também são utilizados para a realização de 

GWAS, partindo da premissa básica de que a variância observada pode variar em 

cada locus em uma distribuição a priori específica. O objetivo do presente estudo foi 

avaliar, por meio de simulações, quais métodos, dentre os avaliados, mais auxiliaria 

na identificação de QTLs para características poligênicas e afetadas por genes de 

efeito maior, apresentando diferentes herdabilidades. Utilizamos os métodos: 

wssGBLUP, com a inclusão ou não de informação adicional fenotípica de animais 

não genotipados e dois distintos ponderadores para os marcadores, onde w1 

representou a mesma ponderação (w1=1) e w2 a ponderação calculada de acordo 

com o processo de iteração anterior (w1) ; Bayes C, assumindo dois valores para π 

(π=0.99 and π=0.999), onde π é a proporção de SNPs não incluída no modelo, 

além do LASSO Bayesiano. Os resultados mostraram que para cenários poligênicos 

o poder de detecção é menor e o uso adicional de fenótipos de animais não 

genotipados pode ajudar na detecção, ainda que com pouca intensidade. Para 

cenários com característica sob efeito maior, houve maior poder na detecção de 

QTL pelos diferentes métodos em comparação aos cenários poligênicos com 

destaque para a leve vantagem do método Bayes C. A inclusão de informação 

fenotípica adicional, entretanto, causou viés nas estimativas e atrapalhou o 

desempenho do wssGBLUP na presença de QTL com efeito maior. O aumento da 



v 
 

herdabilidade para ambas as estruturas melhorou o desempenho dos métodos e o 

poder de mapeamento. O método mais adequado para a detecção de QTL depende 

da estrutura genética e da herdabilidade da característica, não existindo um método 

que seja superior para todos os cenários. 

Palavras-chave: GWAS; Estrutura Genética; wssGBLUP; Métodos Bayesianos. 

 
 



1 
 

Literature Review 

 Several strategies have been adopted to increase productivity and profitability 

in agricultural systems due to human population growth and environmental 

resources limitation. Among them, we highlight the application of selective breeding 

which has been extensively used in animal breeding (Bourdon et al., 2000). 

Statistical methods based on quantitative genetics and, more recently, molecular 

approaches are tools often used to help in this task. 

 Through the advance of technology, such as high density marker panels and 

bioinformatical and statistical techniques, the use of genomic information was 

enabled to estimate more accurately the breeding values of animals showing positive 

results in a relative short-term period as experienced by the dairy cattle sector 

(Hayes et al., 2009). The understanding of how the target trait is expressed and the 

detection of possible genomic regions related to it has been an important subject of 

investigation. Genome-wide association studies (GWAS) were developed aiming to 

obtain statistical associations between the target trait and markers (Goddard and 

Hayes, 2009). 

 However, several traits with economic importance are affected by a large 

number of genes with small effect hardening GWAS. In addition, different factors 

may affect the power of detecting quantitative trait loci (QTL), namely: linkage 

disequilibrium between markers and QTL, trait’s heritability, availability of genotypic 

information and genetic structure, among others (Van den Berg et al., 2013).  

 Different methods and approaches are used in order to perform GWAS. The 

Bayesian methods were primary developed to genomic selection purposes and then 

fitted to GWAS. They usually assume that the variance may fluctuate at each locus 

with a specific prior distribution. Meuwissen et al. (2001) proposed two Bayesian 

methods to estimate marker effects, one based on the inclusion of all markers in the 

model using a scaled t as prior distribution and a second one, with a fixed proportion 

of markers included in the model that leads to a posterior distribution of marker 

effects having higher density at 0, namely Bayes A and Bayes B, respectively. 

Gianola et al. (2009) observed some drawbacks on these methods related to the 

prior hyperparameter impacts on shrinkage of SNPs effects and the prior probability 

π, that a SNP has zero effect is considered as known. Habier et al. (2011) developed 



2 
 

two alternative methods that could overcome these problems. Trying to reduce the 

scale parameter influence, Bayes C was developed   considering the same variance 

for all SNPs, and Bayes D, with a priori scale parameter as unknown. Another point 

exploited was the estimation of π, the proportion of SNPs with zero effect. In Bayes 

A and Bayes B, π was considered as known, thus estimating it directly from the data 

could improve the results, once π may influence the shrinkage of SNP effects. These 

modifications originated Bayes Cπ and Bayes Dπ methods. Van den Berg et al. 

(2013) compared the Bayes Cπ and Bayes C method in a GWAS study and found 

that fixing the π value is more suitable to QTL detection, instead of calculating it from 

the data, especially for polygenic low heritable traits. 

The Bayesian least absolute shrinkage and selection operator (LASSO) 

method has also been used in genomic prediction. Tibshirani (1996) proposed it 

using the sum of the absolute values of the regression coefficients as a penalty in 

regression models, to simultaneously produce variable selection and shrinkage of 

coefficients. This operator has the desirable feature of including in the model only a 

subset of explanatory variables, setting to zero those that have nil effects. De los 

Campos et al. (2009) modified and extended the LASSO to accommodate pedigree 

information and marker data into a single model in the context of QTL analysis. 

Waldman et al. (2013) using real and simulated data showed that LASSO may be an 

important tool in GWAS, although some issues can possibly affect its performance 

towards specific scenarios, i.e. most of LASSO methods used in GWAS consider 

that the sample members are unrelated to each other, which may not be true as 

often genetic studies enroll multiple members of families (Papachristou et al., 2016). 

An alternative approach to perform GWAS is the genomic best linear 

unbiased prediction (GBLUP) which was firstly developed using multiple step 

procedures to predict genomic estimated breeding values (GEBV) in genome 

selection (VanRaden 2008). A limitation for this method is that it uses phenotypes or 

pseudo-phenotypes only from genotyped animals. To overcome this drawback, a 

single-step GBLUP method (ssGBLUP) was proposed allowing using all available 

phenotypic, pedigree and genotypic information simultaneously (Legarra et al., 2009; 

Christensen and Lund 2010). The method integrates the G matrix (genomic 

relationship) and the A matrix (pedigree relationship) into an H matrix. The use of 

ssGBLUP has proven to be successful in studies for different species – beef cattle 

(Lourenco et al., 2015), dairy cattle (VanRaden 2012), pigs (Forni et al., 2011) and 



3 
 

chicken (Chen et al., 2011; Fragomeni et al., 2014). For GWAS, the ssGBLUP has 

also been used (ssGWAS). Dikmen et al. (2013) performed ssGWAS for rectal 

temperature in Holstein Cattle in order to detect association between markers and 

genes with major effects for this trait. Tiezzi et al. (2015) also used ssGWAS trying to 

find QTL associated with clinical mastitis in first parity of U.S. Holstein cows.  

