Analyzing Data Modalities for Cattle Weight Estimation Using Deep Learning Models.

We investigate the impact of different data modalities for cattle weight estimation. For this purpose, we collect and present our own cattle dataset representing the data modalities: RGB, depth, combined RGB and depth, segmentation, and combined segmentation and depth information. We explore a recent vision-transformer-based zero-shot model proposed by Meta AI Research for producing the segmentation data modality and for extracting the cattle-only region from the images. For experimental analysis, we consider three baseline deep learning models. The objective is to assess how the integration of diverse data sources influences the accuracy and robustness of the deep learning models considering four different performance metrics: mean absolute error (MAE), root mean squared error (RMSE), mean absolute percentage error (MAPE), and R-squared (R2). We explore the synergies and challenges associated with each modality and their combined use in enhancing the precision of cattle weight prediction. Through comprehensive experimentation and evaluation, we aim to provide insights into the effectiveness of different data modalities in improving the performance of established deep learning models, facilitating informed decision-making for precision livestock management systems.

Multi-trait genomic prediction in pigs using single and multistep methods based on the absorption of ungenotyped animals.

Many quantitative traits measured in breeding programs are genetically correlated. The genetic correlations between the traits indicate that the measurement of one trait carries information on others. To benefit from this information, multi-trait genomic prediction (MTGP) is preferable to use. However, MTGP is more difficult to implement compared to single-trait genomic prediction (STGP), and even more challenging for the goal to exploit not only the information on other traits but also the information on ungenotyped animals. This could be accomplished using both single and multistep methods. The single-step method was achieved by implementing a single-step genomic best linear unbiased prediction (ssGBLUP) approach using a multi-trait model. Here, we examined a multistep analysis based on an approach called "Absorption" to achieve this goal. The Absorption approach absorbed all available information including the phenotypic information on ungenotyped animals as well as the information on other traits if applicable, into mixed model equations of genotyped animals. The multistep analysis included (1) to apply the Absorption approach that exploits all available information and (2) to implement genomic BLUP (GBLUP) prediction on the absorbed dataset. In this study, the ssGBLUP and multistep analysis were applied to 5 traits in Duroc pigs, which were slaughter percentage, feed consumption from 40 to 120 kg (FC40_120), days of growth from 40 to 120 kg (D40_120), age at 40 kg (A40) and lean meat percentage. The results showed that MTGP yielded higher accuracy than STGP, which on average was 0.057 higher for the multistep method and 0.045 higher for ssGBLUP. The multistep method achieved similar prediction accuracy as ssGBLUP. However, the prediction bias of the multistep method was in general lower than that of ssGBLUP.

A high-throughput study of visceral organs in CT-scanned pigs.

It has been debated whether intensive selection for growth and carcass yield in pig breeding programmes can affect the size of internal organs, and thereby reduce the animal's ability to handle stress and increase the risk of sudden deaths. To explore the respiratory and circulatory system in pigs, a deep learning based computational pipeline was built to extract the size of lungs and hearts from CT-scan images. This pipeline was applied on CT images from 11,000 boar selection candidates acquired during the last decade. Further, heart and lung volumes were analysed genetically and correlated with production traits. Both heart and lung volumes were heritable, with h2 estimated to 0.35 and 0.34, respectively, in Landrace, and 0.28 and 0.4 in Duroc. Both volumes were positively correlated with lean meat percentage, and lung volume was negatively genetically correlated with growth (rg = - 0.48 ± 0.07 for Landrace and rg = - 0.44 ± 0.07 for Duroc). The main findings suggest that the current pig breeding programs could, as an indirect response to selection, affect the size of hearts- and lungs. The presented methods can be used to monitor the development of internal organs in the future.

Accuracy of genomic prediction of maternal traits in pigs using Bayesian variable selection methods.

