In this paper, six univariate and two multivariate best linear unbiased prediction (BLUP) models were tested for the estimation of breeding values (BV) in Holstein Friesian cattle in Serbia. Two univariate models were formed using the numerator relationship matrix (NRM), four using the genomic relationship matrix (GRM). Multivariate models contained only an NRM. Two cases were studied, the first when only first lactations were observed, and the second when all lactations were observed using a repeatability model. A total of 6041 animals were included, and of them, 2565 had data on milk yield (MY), milk fat yield (FY), milk fat content (FC), milk protein yield (PY) and milk protein content (PC). Finally, out of those 2565 cows, 1491 were genotyped. A higher accuracy of BV was obtained when using a combination of NRM and GRM compared to NRM alone in univariate analysis, while multivariate analysis with repeated measures gave the highest accuracy with all 6041 animals. When only genotyped animals were observed, the highest accuracy of the estimated BV was calculated by the ssGBLUPp model, and the lowest by the univariate BLUP model. In conclusion, the current breeding programs in Serbia should be changed to use multivariate analysis with repeated measurements until the optimal size of the reference population, which must include genotyping data on both bulls and cows, is reached.

The use of genomic data makes it possible to examine genetic variability and calculate the genetic parameters of the population in an efficient and precise way. The aim of this research was to analyse linkage disequilibrium (LD), contemporary effective population size (Ne c), haplotype block structure, minor allele frequency (MAF), observed (H O) and expected heterozygosity (H E), calculate the genomic relationship matrix and perform a principal component analysis (PCA) in the Serbian Holstein-Friesian cattle population using SNP data from the GGP Bovine 100K chip. After quality control (QC), 83 208 SNPs and 1 575 cows were retained for further analysis. LD on autosomes had an average value of ≥ 0.2 up to a distance of 50-60 kb (r 2 = 0.211), while on BTX r 2 ≥ 0.2 was represented at distances of 80-90 kb (r 2 = 0.211). LD differed between chromosomes. The average H O for autosomes and X chromosome SNPs was 0.412 and 0.422, respectively. 81.30% of SNPs that passed QC had MAF > 0.2. The total number of haplotype blocks in the studied population was 15 642. On average, blocks contained 2.932 SNPs. The average block length was 32.657 kb and ranged from a minimum of 0.019 kb (BTA21 and BTA26) to a maximum of 999.562 kb (BTX). The estimated value of Ne c in the this cattle population was 142. The results of PCA showed a significant variability of genotypes in the population, but there was no clear stratification of the population. The obtained results will serve as a basis for future genomic analyses such as the detection of QTLs for important economic traits and the implementation of genomic selection.

Summary: This research aims to estimate heritability and repeatability based on the data on milk production traits (MY – milk yield; FY – milk fat yield; FC – milk fat content; PY – milk protein yield and PC – milk protein content) as well as pedigree and genomic information. A total of 6,041 animals were included in the research, while 2,565 of them had data for milk production traits. In order to form a genomic relationship matrix, 58K SNP data were used for a total of 1,491 cows. Several software tools were used in the preparation and analysis of data, which were provided by the Central Breeding Organization, Department of Animal Science, Faculty of Agriculture, University of Novi Sad. PreGSF90, in combination with RENUMF90, was used for quality control of genomic information. Genetic analysis was performed in WOMBAT software by the REML using standard repeatability univariate analysis (BLUPpe) and repeatability models for genomic prediction (GBLUPpe and ssGBLUPpe). In all three analyses, the highest heritability (0.410, 0.378 and 0.389, respectively) and repeatability (0.449, 0.429 and 0.440, respectively) were calculated for FC. Heritability estimates for all other traits were lower. Heritability ranged from 0.158 to 0.185 for MY, from 0.166 to 0.178 for FY, from 0.141 to 0.154 for PY and from 0.135 to 0.221 for PC. Heritability estimates indicate that it is possible to achieve genetic improvement but it is necessary to introduce the best model for prediction of breeding values of cow.

Summary: This paper presents a deep neural network (DNN) approach designed to estimate the milk yield of Holstein-Friesian cattle. The DNN comprised stacked dense (fully connected) layers, each hidden layer followed by a dropout layer. Various configurations of the DNN were tested, incorporating 2 and 3 hidden layers containing 8 to 54 neurons. The experiment involved testing the DNN with different activation functions such as the sigmoid, tanh, and rectified linear unit (ReLU). The dropout rates ranging from 0 to 0.3 were employed, with the output layer using a linear activation function. The DNN models were trained using the Adam, SGD, and RMSprop optimizers, with the root mean square error serving as the loss metric. The training dataset comprised information from a unique database containing records of dairy cows in the Republic of Serbia, totaling 3,406 cows. The input parameters (a total of 27) for the DNN included breeding and milk yield data from the cow’s mother, as well as the father’s ID, whereas the output parameters (a total of 8) consisted of milk yield parameters (a total of 3) and breeding parameters of the cow (a total of 5). Training iterations were conducted using a batch size of 8 over 500, 1000, and 5000 epochs.