However, a remarkable drawback for ssGWAS is the prior assumption of 

equal variance for all markers, which is not necessarily true when the target trait is 

affected by markers with more pronounced effects than others. Wang et al. (2012) 

suggested to weight markers variance according to their importance for the trait, i.e., 

to perform the ssGBLUP principles with the possibility of using all phenotypic (even 

from non-genotyped animals), genotypic and pedigree information available adding 

optimal weights for markers variance corresponding to their contribution to the trait. 

This method, known as weighted ssGBLUP (wssGBLUP), is quite useful for some 

scenarios such as: plenty more phenotypic than genotypic information, a common 

condition in real datasets; and complex models (e.g. non-linear; multiple traits). 

Studies comparing GWAS methods showed that the most suitable method depends 

on the amount of genotypic and phenotypic data available and the proportion of 

markers linked to the genetic variance (Lourenço et al., 2014; Wang et al., 2014; 

Melo et al., 2016). 

GWAS have increased significantly with multiple QTL identified to different 

traits and species. However, a short list of these QTL were validated or reproduced 

by other studies demanding attention for how the results should be interpreted 

(Fragomeni et al., 2014). The data simulation is an important tool to previously know 

important genomic regions and evaluate different methods. For example, Vitezica et 

al. (2011) compared multiple and single-step procedures considering the ability to 

predict the GEBV. Wang et al. (2012) also tested different methods and levels of 

information inclusion observing the accuracy of prediction. Van den Berg et al. 

(2013) evaluated two methods regarding different scenarios and availability of 

information. Melo et al. (2016) verified the power of QTL detection under different 

linkage disequilibrium and using different methods.  

The QTL detection remains a challenging task in real data. For instance, 

despite the decrease in genotyping costs, methods that use less genotypic 

information might be very interesting, since they do not compromise efficiency and 

accuracy to the predictions. In this sense, the present study was developed aiming 



4 
 

to: 1) assess which method present better power for QTL mapping in scenarios with 

different genetic architectures and trait heritability and 2) evaluate the benefit of 

inclusion of additional phenotypic information from non-genotyped animals in the 

QTL mapping. 



5 
 

Introduction 

 From the evolution of new technologies such as, high density marker panels 

and advances in bioinformatics, the mechanisms that affect the expression of traits 

with economic relevance have been more understood in several species (Goddard 

and Hayes 2009). Genome wide association studies (GWAS) are being adopted in 

order to assist in the identification of QTL (quantitative trait loci), allowing the 

detection of statistical associations between important traits and available markers. 

The identification of QTL may allow obtaining more accurate predictions of the 

genetic merit and, as a result, improve the genetic progress (Meuwissen et al., 

2001). 

Number of markers related to the expression of a trait, genetic structure, 

linkage disequilibrium between markers and QTL, trait’s heritability, and the 

availability of phenotypic and genotypic information, are important issues that should 

be analyzed in the QTL mapping (Van den Berg et al., 2013). In terms of genetic 

structure, two main situations that may present difficulties to GWAS are either that a 

trait is controlled by common variants under small phenotypic effect or in contrast, 

several uncommon variants, each with a large effect on the phenotype. In both cases 

the causative mutation may be clustered in a single gene or a small number of genes 

(major effect), or across several genes (polygenic). For this reason, statistical 

methods to correct associate the markers with these causal mutations are needed in 

order to achieve better accuracy in the mapping. 

Several methods are available to perform GWAS but there is no agreement 

whether there is a most suitable method. An approach widely used in GWAS is the 

multiple regression Bayesian models (Fernando and Garrick, 2013). These methods 

are able to produce shrinkage and/or variable selection, including all markers 

simultaneously, assuming different prior distributions. The Bayes A and Bayes B 

(Meuwissen et al., 2001), Bayes C and Bayes Cπ (Habier et al., 2011) and Bayesian 

LASSO (Tibshirani 1996, De los Campos et al., 2009) are examples of Bayesian 

methods often used on GWAS.  

The Bayes B method has a fixed proportion of markers included in the model 

that leads to a posterior distribution of marker effects having higher density at 0.  

Gianola et al. (2009) suggested modifications on this method arguing that even with 

marker-specific variances on their models, the shrinkage of effects will still depend 



6 
 

strongly on the prior distribution. In this sense, Habier et al. (2011) proposed the 

Bayes Cπ method. This method has the prior assumption that marker effects have 

identical and independent mixture distributions and π (the proportion of SNPs with 

zero effect) is treated as unknown with a uniform prior. Van den Berg et al. (2013) 

used this method in a simulation study and concluded that when a small number of 

records are available and/or several QTL are affecting a trait with low heritability, to 

apply a fixed value for π is more efficient for QTL mapping than estimate it direct 

from data. Melo et al. (2016) also used the Bayes Cπ method, with two different 

values for π, and found that applying a higher shrinkage did not necessarily reflect 

better QTL detection.  

The Bayesian least absolute shrinkage and selection operator (LASSO) 

method (Tibshirane, 1996) has also been used in GWAS. Its main desirable feature 

is including just the explanatory fraction of SNPs and setting to zero SNPs with null 

effect. De los Campos et al. (2009) modified and extended the LASSO to 

accommodate pedigree information and marker data into a single model in the 

context of QTL analysis. Waldmann et al. (2013) applied LASSO in real and 

simulated data showing that despite of small number of true QTL were detected 

none false positive was found.  

The single-step GBLUP (ssGBLUP) is another method used in GWAS 

(Misztal et al., 2009; Christensen and Lund, 2010). The main feature of this method 

is to use simultaneously all available pedigree, phenotypic and genotypic 

information. This is possible due to an H matrix that comprises information from G 

matrix (marker information) and A matrix (pedigree information) (Legarra et al., 

2009). A possible drawback for this method is that equal variances are assumed for 

the markers, which is not necessarily true regarding traits with larger QTL effects. In 

this sense, Wang et al. (2012) proposed a method in which the inclusion of all 

sources of information were possible as well as a weighting the marker according to 

their importance for the trait (wssGBLUP – weighted ssGBLUP). Indeed, this method 

has been intensively applied in GWA studies due to the low availability of genotypic 

information which may be considered insecure when using other methods (Wang et 

al., 2012; Wang et al., 2014; Silva et al., 2017). In a simulation study, Melo et al. 

