Using multiple regression, Bayesian networks and artificial neural networks for prediction of total egg production in European quails based on earlier expressed phenotypes

Videla Rodriguez E, Smith VA, van den Brand H, Rodenburg TB, Visser B. A two-step Bayesian network approach to identify key SNPs associated to multiple phenotypic traits in four purebred laying hen lines

When purebred laying hen chicks hatch, they remain at a rearing farm until approximately 17 weeks of age, after which they are transferred to a laying farm. Chicks or pullets are removed from the flocks during these 17 weeks if they display any rearing abnormality. The aim of this study was to investigate associations between single nucleotide polymorphisms (SNPs) and rearing success of 4 purebred White Leghorns layer lines by implementing a Bayesian network approach. Phenotypic traits and SNPs of four purebred genetic White Leghorn layer lines were available for 23,000 rearing batches obtained between 2010 and 2020. Associations between incubation traits (clutch size, embryo mortality), rearing traits (genetic line, first week mortality, rearing abnormalities, natural death, rearing success, pullet flock age, and season) and SNPs were analyzed, using a two-step Bayesian Network (BN) approach. Furthermore, the SNPs were connected to their corresponding genes, which were further explored in bioinformatics databases. BN analysis revealed a total of 28 SNPs associated with some of the traits: ten SNPs were associated with clutch size, another 10 with rearing abnormalities, a single SNP with natural death, and seven SNPs with first week mortality. Exploration via bioinformatics databases showed that one of the SNPs (ENAH) had a protein predicted network composed of 11 other proteins. The major hub of this SNP was CDC42 protein, which has a role in egg production and reproduction. The results highlight the power of BNs in knowledge discovery and how their application in complex biological systems can help getting a deeper understanding of functionality underlying genetic variation of rearing success in laying hens. Improved welfare and production might result from the identified SNPs. Selecting for these SNPs through breeding could reduce stress and increase livability during rearing.

Review: When worlds collide - poultry modeling in the 'Big Data' era

Within poultry production systems, models have provided vital decision support, opportunity analysis, and performance optimization capabilities to nutritionists and producers for decades. In recent years, due to the advancement of digital and sensor technologies, 'Big Data' streams have emerged, optimally positioned to be analyzed by machine-learning (ML) modeling approaches, with strengths in forecasting and prediction. This review explores the evolution of empirical and mechanistic models in poultry production systems, and how these models may interact with new digital tools and technologies. This review will also examine the emergence of ML and Big Data in the poultry production sector, and the emergence of precision feeding and automation of poultry production systems. There are several promising directions for the field, including: (1) application of Big Data analytics (e.g., sensor-based technologies, precision feeding systems) and ML methodologies (e.g., unsupervised and supervised learning algorithms) to feed more precisely to production targets given a 'known' individual animal, and (2) combination and hybridization of data-driven and mechanistic modeling approaches to bridge decision support with improved forecasting capabilities.

Genomic prediction for meat and carcass traits in Nellore cattle using a Markov blanket algorithm

This study was carried out to evaluate the advantage of preselecting SNP markers using Markov blanket algorithm regarding the accuracy of genomic prediction for carcass and meat quality traits in Nellore cattle. This study considered 3675, 3680, 3660 and 524 records of rib eye area (REA), back fat thickness (BF), rump fat (RF), and Warner-Bratzler shear force (WBSF), respectively, from the Nellore Brazil Breeding Program. The animals have been genotyped using low-density SNP panel (30 k), and subsequently imputed for arrays with 777 k SNPs. Four Bayesian specifications of genomic regression models, namely Bayes A, Bayes B, Bayes Cπ and Bayesian Ridge Regression methods were compared in terms of prediction accuracy using a five folds cross-validation. Prediction accuracy for REA, BF and RF was all similar using the Bayesian Alphabet models, ranging from 0.75 to 0.95. For WBSF, the predictive ability was higher using Bayes B (0.47) than other methods (0.39 to 0.42). Although the prediction accuracies using Markov blanket of SNP markers were lower than those using all SNPs, for WBSF the relative gain was lower than 13%. With a subset of informative SNPs markers, identified using Markov blanket, probably, is possible to capture a large proportion of the genetic variance for WBSF. The development of low-density and customized arrays using Markov blanket might be cost-effective to perform a genomic selection for this trait, increasing the number of evaluated animals, improving the management decisions based on genomic information and applying genomic selection on a large scale.