The aim of this study was to compare three methods of genomic prediction: GBLUP, BayesC and BayesGC for genomic prediction of six maternal traits in Landrace sows using a panel of 660 K SNPs. The effects of different priors for the Bayesian methods were also investigated. GBLUP does not take the genetic architecture into account as all SNPs are assumed to have equally sized effects and relies heavily on the relationships between the animals for accurate predictions. Bayesian approaches rely on both fitting SNPs that describe relationships between animals in addition to fitting single SNP effects directly. Both the relationship between the animals and single SNP effects are important for accurate predictions. Maternal traits in sows are often more difficult to record and have lower heritabilities. BayesGC was generally the method with the higher accuracy, although its accuracy was for some traits matched by that of GBLUP and for others by that of BayesC. For piglet mortality within 3 weeks, BayesGC achieved up to 9.2% higher accuracy. For many of the traits, however, the methods did not show significant differences in accuracies.

Correcting for base-population differences and unknown parent groups in single-step genomic predictions of Norwegian Red cattle.

Bias and inflation in genomic evaluation with the single-step methods have been reported in several studies. Incompatibility between the base-populations of the pedigree-based and the genomic relationship matrix (G) could be a reason for these biases. Inappropriate ways of accounting for missing parents could be another reason for biases in genetic evaluations with or without genomic information. To handle these problems, we fitted and evaluated a fixed covariate (J) that contains ones for genotyped animals and zeros for unrelated non-genotyped animals, or pedigree-based regression coefficients for related non-genotyped animals. We also evaluated alternative ways of fitting the J covariate together with genetic groups on biases and stability of breeding value estimates, and of including it into G as a random effect. In a whole vs. partial data set comparison, four scenarios were investigated for the partial data: genotypes missing, phenotypes missing, both genotypes and phenotypes missing, and pedigree missing. Fitting J either as fixed or random reduced level-bias and inflation and increased stability of genomic predictions as compared to the basic model where neither J nor genetic groups were fitted. In most models, genomic predictions were largely biased for scenarios with missing genotype and phenotype information. The biases were reduced for models which combined group and J effects. Models with these corrected group covariates performed better than the recently published model where genetic groups were encapsulated and fitted as random via the Quaas and Pollak transformation. In our Norwegian Red cattle data, a model which combined group and J regression coefficients was preferred because it showed least bias and highest stability of genomic predictions across the scenarios.

Our study dealt with strategies on how to reduce biases (inflation and level-bias) and improve a parameter related to accuracy (stability) of genomic predictions of breeding values that combine genotyped and non-genotyped animals, which are denoted as single-step genomic predictions. We tried to remedy incompatibilities between the pedigree- and the genomics-based relationships matrices by fitting a covariate (J) that corrects for base-population differences that may occur between both relationship matrices. We also evaluated alternative ways to combine the J covariate and genetic group effects to account for missing parental information, which often occurs in practical breeding schemes. We found that fitting J either as fixed or random reduced level-bias and inflation and increased stability of genomic predictions as compared to the basic model where neither J nor genetic groups were fitted. Level-biases and inflation of breeding value estimates were reduced, and stability of genomic predictions improved for models which combined group and J effects. A model which fits group regression coefficients minus the part that could be explained from pedigree was recommended because it showed least bias and highest stability across the scenarios and has theoretical justification.

Predicting survival and longevity of sows using purebred and crossbred data.

Survival and longevity are very important traits in pig breeding. From an economic standpoint, it is favorable to keep the sows for another parity instead of replacing them and, from the animal's perspective, better welfare is achieved if they do not experience health problems. It is challenging to record longevity in purebred (PB) nucleus herds because animals are more likely to be replaced based on breeding value and high replacement rates rather than inability to produce. Crossbred (CB) sows are, however, submitted to lower replacement rates and are more likely to be kept in the farm longer if they can produce large and robust litters. Therefore, the objective of this study was to investigate whether the use of CB phenotypes could improve prediction accuracy of longevity for PBs. In addition, a new definition of survival was investigated. The analyzed data included phenotypes from two PB dam lines and their F1 cross. Three traits were evaluated: 1) whether or not the sow got inseminated for a second litter within 85 d of first farrowing (Longevity 1-2), 2) how many litters the sow can produce within 570 d of first farrowing [Longevity 1-5 (LGY15)], and 3) a repeatability trait that indicates whether or not the sow survived until the next parity (Survival). Traits were evaluated both as the same across breeds and as different between breeds. Results indicated that longevity is not the same trait in PB and CB animals (low genetic correlation). In addition, there were differences between the two PB lines in terms of which trait definition gave the greatest prediction accuracy. The repeatability trait (Survival) gave the greatest prediction accuracy for breed B, but LGY15 gave the greatest prediction accuracy for breed A. Prediction accuracy for CBs was generally poor. The Survival trait is recorded earlier in life than LGY15 and seemed to give a greater prediction accuracy for young animals than LGY15 (until own phenotype was available). Thus, for selection of young animals for breeding, Survival would be the preferred trait definition. In addition, results indicated that lots of data were needed to get accurate estimates of breeding values and that, if CB performance is the breeding goal, CB phenotypes should be used in the genetic evaluation.