(2016) observed that wssGBLUP exhibited better or similar results than Bayes Cπ in 

QTL detection. The authors also noted that the use of phenotypic information from 

non-genotyped animals, in complement to the information of genotyped animals 



7 
 

assisted the QTL detection. However, the authors compared the methods just for a 

polygenic trait with low heritability. It is unclear in which extension the additional 

phenotypic information from non-genotyped animals would improve the power of 

QTL mapping. 

The aim of this study was to evaluate, through simulation, the performance of 

different genome-wide association methods for QTL mapping, considering traits 

under different genetic structures and heritabilities. We also investigated the impact 

of additional phenotypic information from non-genotyped animals on QTL mapping 

for different scenarios. 

 

Material and Methods 

Simulated scenarios 

 The simulation was performed using the QMsim v.1.10 software (Sargolzaei 

and Schenkel, 2013). Four different scenarios (SI to SIV) were simulated being 

comprised by hypothetical traits of low (0.14) and moderate (0.35) heritabilities, 

under polygenic and major gene effects, as described below: 

SI: h2=0.14, polygenic effect; 

SII: h2=0.35, polygenic effect; 

SIII: h2=0.14, major gene effect; 

SIV: h2=0.35, major gene effect. 

To improve the inference power, 20 replicates of each scenario were 

simulated. 

Population structure 

 Following Melo et al. (2016), a historical population was simulated from 

generation zero until generation 1000, with a constant size of 1,000 animals (Figure 

1). Later, from generation 1001 until generation 2020, the population was gradually 

reduced (from 1000 to 200 animals), producing a “bottleneck” effect and, as a result, 

genetic drift and linkage disequilibrium compatible with that reported in real 

populations (O’Brien et al., 2014). The remaining 200 animals, from the last historical 

generation, were selected to expand the population. In the expansion process, it was 

assumed random union of gametes in the matings, absence of selection and 

exponential growth in the number of females, with a 100% replacement rate during 



8 
 

each generation and a mean of five animals per female. Six generations of 

expansion were generated, resulting in 16,000 animals (8,000 females). After the 

expanded phase, 240 males and 6,000 females were randomly selected and mated, 

constituting the founders of the selection population, simulated for 15 generations. At 

each generation of the selection population, males and females selected were 

randomly mated, generating a single progeny with equal probability of being male or 

female. The replacement rate for males and females were kept in 20% and the 

selection criteria was based on the expected breeding value. The 15 generations of 

selection population resulted in phenotypic information of 90,000 animals. For the 

genotypic data, 2,000 animals from the last three generations of selection population 

were randomly selected and had their genotypes used in GWAS. 

 

 

Figure 1. Population structure simulated in all scenarios.  

Genome simulation 

 It was assumed that the QTL explained 100% of the genetic variance. The 

genome had approximately 2,333 cM of length, 735,293 markers and 7,000 QTL. 

The number of marker and QTL ranged per chromosome from 12,931 to 46,495 and 

121 to 438, respectively, being randomly distributed over 29 autosomes. All markers 

were bi-allelic mimicking the commercial panels available. For the QTL, the number 

of alleles ranged randomly from two to four. A mutation rate of 10-4 for markers and 

QTL were admitted in the historical population. A total of 335,000 markers (MAF ≥ 

0.2) and 1,000 QTL were selected from the last historical generation composing the 

genotypic dataset for the selected population. The average distance among markers 



9 
 

was 0.007 cM. For the polygenic traits, the QTL effects were sampled from a gamma 

distribution with shape 0.4 (Hayes and Goddard, 2001).  

For the major gene effect traits, a polygenic trait was simulated through 

QMSim as described previously. Then, five QTL were randomly chose each one 

explaining 7% of additive genetic variance (35% major effect). This effect was 

produced in the first generation of the selection population replacing the pure 

polygenic effect by the major effect structure and then passed through all sub-

sequent generations. The remaining 995 QTL were randomly sampled using a 

gamma distribution with shape 0.4 accounting for the 65% of the variance such as in 

the polygenic scenario. These procedures were performed using specific routines 

developed in the R software, version 3.4 (2017).  

The phenotypes were the sum of QTL plus an error following normal 

distribution with mean zero and variance of 0.86 and 0.65, for the low and moderate 

heritability traits, respectively. The phenotypic variance of traits was standardised in 

one. 

GWAS analyses 

 Three different methods were compared in terms of QTL detection, namely: 

wssGBLUP (Wang et al., 2012), Bayes C (Habier et al., 2011) and Bayesian LASSO 

(Tibshirani, 1996; De los Campos et al., 2009). 

 The wssGBLUP method was adopted based on the model: y = Xβ + Zaa + e, 

where y is the vector of phenotypic observations, X is the incidence matrix that 

relates the phenotypes to the fixed effects, β is the vector of fixed effects (overall 

mean), Za is the incidence matrix that relates the animals to the phenotypes, a is the 

vector of direct additive genetic effects and e is the vector of residuals. The 

covariance between a and e was admitted as zero and their variances Ha
2 and Ie

2, 

respectively, where a
2 and e

2 are the direct additive and residual variances, 

respectively. H is the matrix that combines pedigree and genomic information 

(Aguilar et al., 2010) and I is an identity matrix. 

 The solutions for the SNP effects (û) were obtained according to VanRaden et 

al. (2009) and Stranden and Garrick (2009), as: û = DZ’[ ZDZ’ ]-1âg, where D is the 

diagonal matrix with weighting factors for the SNPs, Z is a matrix that relates the 

genotypes in each locus, and âg is the vector of breeding values predicted for the 

genotyped animals. The D matrix, SNP effects and breeding values were iteratively 



10 
 

calculated according to Wang et al. (2012). Two iteration processes (w1 and w2) 

were performed for each scenario. The w1 represented the situation that the same 

weight (w1=1) were attributed for all SNPs. For the w2 iteration, di was calculated 

according to: di = ûi
2pi(1- pi), where ûi is the allele substitution effect of the ith marker 

estimated from the previous iteration (w1), and pi is the allele frequency of the 

second allele of the ith marker. As a result, a greater shrinkage was applied in w2 for 

the SNPs explaining lower variance and, consequently, an increasing proportion of 

variance was explained by the remaining markers. 