In vivo prediction of abdominal fat and breast muscle in broiler chicken using live body measurements based on machine learning

The purpose of this study was to predict the carcass characteristics of broilers using support vector regression (SVR) and artificial neural network (ANN) model methods. Data were obtained from 176 yellow feather broilers aged 100-day-old (90 males and 86 females). The input variables were live body measurements, including external measurements and B-ultrasound measurements. The predictors of the model were the weight of abdominal fat and breast muscle in male and female broilers, respectively. After descriptive statistics and correlation analysis, the datasets were randomly divided into train set and test set according to the ratio of 7:3 to establish the model. The results of this study demonstrated that it is feasible to use machine learning methods to predict carcass characteristics of broilers based on live body measurements. Compared with the ANN method, the SVR method achieved better prediction results, for predicting breast muscle (male: R² = 0.950; female: R² = 0.955) and abdominal fat (male: R² = 0.802; female: R² = 0.944) in the test set. Consequently, the SVR method can be considered to predict breast muscle and abdominal fat of broiler chickens, except for abdominal fat in male broilers. However, further revaluation of the SVR method is suggested.

Prediction of dry matter intake and gross feed efficiency using milk production and live weight in first-parity Holstein cows

Direct measurement of dry matter intake (DMI) presents a major challenge in estimating gross feed efficiency (GFE) in dairy cattle. This challenge can, however, be resolved through the prediction of DMI and GFE from easy-to-measure traits such as milk production (i.e. milk yield, energy-corrected milk (ECM), butterfat, protein, lactose) and live weight (LW). The main objective of this study was, therefore, to investigate the feasibility of predicting dry matter intake and gross feed efficiency for first-parity Holstein cows using milk production traits and LW. Data comprised of 30 daily measurements of DMI and milk production traits, and 25 daily LW records of a group of 100 first-parity Holstein cows, fed a total mixed ration. Gross feed efficiency was calculated as kg ECM divided by kg DMI. The initial step was to estimate correlations of milk production traits and LW with DMI and GFE, to identify the best potential predictors of DMI and GFE. Subsequently, a forward stepwise regression analysis was used to develop models to predict DMI and GFE from LW and milk production traits, followed by within-herd validations. Means for DMI, butterfat yield (BFY) and LW were 21.91 ± 2.77 kg/day, 0.95 ± 0.14 kg/day and 572 ± 15.58 kg/day, respectively. Mean GFE was 1.32 ± 0.22. Dry matter intake had positive correlations with milk yield (MY) (r = 0.32, p < 0.001) and LW (r = 0.76, p < 0.0001) and an antagonistic association with butterfat percent (BFP) (r =  - 0.55, p < 0.001). On the other hand, GFE was positively associated with MY (r = 0.36, p < 0.001), BFP (r = 0.53, p < 0.001) and BFY (r = 0.83, p < 0.0001), and negatively correlated with LW (r =  - 0.23, p > 0.05). Dry matter intake was predicted reliably by a model comprising of only LW and MY (R² = 0.79; root mean squared error (RMSE) = 1.05 kg/day). A model that included BFY, MY and LW had the highest ability to predict GFE (R² = 0.98; RMSE = 0.05). Live weight and BFY were the main predictor traits for DMI and GFE, respectively. The best models for predicting DMI and GFE were as follows: DMI (kg/day) =  - 54.21 - 0.192 × MY (kg/day) + 0.146 × LW (kg/day) and GFE (kg/day) = 4.120 + 0.024 × MY (kg/day) + 1.000 × BFY (kg/day) - 0.008 × LW (kg/day). Thus, daily DMI (kg/day) and GFE can be reliably predicted from LW and milk production traits using these developed models in first-parity Holstein cows. This presents a big promise to generate large quantities of data of individual cow DMI and GFE, which can be used to implement genetic improvement of feed efficiency.