Meta-analysis for milk fat and protein percentage using imputed sequence variant genotypes in 94,321 cattle from eight cattle breeds.

BACKGROUND: Sequence-based genome-wide association studies (GWAS) provide high statistical power to identify candidate causal mutations when a large number of individuals with both sequence variant genotypes and phenotypes is available. A meta-analysis combines summary statistics from multiple GWAS and increases the power to detect trait-associated variants without requiring access to data at the individual level of the GWAS mapping cohorts. Because linkage disequilibrium between adjacent markers is conserved only over short distances across breeds, a multi-breed meta-analysis can improve mapping precision. RESULTS: To maximise the power to identify quantitative trait loci (QTL), we combined the results of nine within-population GWAS that used imputed sequence variant genotypes of 94,321 cattle from eight breeds, to perform a large-scale meta-analysis for fat and protein percentage in cattle. The meta-analysis detected (p ≤ 10-8) 138 QTL for fat percentage and 176 QTL for protein percentage. This was more than the number of QTL detected in all within-population GWAS together (124 QTL for fat percentage and 104 QTL for protein percentage). Among all the lead variants, 100 QTL for fat percentage and 114 QTL for protein percentage had the same direction of effect in all within-population GWAS. This indicates either persistence of the linkage phase between the causal variant and the lead variant across breeds or that some of the lead variants might indeed be causal or tightly linked with causal variants. The percentage of intergenic variants was substantially lower for significant variants than for non-significant variants, and significant variants had mostly moderate to high minor allele frequencies. Significant variants were also clustered in genes that are known to be relevant for fat and protein percentages in milk. CONCLUSIONS: Our study identified a large number of QTL associated with fat and protein percentage in dairy cattle. We demonstrated that large-scale multi-breed meta-analysis reveals more QTL at the nucleotide resolution than within-population GWAS. Significant variants were more often located in genic regions than non-significant variants and a large part of them was located in potentially regulatory regions.

Modelling the shape of the pig scapula.

BACKGROUND: The shape of pig scapula is complex and is important for sow robustness and health. To better understand the relationship between 3D shape of the scapula and functional traits, it is necessary to build a model that explains most of the morphological variation between animals. This requires point correspondence, i.e. a map that explains which points represent the same piece of tissue among individuals. The objective of this study was to further develop an automated computational pipeline for the segmentation of computed tomography (CT) scans to incorporate 3D modelling of the scapula, and to develop a genetic prediction model for 3D morphology. RESULTS: The surface voxels of the scapula were identified on 2143 CT-scanned pigs, and point correspondence was established by predicting the coordinates of 1234 semi-landmarks on each animal, using the coherent point drift algorithm. A subsequent principal component analysis showed that the first 10 principal components covered more than 80% of the total variation in 3D shape of the scapula. Using principal component scores as phenotypes in a genetic model, estimates of heritability ranged from 0.4 to 0.8 (with standard errors from 0.07 to 0.08). To validate the entire computational pipeline, a statistical model was trained to predict scapula shape based on marker genotype data. The mean prediction reliability averaged over the whole scapula was equal to 0.18 (standard deviation = 0.05) with a higher reliability in convex than in concave regions. CONCLUSIONS: Estimates of heritability of the principal components were high and indicated that the computational pipeline that processes CT data to principal component phenotypes was associated with little error. Furthermore, we showed that it is possible to predict the 3D shape of scapula based on marker genotype data. Taken together, these results show that the proposed computational pipeline closes the gap between a point cloud representing the shape of an animal and its underlying genetic components.