In addition to w1 and w2 iteration processes as weighting, analyses including 

all phenotypic information available (even from non-genotyped animals) (C) and 

using only phenotypic information from genotyped animals (S) were applied. Thus, 

four different analyses using wssGBLUP were performed for each scenario. These 

analyses were performed using the BLUPF90 family programs (Misztal et al., 2012). 

 The Bayes C method was based on the model: y = 1µ + ∑i=1
ngibii + e, where 

the vectors y and e, are the vectors previously described, 1 is a vector of ones, µ is 

the overall mean, gi is the vector of genotypes for the ith SNP, bi is the allelic 

substitution effect, I is an indicator variable (0,1) sampled from a binomial 

distribution with n and π parameters, which n is the number of SNPs and π is the 

proportion of SNPs not included in the model. Two π values were tested: almost 

fixed to 0.99 (Bayes CπI) and 0.999 (Bayes CπII), assuming a prior beta distribution 

with α = 108 and β =1010 or β =1011, respectively. A scaled inverse chis-squared 

prior distribution was assumed for SNP variance effects (g
2) and residual variance 

(e
2). The Markov chain and Monte Carlo algorithm (MCMC) were applied to the 

Bayes C method performed on GS3 software (Legarra et al., 2014). A chain with 

550,000 iterations, burn-in period of 50,000, and thinning interval of 50 iterations 

were adopted for this procedure.  

The Bayesian LASSO method was implemented in a linear mixed model 

assuming an exponential prior distribution for variances of SNP marker effects. The 

LASSO prior distribution for an individual SNP (ai) follows: P(ai|2)~N(0,i
2) and 

Pr(i
2)=0.52exp(-2|i

2). This parameterization process means that individual 

variances for each SNP (i.e. i
2) are estimated, according to a regularization 

parameter (), which was estimated by using a prior gamma distribution restricted 



11 
 

between 0 and 107. The MCMC features were the same performed in the Bayes C 

method.  

 To summarize, for each scenario 7 different analyses were performed: 

wssGBLUP assuming two weights (w1 and w2), with different levels of information 

inclusion: all phenotypic information available (C) and phenotypic information only 

from genotyped animals (S); Bayes C assuming two different values for π (π=0.99 

and π=0.999) and the Bayesian LASSO, as described above. As 4 different 

scenarios and 20 replicates per scenario were simulated, a total of 560 (4x20x7) 

GWAS analyses were performed.  

Comparison criteria 

 The linkage disequilibrium decay (LD) between any two loci was surveyed 

and compared to similar study for all scenarios using the r2 parameter which is 

supplied by a QMSim feature. The following criteria was used to compare the 

methods: number of QTL that explained more than 1% of the genetic variance 

(NtopQ); number of windows (1 Mb) with greater proportion of variance explained by 

the markers (topM); sum of all genetic variance explained by the NtopQ (P_topQ) 

and superior marker windows (P_topM); maximum percentage explained by a 

topQTL (P_1stQ) and for the superior marker window (P_1stM); number of true QTL 

identified (NtrueQ), i.e. the number of topQ identified by a topM no further than 1Mb 

from the true QTL position. The P_1stTrue value indicates the number of times that a 

method was the best in 20 replicates i.e. which method(s) showed the highest ability 

to detect QTL in each replicate. In case of draw among methods, one point was 

attributed to each method, thus the sum of all methods surplus 100%. 

 

Results and Discussion 

 The average linkage disequilibrium decays over the 20 replicates for each 

scenario are shown in Figure 2. As expected, no expressive variation among the 

scenarios was found, since they were simulated through the same population 

structure. The same LD pattern was observed by Melo et al. (2016) and Pérez 

O’Brien et al. (2014) for a simulated and real population, respectively, highlighting 

the adequacy of the simulated populations. 



12 
 

Figure 3 shows how the QTL effect was distributed on the genome post-

simulation. For the polygenic trait with lower and higher heritability, SI and SII, 

respectively, it is possible to notice the stronger peaks in SII due to higher proportion 

of additive genetic variance. For SIII and SIV, we were able to produce the desired 

major effect (5 QTL summing up for 35% of total genetic variance). Often these 5 

QTL presented in the major effect structure scenarios were the only topQ (greater 

than 1% of total variance) found, however, in some replicates the “polygenic part” 

exhibited topQs as well. 

 

  

Figure 2. Linkage disequilibrium (LD) decay of all scenarios (SI to SIV) 

simulated. Average LD, expressed in r2, according to varying distances 

between markers (Mb). 

 

 As observed by Melo et al. (2016), in the polygenic scenario under low 

heritability, independent of the statistical method adopted, the QTL detection was 

poor. On average, 17.55 QTL were simulated over the 20 replicates and only 2.6 

(14.8%) to 3.75 (21.4%) QTL were detected, depending on the method adopted 

(Table1). This result is in agreement with Van den Berg et al. (2013) who highlighted 

the difficulty of QTL mapping for low heritable polygenic traits. 

 Although the comparison among methods was difficult due to high standard 

deviations on the means, the relative number of best detection (P_1stTrue) helped to 



13 
 

show that CW1 model presented better results for QTL mapping as the best method 

in 45% of replicates. The slight superiority of CW1 may be explained by the 

wssGBLUP model premises which fit to an infinitesimal model, i.e., many markers 

presenting small effects. 

 The effect of additional phenotypic information helped slightly the QTL 

detection. For CW1, on average, the use of this extra information, allowed the 

detection of 21.36% (3.75 out of 17.55) NtopQ, compared to 15.10% (2.65 out of 

17.55) when this information was ignored (SW1). These results are in agreement 

with Melo et al. (2016) with 17.36% and 11.98% also for w1, for inclusion and non-

inclusion of additional information, respectively. The same pattern was observed for 

w2 in this scenario. 

 

 

Figure 3. Examples of QTL variance and distribution for random replicates 

simulated for all scenarios present in the last selection generation. Position in 

the x-axis refers to position of the QTL in cM irrespective of the chromosome. 



14 
 

 Using a stronger shrinkage on Bayes C (Bayes CπII) resulted in greater 

proportion of the genetic variance being captured by the leading SNPs (P_topM). 

However, this did not reflect in greater power to QTL detection, when compared to a 

weaker shrinkage (Bayes CπI). Similarly, using w2 in the wssGBLUP method 

increased the genetic variance captured by the leading SNPs, but the mean NtrueQ 

were similar between w1 and w2 or superior for w1 (Table 1: contrasts SW1xSW2 

and CW1xCW2).  

 

Table 1. Mean (SD) for QTL and marker statistics1 using Bayes C, Bayesian 

LASSO and weighted single step GBLUP for SI (h2=0.14, polygenic effect), over 

20 replicates. 