Can machine learning algorithms perform better than multiple linear regression in predicting nitrogen excretion from lactating dairy cows

This study aims to compare the performance of multiple linear regression and machine learning algorithms for predicting manure nitrogen excretion in lactating dairy cows, and to develop new machine learning prediction models for MN excretion. Dataset used were collated from 43 total diet digestibility studies with 951 lactating dairy cows. Prediction models for MN were developed and evaluated using MLR technique and three machine learning algorithms, artificial neural networks, random forest regression and support vector regression. The ANN model produced a lower RMSE and a higher CCC, compared to the MLR, RFR and SVR model, in the tenfold cross validation. Meanwhile, a hybrid knowledge-based and data-driven approach was developed and implemented to selecting features in this study. Results showed that the performance of ANN models were greatly improved by the turning process of selection of features and learning algorithms. The proposed new ANN models for prediction of MN were developed using nitrogen intake as the primary predictor. Alternative models were also developed based on live weight and milk yield for use in the condition where nitrogen intake data are not available (e.g., in some commercial farms). These new models provide benchmark information for prediction and mitigation of nitrogen excretion under typical dairy production conditions managed within grassland-based dairy systems.

Practical application of a Bayesian network approach to poultry epigenetics and stress

Background

Relationships among genetic or epigenetic features can be explored by learning probabilistic networks and unravelling the dependencies among a set of given genetic/epigenetic features. Bayesian networks (BNs) consist of nodes that represent the variables and arcs that represent the probabilistic relationships between the variables. However, practical guidance on how to make choices among the wide array of possibilities in Bayesian network analysis is limited. Our study aimed to apply a BN approach, while clearly laying out our analysis choices as an example for future researchers, in order to provide further insights into the relationships among epigenetic features and a stressful condition in chickens (Gallus gallus).

Results

Chickens raised under control conditions (n = 22) and chickens exposed to a social isolation protocol (n = 24) were used to identify differentially methylated regions (DMRs). A total of 60 DMRs were selected by a threshold, after bioinformatic pre-processing and analysis. The treatment was included as a binary variable (control = 0; stress = 1). Thereafter, a BN approach was applied: initially, a pre-filtering test was used for identifying pairs of features that must not be included in the process of learning the structure of the network; then, the average probability values for each arc of being part of the network were calculated; and finally, the arcs that were part of the consensus network were selected. The structure of the BN consisted of 47 out of 61 features (60 DMRs and the stressful condition), displaying 43 functional relationships. The stress condition was connected to two DMRs, one of them playing a role in tight and adhesive intracellular junctions in organs such as ovary, intestine, and brain.

Conclusions

We clearly explain our steps in making each analysis choice, from discrete BN models to final generation of a consensus network from multiple model averaging searches. The epigenetic BN unravelled functional relationships among the DMRs, as well as epigenetic features in close association with the stressful condition the chickens were exposed to. The DMRs interacting with the stress condition could be further explored in future studies as possible biomarkers of stress in poultry species.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12859-022-04800-0.

A Bayesian network structure learning approach to identify genes associated with stress in spleens of chickens

Differences in the expression patterns of genes have been used to measure the effects of non-stress or stress conditions in poultry species. However, the list of genes identified can be extensive and they might be related to several biological systems. Therefore, the aim of this study was to identify a small set of genes closely associated with stress in a poultry animal model, the chicken (Gallus gallus), by reusing and combining data previously published together with bioinformatic analysis and Bayesian networks in a multi-step approach. Two datasets were collected from publicly available repositories and pre-processed. Bioinformatics analyses were performed to identify genes common to both datasets that showed differential expression patterns between non-stress and stress conditions. Bayesian networks were learnt using a Simulated Annealing algorithm implemented in the software Banjo. The structure of the Bayesian network consisted of 16 out of 19 genes together with the stress condition. Network structure showed CARD19 directly connected to the stress condition plus highlighted CYGB, BRAT1, and EPN3 as relevant, suggesting these genes could play a role in stress. The biological functionality of these genes is related to damage, apoptosis, and oxygen provision, and they could potentially be further explored as biomarkers of stress.