Level-biases in estimated breeding values due to the use of different SNP panels over time in ssGBLUP.

BACKGROUND: The main aim of single-step genomic predictions was to facilitate optimal selection in populations consisting of both genotyped and non-genotyped individuals. However, in spite of intensive research, biases still occur, which make it difficult to perform optimal selection across groups of animals. The objective of this study was to investigate whether incomplete genotype datasets with errors could be a potential source of level-bias between genotyped and non-genotyped animals and between animals genotyped on different single nucleotide polymorphism (SNP) panels in single-step genomic predictions. RESULTS: Incomplete and erroneous genotypes of young animals caused biases in breeding values between groups of animals. Systematic noise or missing data for less than 1% of the SNPs in the genotype data had substantial effects on the differences in breeding values between genotyped and non-genotyped animals, and between animals genotyped on different chips. The breeding values of young genotyped individuals were biased upward, and the magnitude was up to 0.8 genetic standard deviations, compared with breeding values of non-genotyped individuals. Similarly, the magnitude of a small value added to the diagonal of the genomic relationship matrix affected the level of average breeding values between groups of genotyped and non-genotyped animals. Cross-validation accuracies and regression coefficients were not sensitive to these factors. CONCLUSIONS: Because, historically, different SNP chips have been used for genotyping different parts of a population, fine-tuning of imputation within and across SNP chips and handling of missing genotypes are crucial for reducing bias. Although all the SNPs used for estimating breeding values are present on the chip used for genotyping young animals, incompleteness and some genotype errors might lead to level-biases in breeding values.

Effects of heterozygosity on performance of purebred and crossbred pigs.

BACKGROUND: In pigs, crossbreeding aims at exploiting heterosis, but heterosis is difficult to quantify. Heterozygosity at genetic markers is easier to measure and could potentially be used as an indicator of heterosis. The objective of this study was to investigate the effect of heterozygosity on various maternal and production traits in purebred and crossbred pigs. The proportion of heterozygosity at genetic markers across the genome for each individual was included in the prediction model as a fixed regression across or within breeds. RESULTS: Estimates of regression coefficients of heterozygosity showed large effects for some traits. For maternal traits, regression coefficient estimates were always in a favourable direction, while for production, meat and slaughter quality traits, they were both favourable and unfavourable. Traits with the largest estimated effects of heterozygosity were total number born, litter weight at 3 weeks, weight at 150 days, and age at 40 kg. Estimates of regression coefficients on heterozygosity differed between breeds. Traits with the largest effect of heterozygosity also showed a significant (P < 0.05) increase in prediction accuracy when heterozygosity was included in the model compared to the model without heterozygosity. CONCLUSIONS: For traits with the largest estimates of regression coefficients on heterozygosity, the inclusion of heterozygosity in the model improved prediction accuracy. Using models that include heterozygosity would result in selecting different animals for breeding, which has the potential to improve genetic gain for these traits. This is most beneficial when crossbreds or several breeds are included in the estimation of breeding values and is relevant to all species, not only pigs. Thus, our results show that including heterozygosity in the model is beneficial for some traits, likely due to dominant gene action.

Towards causally cohesive genotype-phenotype modelling for characterization of the soft-tissue mechanics of the heart in normal and pathological geometries.

A scientific understanding of individual variation is key to personalized medicine, integrating genotypic and phenotypic information via computational physiology. Genetic effects are often context-dependent, differing between genetic backgrounds or physiological states such as disease. Here, we analyse in silico genotype-phenotype maps (GP map) for a soft-tissue mechanics model of the passive inflation phase of the heartbeat, contrasting the effects of microstructural and other low-level parameters assumed to be genetically influenced, under normal, concentrically hypertrophic and eccentrically hypertrophic geometries. For a large number of parameter scenarios, representing mock genetic variation in low-level parameters, we computed phenotypes describing the deformation of the heart during inflation. The GP map was characterized by variance decompositions for each phenotype with respect to each parameter. As hypothesized, the concentric geometry allowed more low-level parameters to contribute to variation in shape phenotypes. In addition, the relative importance of overall stiffness and fibre stiffness differed between geometries. Otherwise, the GP map was largely similar for the different heart geometries, with little genetic interaction between the parameters included in this study. We argue that personalized medicine can benefit from a combination of causally cohesive genotype-phenotype modelling, and strategic phenotyping that captures effect modifiers not explicitly included in the mechanistic model.