Method2 P_topM (%)  P_1stM (%)  NtrueQ P_1stTrue (%)3 

Bayes CπI 7.26 (1.09) 0.87 (0.26) 2.95 (2.01) 15.00% 

Bayes CπII 37.59 (12.56) 9.96 (7.80) 2.95 (2.35) 15.00% 

LASSO 5.25 (0.63) 0.50 (0.09) 2.60 (1.76) 20.00% 

CW1 5.17 (0.06) 0.50 (0.09) 3.75 (1.74) 45.00% 

CW2 16.84 (2.48) 2.09 (0.58) 3.00 (2.00) 30.00% 

SW1 5.42 (0.68) 0.52 (0.10) 2.65 (1.60) 10.00% 

SW2 21.23 (4.43) 3.22 (1.53) 2.85 (1.95) 10.00% 

True Values P_topQ (%)  P_1stQ (%)  NtopQ  

31.30 (3.55) 3.96 (1.42) 17.55 (2.01)  

1 Pvar_topMRKw (P_topQ): sum of all genetic variance explained by the NtopQ (P_topQ) 
and superior marker windows (P_topM); 
P_1stM (P_1stQ): maximum percentage explained by a topQ (P_1stQ) and for the superior 
marker window (P_1stM); 
NtrueQ (NtopQ): number of true QTL identified (NtrueQ) and number of QTL that 
explained more than 1% of the genetic variance (NtopQ); 

2 Bayes CπI (Bayes CπII): Bayes C assuming two different values for π, π=0.99, 
(BayesCπI) and π=0.999, (Bayes CπII); 
LASSO: Bayesian LASSO; 
CW1 (CW2): wssGBLUP with the inclusion of all phenotypic information, including from 
non genotyped animals and applying different weights w1 (CW1) and w2 (CW2); 
SW1 (SW2): wssGBLUP with the inclusion of phenotypic information only from genotyped 
animals and applying different weights w1 (SW1) and w2 (SW2); 

3 P_1stTrue: number of times that each method showed the highest NtrueQ. 

 

 For SII, the number of NtopQ, proportion of maximum variance explained by a 

single QTL and sum of topQ variances were very similar to SI: 30.07%, 3.83% and 

17.25, respectively (Table 2). It is also possible to observe a greater power of QTL 

detection for all methods due to heritability increase, compared to SI. For SII, the 



15 
 

QTL detection (NtrueQ) ranged from 31.20% (5.40 out of 17.25) to 36.2% (6.25 out 

of 17.25). The influence of heritability on QTL detection was also highlighted by Van 

den Berg et al. (2013), where the number of false positives decreased as the 

heritability increased. 

 

Table 2. Mean (SD) for QTL and marker statistics1 using Bayes C, Bayesian 

LASSO and weighted single step GBLUP for SII (h2=0.35, polygenic effect), 

over 20 replicates. 

Method2 P_topM (%)  P_1stM (%)  NtrueQ P_1stTrue (%)3 

Bayes CπI 12.90 (3.87) 2.32 (2.58) 6.11 (3.74) 40.00% 

Bayes CπII 66.57 (7.93) 22.16 (10.63) 6.25 (3.04) 30.00% 

LASSO 5.50 (0.92) 0.57 (0.16) 5.80 (3.37) 20.00% 

CW1 5.51 (1.00) 0.57 (0.20) 6.20 (2.97) 25.00% 

CW2 16.72 (4.82) 2.61 (1.38) 5.45 (2.35) 15.00% 

SW1 5.53 (0.93) 0.57 (0.16) 5.65 (3.39) 20.00% 

SW2 17.58 (2.66) 2.55 (1.41) 5.40 (2.95) 0.00% 

True 

Values 

P_topQ (%)  P_1stQ (%)  NtopQ  

30.07 (5.68) 3.83 (1.13) 17.25 (3.70)  

1 Pvar_topMRKw (P_topQ): sum of all genetic variance explained by the NtopQ (P_topQ) 
and superior marker windows (P_topM); 
P_1stM (P_1stQ): maximum percentage explained by a topQ (P_1stQ) and for the superior 
marker window (P_1stM); 
NtrueQ (NtopQ): number of true QTL identified (NtrueQ) and number of QTL that 
explained more than 1% of the genetic variance (NtopQ); 

2 Bayes CπI (Bayes CπII): Bayes C assuming two different values for π, π=0.99, 
(BayesCπI) and π=0.999, (Bayes CπII); 
LASSO: Bayesian LASSO; 
CW1 (CW2): wssGBLUP with the inclusion of all phenotypic information, including from 
non genotyped animals and applying different weights w1 (CW1) and w2 (CW2); 
SW1 (SW2): wssGBLUP with the inclusion of phenotypic information only from genotyped 
animals and applying different weights w1 (SW1) and w2 (SW2); 

3 P_1stTrue: number of times that each method showed the highest NtrueQ. 

 

 Again, Bayes CπII, CW2 and SW2 captured greater part of genetic variance 

in contrast to the other methods. However, Bayes CπI and Bayes CπII sharply 

increased this amount of genetic variance captured (56.28 and 56.47%, respectively, 

more in comparison to the same methods on SI). For CW2 and SW2 the shrinkage 

did not produce the same effect on the variance captured and it remained practically 

constant in relation to SI statistics. 



16 
 

 In comparison to SI, there was more similarity among the methods regarding 

NtrueQ. The benefit of additional phenotypic information in wssGBLUP on QTL 

detection seemed less pronounced. For w1 iteration, 35.94% (6.20 out of 17.25) 

NtopQ were detected (CW1) while, ignoring this information (SW1) this number 

reduced only to 32.75% (5.65 out of 17.25). It most likely happened because of the 

additional phenotype information from related animals has less impact on the 

accuracy of estimated breeding values, and as a result, less importance on the 

estimatives of SNP effects, when compared to a lower heritability trait. Similar 

tendency was found for w2 iteration with 5.45 and 5.40 for CW2 and SW2, 

respectively.   

For SIII (Table 3) the results suggested that all methods evaluated had 

superior power for QTL detection when the trait presents major QTL effect. On 

average, NtrueQ detected ranged from 31.1% (1.65 out of 5.30) to 53.8% (2.85 out 

of 5.30). This result was expected due to the major effect simulated (5 QTL 

explaining 7% of total genetic variance each) concentrating the effect and favouring 

all methods in the estimation of SNP effects. 