Using an artificial neural network to predict the probability of oviposition events of precision-fed broiler breeder hens

Identifying daily oviposition events for individual broiler breeders is important for bird management. Identifying non-laying birds in a flock that might be caused by improper nutrition or diseases can guide diet changes or disease treatments for these individuals. Oviposition depends on follicle maturation and egg formation, and follicle maturation can be variable. As such, the day and time of oviposition events of individual birds in a free-run flock can be hard to predict. Based on a precision feeding (PF) system that can record the feeding activity of individual birds, a recent study reported a machine learning model to predict daily egg-laying events of broiler breeders. However, there were 2 limitations in that study: 1) It could only be used to identify daily egg-laying events on a subsequent day; 2) The prediction outputs that were binary labels were unable to indicate more details among the outputs with the same label. To improve the previous approach, the current study aimed to predict and output the probability of daily oviposition events occurring using a specific time point in 1 day. In the current study, 706 egg-laying events recorded by nest boxes with radio frequency identification of hens and 706 randomly selected no-egg-laying events were used. The anchor point was newly defined in the current study as a specific time point in 1 day, and 26 features around the anchor point were created for all events (706 egg-laying events and 706 no-egg-laying events). A feed-forward artificial neural network (ANN) model was built for prediction. The area under the receiver operating characteristic (ROC) curve was 0.9409, indicating that the model had an outstanding classification performance. The ANN model could predict oviposition events on the current day, and the output was a probability that could be informative to indicate the likelihood of an oviposition event for an individual breeder. In situations where total egg production was known for a group, the ANN model could predict the probability of daily oviposition events occurring of all individual birds and then rank them to choose those most likely to have laid an egg.

Application of Bayesian networks to the prediction of the AMEn: a new methodology in broiler nutrition

Designing balanced rations for broilers depends on precise knowledge of nitrogen-corrected apparent metabolizable energy (AMEn) and the chemical composition of the feedstuffs. The equations that include the measurements of the chemical composition of the feedstuff can be used in the prediction of AMEn. In the literature, there are studies that obtained prediction equations through multiple regression, meta-analysis, and neural networks. However, other statistical methodologies with promising potential can be used to obtain better predictions of energy values. The objective of the present study was to propose and evaluate the use of Bayesian networks (BN) to the prediction of the AMEn values of energy and protein feedstuffs of vegetable origin used in the formulation of broiler rations. In addition, verify that the predictions of energy values using this methodology are the most accurate and, consequently, are recommended to Animal Science professionals area for the preparation of balanced feeds. BN are models that consist of graphical and probabilistic representations of conditional and joint distributions of the random variables. BN uses machine learning algorithms, being a methodology of artificial intelligence. The bnlearn package in R software was used to predict AMEn from the following covariates: crude protein, crude fiber, ethereal extract, mineral matter, as well as food category, i.e., energy (corn, corn by-products, and others) or protein (soybean, soy by-products, and others) and the type of animal (chick or cockerel). The data come from 568 feeding experiments carried out in Brazil. Additional data from metabolic experiments were obtained from the Federal University of Lavras (UFLA) – Lavras, Minas Gerais, Brazil. The model with the highest accuracy (mean squared error = 66529.8 and multiple coefficients of determination = 0.87) was fitted with the max-min hill climbing algorithm (MMHC) using 80% and 20% of the data for training and test sets, respectively. The accuracy of the models was evaluated based on their values of mean squared error, mean absolute deviation, and mean absolute percentage error. The equations proposed by a new methodology in avian nutrition can be used by the broiler industry in the determination of rations.

Boosted trees to predict pneumonia, growth, and meat percentage of growing-finishing pigs1

In pig production, efficiency is benefiting from uniform growth in pens resulting in single deliveries from a pen of possibly all animals in the targeted weight range. Abnormalities, like pneumonia or aberrant growth, reduce production efficiency as it reduces the uniformity and might cause multiple deliveries per batch and pigs delivered with a low meat yield or outside the targeted weight range. Early identification of pigs prone to develop these abnormalities, for example, at the onset of the growing-finishing phase, would help to prevent heterogeneous pens through management interventions. Data about previous production cycles at the farm combined with data from the piglet’s own history may help in identifying these abnormalities. The aim of this study, therefore, was to predict at the onset of the growing-finishing phase, that is, at 3 mo in advance, deviant pigs at slaughter with a machine-learning technique called boosted trees. The dataset used was extracted from the farm management system of a research center. It contained over 70,000 records of individual pigs born between 2004 and 2016, including information on, for example, offspring, litter size, transfer dates between production stages, their respective locations within the barns, and individual live-weights at several production stages. Results obtained on an independent test set showed that at a 90% specificity rate, the sensitivity was 16% for low meat percentage, 20% for pneumonia and 36% for low lifetime growth rate. For low lifetime growth rate, this meant an almost three times increase in positive predictive value compared to the current situation. From these results, it was concluded that routine performance information available at the onset of the growing-finishing phase combined with data about previous production cycles formed a moderate base to identify pigs prone to develop pneumonia (AUC > 0.60) and a good base to identify pigs prone to develop growth aberrations (AUC > 0.70) during the growing-finishing phase. The mentioned information, however, was not a sufficient base to identify pigs prone to develop low meat percentage (AUC < 0.60). The shown ability to identify growth aberrations and pneumonia can be considered a good first step towards the development of an early warning system for pigs in the growing-finishing phase.