A computational pipeline for quantification of mouse myocardial stiffness parameters.

The mouse is an important model for theoretical-experimental cardiac research, and biophysically based whole organ models of the mouse heart are now within reach. However, the passive material properties of mouse myocardium have not been much studied. We present an experimental setup and associated computational pipeline to quantify these stiffness properties. A mouse heart was excised and the left ventricle experimentally inflated from 0 to 1.44kPa in eleven steps, and the resulting deformation was estimated by echocardiography and speckle tracking. An in silico counterpart to this experiment was built using finite element methods and data on ventricular tissue microstructure from diffusion tensor MRI. This model assumed a hyperelastic, transversely isotropic material law to describe the force-deformation relationship, and was simulated for many parameter scenarios, covering the relevant range of parameter space. To identify well-fitting parameter scenarios, we compared experimental and simulated outcomes across the whole range of pressures, based partly on gross phenotypes (volume, elastic energy, and short- and long-axis diameter), and partly on node positions in the geometrical mesh. This identified a narrow region of experimentally compatible values of the material parameters. Estimation turned out to be more precise when based on changes in gross phenotypes, compared to the prevailing practice of using displacements of the material points. We conclude that the presented experimental setup and computational pipeline is a viable method that deserves wider application.

OpenCMISS: a multi-physics & multi-scale computational infrastructure for the VPH/Physiome project.

The VPH/Physiome Project is developing the model encoding standards CellML (cellml.org) and FieldML (fieldml.org) as well as web-accessible model repositories based on these standards (models.physiome.org). Freely available open source computational modelling software is also being developed to solve the partial differential equations described by the models and to visualise results. The OpenCMISS code (opencmiss.org), described here, has been developed by the authors over the last six years to replace the CMISS code that has supported a number of organ system Physiome projects. OpenCMISS is designed to encompass multiple sets of physical equations and to link subcellular and tissue-level biophysical processes into organ-level processes. In the Heart Physiome project, for example, the large deformation mechanics of the myocardial wall need to be coupled to both ventricular flow and embedded coronary flow, and the reaction-diffusion equations that govern the propagation of electrical waves through myocardial tissue need to be coupled with equations that describe the ion channel currents that flow through the cardiac cell membranes. In this paper we discuss the design principles and distributed memory architecture behind the OpenCMISS code. We also discuss the design of the interfaces that link the sets of physical equations across common boundaries (such as fluid-structure coupling), or between spatial fields over the same domain (such as coupled electromechanics), and the concepts behind CellML and FieldML that are embodied in the OpenCMISS data structures. We show how all of these provide a flexible infrastructure for combining models developed across the VPH/Physiome community.

Neural network firing-rate models on integral form: effects of temporal coupling kernels on equilibrium-state stability.

Firing-rate models describing neural-network activity can be formulated in terms of differential equations for the synaptic drive from neurons. Such models are typically derived from more general models based on Volterra integral equations assuming exponentially decaying temporal coupling kernels describing the coupling of pre- and postsynaptic activities. Here we study models with other choices of temporal coupling kernels. In particular, we investigate the stability properties of constant solutions of two-population Volterra models by studying the equilibrium solutions of the corresponding autonomous dynamical systems, derived using the linear chain trick, by means of the Routh-Hurwitz criterion. In the four investigated synaptic-drive models with identical equilibrium points we find that the choice of temporal coupling kernels significantly affects the equilibrium-point stability properties. A model with an alpha-function replacing the standard exponentially decaying function in the inhibitory coupling kernel is in most of our examples found to be most prone to instability, while the opposite situation with an alpha-function describing the excitatory kernel is found to be least prone to instability. The standard model with exponentially decaying coupling kernels is typically found to be an intermediate case. We further find that stability is promoted by increasing the weight of self-inhibition or shortening the time constant of the inhibition.