 For this scenario, the Bayes C method showed results slight superior than the 

other methods as showed by 75% and 60% of best detection (P_1stTrue) for Bayes 

CπI and Bayes CπII, respectively. Most likely the assumptions of Bayes C method 

were better fitted to the genetic structure simulated. Mehrban et al. (2017) comparing 

methods for genomic selection found that Bayes C exhibits more accuracy for 

carcass weight in Hanwoo beef cattle, a possible indicative of major gene effects. 

This had been supported by Lee et al. (2013) through an association study detecting 

6 SNP loci for carcass weight with some of these markers accounting for more than 

10% of total genetic variance. As for the other scenarios, the increase on shrinkage 

(Bayes CπII x Bayes CπI) was not associated to a better QTL detection. This 

probably happened due to proportionally few genotypes and low heritability trait on 

this scenario.   

 Distinct from the polygenic scenarios, the additional phenotypic information 

tended to reduce the power of QTL detection in SIII (Table 3: contrasts CW1xSW1 

and CW2xSW2, for the different levels of information included in the wssGBLUP 

method). Most likely, the extra phenotypic information compromised the QTL 

detection by introducing “noise” on wssGBLUP estimations. Close relatives not 

necessarily present the same genotype regarding the major QTL configuration, and 



17 
 

using their phenotypic information without knowing their genotypes may hamper QTL 

mapping.  

 Although, exactly 5 major QTL with 7% of total genetic variance were 

simulated, on average 5.3 QTL were found over 20 replicates. The surplus 0.3 is 

associated to others QTL that in determined replicates explained more than 1% of 

the genetic variance being accounted for the total variance explained by all QTL. The 

milder reduction on the proportion of total genetic variance explained by the major 

QTL (6.95 instead of 7.00) may indicate that the polygenic part also exhibited 

NtopQ(5.30, instead of exactly 5) along the 20 replicates. 

 

Table 3. Mean (SD) for QTL and marker statistics1 using Bayes C, Bayesian 

LASSO and weighted single step GBLUP for SIII (h2=0.14, major gene effect), 

over 20 replicates. 

Method2 P_topM (%)  P_1stM (%)  NtrueQ P_1stTrue (%)3 

Bayes CπI 5.49 (1.60) 2.00 (1.01) 2.85(1.01) 75.00 

Bayes CπII 51.07 (16.66) 29.47 (16.72) 2.65(0.99) 60.00 

LASSO 2.75 (0.57) 0.73 (0.24) 2.26(1.24) 30.00 

CW1 2.26 (0.62) 0.59 (0.28) 1.90(0.41) 25.00 

CW2 9.85 (2.70) 3.03 (1.16) 1.65(1.18) 15.00 

SW1 3.48 (2.89) 1.06 (1.42) 2.40(1.27) 40.00 

SW2 16.69 (5.97) 6.58 (3.81) 2.45(1.10) 35.00 

True 

Values 

P_topQ (%)  P_1stQ (%)  NtopQ  

35.08 (1.39) 6.95 (0.29) 5.30 (0.66)  

1 Pvar_topMRKw (P_topQ): sum of all genetic variance explained by the NtopQ (P_topQ) 
and superior marker windows (P_topM); 
P_1stM (P_1stQ): maximum percentage explained by a topQ (P_1stQ) and for the superior 
marker window (P_1stM); 
NtrueQ (NtopQ): number of true QTL identified (NtrueQ) and number of QTL that 
explained more than 1% of the genetic variance (NtopQ); 

2 Bayes CπI (Bayes CπII): Bayes C assuming two different values for π, π=0.99, 
(BayesCπI) and π=0.999, (Bayes CπII); 
LASSO: Bayesian LASSO; 
CW1 (CW2): wssGBLUP with the inclusion of all phenotypic information, including from 
non genotyped animals and applying different weights w1 (CW1) and w2 (CW2); 
SW1 (SW2): wssGBLUP with the inclusion of phenotypic information only from genotyped 
animals and applying different weights w1 (SW1) and w2 (SW2); 

3 P_1stTrue: number of times that each method showed the highest NtrueQ. 

 

 



18 
 

 The results on Table 4 shows that all methods evaluated outperformed the 

QTL detection when the trait is under major effect and higher heritability. Over all 

replicates, the extension of QTL detection was from 40.37% (2.20 out of 5.45) to 

72.48% (3.95 out of 5.45). This result was also expected due to the increase on 

additive genetic variance.  

 The Bayes C method showed greater power to QTL detection compared to 

the other methods. Again, the genetic structure may have favoured the variance of 

SNPs captured by this method. However, in scenario SIV, the difference between 

stronger and weak shrinkage, regarding the variance captured by top markers, 

seemed to be less pronounced than in SIII. 

 

Table 4. Mean (SD) for QTL and marker statistics using Bayes C, Bayesian 

LASSO and weighted single step GBLUP for SIV (h2=0.35, major gene effect), 

over 20 replicates. 

Method P_topM(%)  P_1stM(%)  NtrueQ P_1stTrue (%)3 

Bayes CπI 19.52 (12.18) 10.37 (9.68) 3.90 (1.37) 70.00 

Bayes CπII 69.76 (11.06) 37.40 (16.32) 3.80 (1.06) 55.00 

LASSO 3.25 (1.11) 0.84 (0.35) 3.00 (1.63) 30.00 

CW1 2.35 (0.72) 0.57 (0.14) 2.20 (1.61) 10.00 

CW2 9.18 (4.18) 3.01 (1.57) 2.35 (1.73) 25.00 

SW1 3.42 (0.87) 0.88 (0.25) 3.45 (1.50) 25.00 

SW2 16.13 (4.59) 6.24 (3.13) 3.95 (1.05) 40.00 

True 

Values 

P_topQ(%)  P_1stQ (%)  NtopQ  

35.87 (1.88) 7.07 (0.31) 5.45 (0.69)  

1 Pvar_topMRKw (P_topQ): sum of all genetic variance explained by the NtopQ (P_topQ) 
and superior marker windows (P_topM); 
P_1stM (P_1stQ): maximum percentage explained by a topQ (P_1stQ) and for the superior 
marker window (P_1stM); 
NtrueQ (NtopQ): number of true QTL identified (NtrueQ) and number of QTL that 
explained more than 1% of the genetic variance (NtopQ); 

2 Bayes CπI (Bayes CπII): Bayes C assuming two different values for π, π=0.99, 
(BayesCπI) and π=0.999, (Bayes CπII); 
LASSO: Bayesian LASSO; 
CW1 (CW2): wssGBLUP with the inclusion of all phenotypic information, including from 
non genotyped animals and applying different weights w1 (CW1) and w2 (CW2); 
SW1 (SW2): wssGBLUP with the inclusion of phenotypic information only from genotyped 
animals and applying different weights w1 (SW1) and w2 (SW2); 

3 P_1stTrue: number of times that each method showed the highest NtrueQ. 

 



19 
 

 As for SIII, the use of additional phenotypic information compromised 

wssGBLUP performance in terms of QTL detection. Again, this result is related to the 

fact that the major QTL effect may vary in the relatives included as extra information. 