Causal phenotypic networks for egg traits in an F2 chicken population

Traditional single-trait genetic analyses, such as quantitative trait locus (QTL) and genome-wide association studies (GWAS), have been used to understand genotype-phenotype relationships for egg traits in chickens. Even though these techniques can detect potential genes of major effect, they cannot reveal cryptic causal relationships among QTLs and phenotypes. Thus, to better understand the relationships involving multiple genes and phenotypes of interest, other data analysis techniques must be used. Here, we utilized a QTL-directed dependency graph (QDG) mapping approach for a joint analysis of chicken egg traits, so that functional relationships and potential causal effects between them could be investigated. The QDG mapping identified a total of 17 QTLs affecting 24 egg traits that formed three independent networks of phenotypic trait groups (eggshell color, egg production, and size and weight of egg components), clearly distinguishing direct and indirect effects of QTLs towards correlated traits. For example, the network of size and weight of egg components contained 13 QTLs and 18 traits that are densely connected to each other. This indicates complex relationships between genotype and phenotype involving both direct and indirect effects of QTLs on the studied traits. Most of the QTLs were commonly identified by both the traditional (single-trait) mapping and the QDG approach. The network analysis, however, offers additional insight regarding the source and characterization of pleiotropy affecting egg traits. As such, the QDG analysis provides a substantial step forward, revealing cryptic relationships among QTLs and phenotypes, especially regarding direct and indirect QTL effects as well as potential causal relationships between traits, which can be used, for example, to optimize management practices and breeding strategies for the improvement of the traits.

Revealing the structure of the associations between housing system, facilities, management and welfare of commercial laying hens using Additive Bayesian Networks

After the ban of battery cages in 1988, a welfare control programme for laying hens was developed in Sweden. Its goal was to monitor and ensure that animal welfare was not negatively affected by the new housing systems. The present observational study provides an overview of the current welfare status of commercial layer flocks in Sweden and explores the complexity of welfare aspects by investigating and interpreting the inter-relationships between housing system, production type (i.e. organic or conventional), facilities, management and animal welfare indicators. For this purpose, a machine learning procedure referred to as structure discovery was applied to data collected through the welfare programme during 2010-2014 in 397 flocks housed in 193 different farms. Seventeen variables were fitted to an Additive Bayesian Network model. The optimal model was identified by an exhaustive search of the data iterated across incremental parent limits, accounting for prior knowledge about causality, potential over-dispersion and clustering. The resulting Directed Acyclic Graph shows the inter-relationships among the variables. The animal-based welfare indicators included in this study - flock mortality, feather condition and mite infestation - were indirectly associated with each other. Of these, severe mite infestations were rare (4% of inspected flocks) and mortality was below the acceptable threshold (< 0.6%). Feather condition scored unsatisfactory in 21% of the inspected flocks; however, it seemed to be only associated to the age of the flock, ruling out any direct connection with managerial and housing variables. The environment-based welfare indicators - lighting and air quality - were an issue in 5 and 8% of the flocks, respectively, and showed a complex inter-relationship with several managerial and housing variables leaving room for several options for intervention. Additive Bayesian Network modelling outlined graphically the underlying process that generated the observed data. In contrast to ordinary regression, it aimed at accounting for conditional independency among variables, facilitating causal interpretation.