Both w2 iterations (CW2 x SW2) showed more power in the detection of QTL, 

compared to w1. These results are in agreement with Zhang et al. (2016) study, in 

which, the use of different weights for SNPs is more effective for traits influenced by 

few QTL than in polygenic structures resulting in more power in QTL detection, .  

 The LASSO method presented steady results over all the scenarios in 

comparison to other methods. However, the influence of increase on the heritability 

and the shift to the major effect structure resulted in increase on QTL mapping, same 

pattern observed for the other methods. Its intermediary performance may be 

explained by the LASSO assumption that this method cannot select more predictor 

variables than the sample size. This could potentially be a problem in our study that 

involves several more predictor variables than response variables. However, some 

studies state that the double exponential prior used by LASSO may present a 

remarkable advantage by modelling multiple markers at same time. This method is 

able to distinct trait supplier loci from others that are in high linkage disequilibrium 

with those loci. (Papachristou et al., 2016; Motyer et al., 2011). 

 The proportion of times that each method was the best (P_1stTrue) 

summarizes the results obtained on true QTL detection means highlighting: the 

benefit of using additional phenotypic information (CW1) on scenarios under 

complex traits (SI); the similarity among methods when the heritability is increased 

under polygenic structures (SII); the remarkable performance of Bayes C method, 

mainly with π almost fixed to 0.99, when the trait is under major structures (SIII and 

SIV). The most suitable method for each scenario relies on further studies regarding, 

for instance, other genetic structures, varying number of genotyped animals and 

different population structures. By investigating the methods’ strength and weakness, 

the QTL mapping will become more accurate and helpful. 

 

Conclusions 

 The most suitable method to perform GWAS relies on the genetic structure 

and trait’s heritability.  



20 
 

 Despite the small differences among methods in the QTL detection, for 

polygenic traits, the wssGBLUP method seemed to show better results in 

comparison to the other methods mainly in the low heritability scenario. 

 For traits with major gene effects, the Bayes C method is expected to present 

better results, compared to the other methods evaluated. 

 The use of additional phenotypic information from non genotyped animals 

implies in better QTL detection in polygenic traits. However, for traits under major 

gene effect, it may negatively affect the QTL detection.  

References 

Aguilar, I., Misztal, I., Johnson, D. L., Legarra, A., Tsuruta, S., Lawlor, T. J. (2010). 

Hot topic: A unified approach to utilize phenotypic, full pedigree, and genomic 

information for genetic evaluation of Holstein final score. Journal of dairy science, 

93(2), 743-752. 

Bourdon, R.M. (1999) Understanding animal breeding. 2nd ed. Englewood Cliffs, NJ: 

Prentice Hall. 538p. 

Chen, C. Y., Misztal, I., Aguilar, I., Legarra, A., Muir, W. M. (2011). Effect of different 

genomic relationship matrices on accuracy and scale. Journal of animal science, 

89(9), 2673-2679. 

Christensen, O. F., Lund, M. S. (2010). Genomic prediction when some animals are 

not genotyped. Genetics Selection Evolution, 42(1), 2. 

De Los Campos, G., Naya, H., Gianola, D., Crossa, J., Legarra, A., Manfredi, E., 

Weigel, K., Cotes, J. M. (2009). Predicting quantitative traits with regression models 

for dense molecular markers and pedigree. Genetics, 182(1), 375-385. 

Dikmen, S., Cole, J. B., Null, D. J., Hansen, P. J. (2013). Genome-wide association 

mapping for identification of quantitative trait loci for rectal temperature during heat 

stress in Holstein cattle. PLoS One, 8(7). 

Fernando, R. L., Garrick, D. (2013). Bayesian methods applied to GWAS. Genome-

wide association studies and genomic prediction, 237-274. 

Forni, S., Aguilar, I., Misztal, I. (2011). Different genomic relationship matrices for 

single-step analysis using phenotypic, pedigree and genomic information. Genetics 

Selection Evolution, 43(1), 1. 



21 
 

Fragomeni, B. O., Misztal, I., Lourenco, D. L., Aguilar, I., Okimoto, R.,  Muir, W. M. 

(2014). Changes in variance explained by top SNP windows over generations for 

three traits in broiler chicken. Frontiers in genetics, 5. 

Gianola D., De Los Campos G., Hill W.G., Manfredi E., Fernando R.L. 

(2009).Additive genetic variability and the Bayesian alphabet. Genetics, 183: 347–

363. 

Goddard, M. E., Hayes, B. J. (2009). Mapping genes for complex traits in domestic 

animals and their use in breeding programmes. Nature Reviews Genetics, 10(6), 

381-391. 

Habier D., Fernando R.L., Kizilkaya K., Garrick, D.J. (2011).Extension of the 

Bayesian alphabet for genomic selection. BMC Bioinformatics, 12:186. 

Hayes, B. E. N., Goddard, M. E. (2001). The distribution of the effects of genes 

affecting quantitative traits in livestock.Genetics Selection Evolution, 33(3), 1. 

Hayes, B. J., Bowman, P. J., Chamberlain, A. J., Goddard, M. E. (2009). Invited 

review: Genomic selection in dairy cattle: Progress and challenges. Journal of dairy 

science, 92(2), 433-443. 

Lee, S. H., Choi, B. H., Lim, D., Gondro, C., Cho, Y. M., Dang, C. G., Sharma, A., 

Jang, G. W., Lee, K. T., Yoon, D., Lee, H. K., Yeon, S. H., Yang, B. S., Kang, H. S., 

Hong, S. K. (2013). Genome-wide association study identifies major loci for carcass 

weight on BTA14 in Hanwoo (Korean cattle). PLoS One, 8(10). 

Legarra, A., Aguilar, I.,Misztal, I. (2009). A relationship matrix including full pedigree 

and genomic information. Journal of Dairy Science, v. 92, n. 9, p. 4656-4663. 

Legarra, A.; Ricard, A.; Filangi, O. GS3. Manual. France. [S.I.: s.n.], 2014, 24p. 

Available in: <http://snp.toulouse.inra.fr/~alegarra/manualgs3_last.pdf>. Accessed in 

Oct. 2017. 

Lourenco, D. A. L., Tsuruta, S., Fragomeni, B. O., Masuda, Y., Aguilar, I., Legarra, 

A., Bertrand, J. K., Amen, T. S., Wang, L., Moser, D. W.,  Misztal, I. (2015). Genetic 

evaluation using single-step genomic best linear unbiased predictor in American 

Angus. Journal of animal science, 93(6), 2653-2662. 

Mehrban, H., Lee, D. H., Moradi, M. H., IlCho, C., Naserkheil, M., Ibáñez-Escriche, 

N. (2017). Predictive performance of genomic selection methods for carcass traits in 

Hanwoo beef cattle: impacts of the genetic architecture. Genetics Selection 

Evolution, 49(1), 1. 



22 
 

Melo, T. P., Takada, L., Baldi, F., Oliveira, H. N., Dias, M. M., Neves, H. H., 

Schenkel F.S., Albuquerque, L.G., Carvalheiro, R. (2016). Assessing the value of 

phenotypic information from non-genotyped animals for QTL mapping of complex 

traits in real and simulated populations.BMC genetics, 17(1), 89. 

Meuwissen, T. H. E., Hayes, B. J., Goddard, M. E. (2001) Prediction of total genetic 

value using genome-wide dense marker maps. Genetics, v. 157, n. 4, p. 1819-1829. 

Misztal, I. BLUPF90-a flexible mixed model program in Fortran 90. User Manual. 

Animal and Dairy Science. Athens, GA, USA: University of Georgia; [S.I.: s.n.], 2012. 

Available in: http://nce.ads.uga.edu/wiki/lib/exe/fetch.php?media=blupf90.pdf . 

Accessed in: nov. 2017.  

Misztal, I., Legarra, A., Aguilar, I. (2009). Computing procedures for genetic 

evaluation including phenotypic, full pedigree, and genomic information. Journal of 

Dairy Science, 92(9), 4648-4655. 

Motyer, A. J., McKendry, C., Galbraith, S., Wilson, S. R. (2011). LASSO model 

selection with post-processing for a genome-wide association study data set. In BMC 

proceedings (Vol. 5, No. 9, p. S24). BioMed Central. 

O’Brien, A. M. P., Mészáros, G., Utsunomiya, Y. T., Sonstegard, T. S., Garcia, J. F., 

Van Tassell, C. P., Carvalheiro, R., Silva, M. V. B. Sölkner, J. (2014). Linkage 

disequilibrium levels in Bosindicus and Bostaurus cattle using medium and high 

density SNP chip data and different minor allele frequency distributions. Livestock 

Science, 166, 121-132. 

Papachristou, C., Ober, C., Abney, M. (2016). A LASSO penalized regression 

approach for genome-wide association analyses using related individuals: 

application to the Genetic Analysis Workshop 19 simulated data. In BMC 

proceedings (Vol. 10, No. 7, p. 53). BioMed Central. 

R Core Team (2017). R: A language and environment for statistical computing. R 

Foundation for Statistical Computing, Vienna, Austria.URL https://www.R-

project.org/. 

Sargolzaei, M., Schenkel, F. S. (2013).QMSim: User’s Guide [S.I.: s.n.], p.77. 

Silva, R. M. O., Vallejo, R. L., Evenhuis, J. P., Leeds, T. D., Gao, G., Parsons, J. E., 

Martin, K. E., Lourenco, D. A. L., Palti, Y. (2017). 209 Prospecting genomic regions 

associated with columnaris disease in two rainbow trout breeding populations. 

Journal of Animal Science, 95(supplement 4), 103-104. 



23 
 

Stranden, I., Garrick, D. J. (2009). Derivation of equivalent computing algorithms for 

genomic predictions and reliabilities of animal merit. Journal of dairy science, 92(6), 

2971-2975. 

Tibshirani, R. (1996). Regression shrinkage and selection via the lasso. Journal of 

the Royal Statistical Society.Series B (Methodological), 267-288. 

Tiezzi, F., Parker-Gaddis, K. L., Cole, J. B., Clay, J. S., Maltecca, C. (2015). A 

genome-wide association study for clinical mastitis in first parity US Holstein cows 

using single-step approach and genomic matrix re-weighting procedure. PLoS One, 

10(2). 

Van den Berg, I., Fritz, S., Boichard, D. (2013). QTL fine mapping with Bayes C (π): 

a simulation study. Genetics Selection Evolution, 45(1), 19. 

VanRaden, P. M. (2012). Avoiding bias from genomic pre-selection in converting 

daughter information across countries. In: INTERNATIONAL BULL EVALUATION 

SERVICE, 2., n. 45, p. 1-5.Verona, Italy.  

VanRaden, P. M., Van Tassell, C. P., Wiggans, G. R., Sonstegard, T. S., Schnabel, 

R. D., Taylor, J. F., Schenkel, F. S. (2009). Invited review: Reliability of genomic 

predictions for North American Holstein bulls. Journal of dairy science, 92(1), 16-24. 

Vitezica, Z. G., Aguilar, I., Misztal, I., Legarra, A. (2011). Bias in genomic predictions 

for populations under selection. Genetics Research, 93(5), 357-366. 

Waldmann, P., Mészáros, G., Gredler, B., Fuerst, C., Sölkner, J. (2013). Evaluation 

of the lasso and the elastic net in genome-wide association studies. Frontiers in 

genetics, 4. 

Wang, H., Misztal, I., Aguilar, I., Legarra, A., Fernando, R. L., Vitezica, Z., Okimoto, 

R., Wing, T., Hawken, R., Muir, W. M. (2014). Genome-wide association mapping 

including phenotypes from relatives without genotypes in a single-step (ssGWAS) for 

6-week body weight in broiler chickens. Frontiers in genetics, 5. 

Wang, H., Misztal, I., Aguilar, I., Legarra, A., Muir, W. M. (2012). Genome-wide 

association mapping including phenotypes from relatives without genotypes. 

Genetics Research, 94(2), 73-83. 

Zhang, X., Lourenco, D., Aguilar, I., Legarra, A., Misztal, I. (2016). Weighting 

strategies for single-step genomic BLUP: an iterative approach for accurate 

calculation of GEBV and GWAS. Frontiers in genetics, 7.