Conceptual framework for investigating causal effects from observational data in livestock

Understanding causal mechanisms among variables is critical to efficient management of complex biological systems such as animal agriculture production. The increasing availability of data from commercial livestock operations offers unique opportunities for attaining causal insight, despite the inherently observational nature of these data. Causal claims based on observational data are substantiated by recent theoretical and methodological developments in the rapidly evolving field of causal inference. Thus, the objectives of this review are as follows: 1) to introduce a unifying conceptual framework for investigating causal effects from observational data in livestock, 2) to illustrate its implementation in the context of the animal sciences, and 3) to discuss opportunities and challenges associated with this framework. Foundational to the proposed conceptual framework are graphical objects known as directed acyclic graphs (DAGs). As mathematical constructs and practical tools, DAGs encode putative structural mechanisms underlying causal models together with their probabilistic implications. The process of DAG elicitation and causal identification is central to any causal claims based on observational data. We further discuss necessary causal assumptions and associated limitations to causal inference. Last, we provide practical recommendations to facilitate implementation of causal inference from observational data in the context of the animal sciences.

Implementing structural equation models to observational data from feedlot production systems

The objective of this study was to illustrate the implementation of a mixed-model-based structural equation modeling (SEM) approach to observational data in the context of feedlot production systems. Different from traditional multiple-trait models, SEMs allow assessment of potential causal interrelationships between outcomes and can effectively discriminate between direct and indirect effects. For illustration, we focused on feedlot performance and its relationship to health outcomes related to Bovine Respiratory Disease (BRD), which accounts for approximately 75% of morbidity and 50-80% of deaths in feedlots. Our data consisted of 1430 lots representing 178,983 cattle from 9 feedlot operations located across the US Great Plains. We explored functional links between arrival weight (AW; i = 1), BRD-related treatment costs (Trt$; as a proxy for health; i = 2) and average daily weight gain (ADG; as an indicator of productive performance i = 3), accounting for the fixed effect of sex and correlation patterns due to the clustering of lots within feedlots. We proposed competing plausible causal models based on expert knowledge. The best fitting model selected for inference supported direct effects of AW on ADG as well as indirect effects of AW on ADG mediated by Trt$. Direct effects from outcome i' to outcome i are quantified by the structural coefficient λ_ii', such that every unit increase in kg/head of AW had a direct effect of increasing ADG by approximately (estimate ± standard error) λˆ₃₁=0.002±0.0001 kg/head/day and also a direct effect of reducing Trt$ by an estimated λˆ₂₁=$0.08±0.006 USD per head. In addition, every $1 USD spent on Trt$ directly decreased ADG by an estimated λˆ₃₂=0.004±0.0006 kg/head/day. From these estimates, we show how to compute the indirect, Trt$-mediated, effect of AW on ADG, as well as the overall effect of AW on ADG, including both direct and indirect effects. We further compared estimates of SEM-based effects with those obtained from standard linear regression mixed models and demonstrated the additional advantage of explicitly distinguishing direct and indirect components of an overall regression effect using SEMs. Understanding the direct and indirect mechanisms of interplay between health and performance outcomes may provide valuable insight into production systems.

Bayesian Networks Illustrate Genomic and Residual Trait Connections in Maize (Zea mays L.)

Relationships among traits were investigated on the genomic and residual levels using novel methodology. This included inference on these relationships via Bayesian networks and an assessment of the networks with structural equation models. The methodology employed three steps. First, a Bayesian multiple-trait Gaussian model was fitted to the data to decompose phenotypic values into their genomic and residual components. Second, genomic and residual network structures among traits were learned from estimates of these two components. Network learning was performed using six different algorithmic settings for comparison, of which two were score-based and four were constraint-based approaches. Third, structural equation model analyses ranked the networks in terms of goodness of fit and predictive ability, and compared them with the standard multiple-trait fully recursive network. The methodology was applied to experimental data representing the European heterotic maize pools Dent and Flint (Zea mays L.). Inferences on genomic and residual trait connections were depicted separately as directed acyclic graphs. These graphs provide information beyond mere pairwise genetic or residual associations between traits, illustrating for example conditional independencies and hinting at potential causal links among traits. Network analysis suggested some genetic correlations as potentially spurious. Genomic and residual networks were compared between Dent and Flint.