Predicting dairy herd resilience on farms with conventional milking systems.

This research paper addresses the problem that, thus far, there is no method available to predict herd resilience for farms that do not use automated milking systems (AMS). Recently, a methodology was developed to estimate both individual cow as well as herd resilience using daily milk yield observations at individual cow level from farms with AMS. This AMS-based method, however, is not suitable on farms that use conventional milking systems (CMS) where such individual cow milk yield observations are lacking. Therefore, this research aimed at predicting herd resilience using herd performance data that is commonly available on CMS farms. To do so, data consisting of 585 Dutch AMS farms where herd resilience estimates using the AMS-based method were available was examined. To predict herd resilience with herd performance data, only those data that are also commonly available on CMS farms were used in a 5-fold cross validation Random Forest model. These herd resilience estimates were subsequently compared with the AMS-based herd resilience estimates. Results showed that it is possible to predict with a 69.9% probability whether a herd performs with above or below average herd resilience using only variables available on CMS farms. Especially, the proportion of cows with an indication of rumen acidosis, proportion of cows with an elevated somatic cell count and the fluctuation in herd size over the years are good predictors of herd resilience. Since herd management decisions appear to affect herd resilience, a lower predicted herd resilience could be taken as a general indication that tactical or strategic management changes could be taken to improve the herd resilience.

Differences across herds with different dairy breeds in daily milk yield based proxies for resilience.

Global sustainability issues such as climate change, biodiversity loss and food security require food systems to become more resource efficient and better embedded in the local environment. This needs a transition towards more diverse, circular and low-input dairy farming systems with animals best suited to the specific environmental conditions. When varying environmental challenges are posed to animals, cows need to become resilient to disturbances they face. This resilience of dairy cows for disturbances can be quantified using sensor features and resilience indicators derived from daily milk yield records. The aim of this study was to explore milk yield based sensor features and resilience indicators for different cattle groups according to their breeds and herds. To this end, we calculated 40 different features to describe the dynamics and variability in milk production of first parity dairy cows. After correction for milk production level, we found that various aspects of the milk yield dynamics, milk yield variability and perturbation characteristics indeed differed across herds and breeds. On farms with a lower breed proportion of Holstein Friesian across cows, there was more variability in the milk yield, but perturbations were less severe upon critical disturbances. Non-Holstein Friesian breeds had a more stable milk production with less (severe) perturbations. These differences can be attributed to differences in genetics, environments, or both. This study demonstrates the potential to use milk yield sensor features and resilience indicators as a tool to quantify how cows cope with more dynamic production conditions and select animals for features that best suit a farms' breeding goal and specific environment.

Classifying aneuploidy in genotype intensity data using deep learning.

Aneuploidy is the loss or gain of one or more chromosomes. Although it is a rare phenomenon in liveborn individuals, it is observed in livestock breeding populations. These breeding populations are often routinely genotyped and the genotype intensity data from single nucleotide polymorphism (SNP) arrays can be exploited to identify aneuploidy cases. This identification is a time-consuming and costly task, because it is often performed by visual inspection of the data per chromosome, usually done in plots of the intensity data by an expert. Therefore, we wanted to explore the feasibility of automated image classification to replace (part of) the visual detection procedure for any diploid species. The aim of this study was to develop a deep learning Convolutional Neural Network (CNN) classification model based on chromosome level plots of SNP array intensity data that can classify the images into disomic, monosomic and trisomic cases. A multispecies dataset enriched for aneuploidy cases was collected containing genotype intensity data of 3321 disomic, 1759 monosomic and 164 trisomic chromosomes. The final CNN model had an accuracy of 99.9%, overall precision was 1, recall was 0.98 and the F1 score was 0.99 for classifying images from intensity data. The high precision assures that cases detected are most likely true cases, however, some trisomy cases may be missed (the recall of the class trisomic was 0.94). This supervised CNN model performed much better than an unsupervised k-means clustering, which reached an accuracy of 0.73 and had especially difficult to classify trisomic cases correctly. The developed CNN classification model provides high accuracy to classify aneuploidy cases based on images of plotted X and Y genotype intensity values. The classification model can be used as a tool for routine screening in large diploid populations that are genotyped to get a better understanding of the incidence and inheritance, and in addition, avoid anomalies in breeding candidates.

A data-driven prediction of lifetime resilience of dairy cows using commercial sensor data collected during first lactation.

Reliable prediction of lifetime resilience early in life can contribute to improved management decisions of dairy farmers. Several studies have shown that time series sensor data can be used to predict lifetime resilience rankings. However, such predictions generally require the translation of sensor data into biologically meaningful sensor features, which involve proper feature definitions and a lot of preprocessing. The objective of this study was to investigate the hypothesis that data-driven random forest algorithms can equal or improve the prediction of lifetime resilience scores compared with ordinal logistic regression, and that these algorithms require considerably less effort for data preprocessing. We studied this by developing prediction models that forecast lifetime resilience of a cow early in her productive life using sensor data from the first lactation. We used an existing data set from a Dutch experimental herd, with data of culled cows for which birth dates, insemination dates, calving dates, culling dates, and health treatments were available to calculate lifetime resilience scores. Moreover, 4 types of first-lactation sensor data, converted to daily aggregated values, were available: milk yield, body weight, activity, and rumination. For each sensor, 14 sensor features were calculated, of which part were based on absolute daily values and part on relative to herd average values. First, we predicted lifetime resilience rank with stepwise logistic regression using sensor features as predictors and a P-value of <0.2 as the cut-off. Next, we applied a random forest with the 6 features that remained in the final logistic regression model. We then applied a random forest with all sensor features, and finally applied a random forest with daily aggregated values as features. All models were validated with stratified 10-fold cross-validation with 90% of the records in the training set and 10% in the validation set. Model performances expressed in percentage of correctly classified cows (accuracy) and percentage of cows being critically misclassified (i.e., high as low and vice versa) ± standard deviation were 45.1 ± 8.1% and 10.8% with the ordinal logistic regression model, 45.7 ± 8.4% and 16.0% with the random forest using the same 6 features as the logistic regression model, 48.4 ± 6.7% and 10.0% for the random forest with all sensor features, and 50.5 ± 6.3% and 8.4% for the random forest with daily sensor values. This random forest also revealed that data collected in early and late stages of first lactation seem to be of particular importance in the prediction compared with that in mid lactation. Accuracies of the models were not significantly different, but the percentage of critically misclassified cows was significantly higher for the second model than for the other models. We concluded that a data-driven random forest algorithm with daily aggregated sensor data as input can be used for the prediction of lifetime resilience classification with an overall accuracy of ~50%, and provides at least as good prediction as models with sensor features as input.

Machine learning to further improve the decision which boar ejaculates to process into artificial insemination doses.

Current artificial insemination (AI) laboratory practices assess semen quality of each boar ejaculate to decide which ones to process into AI doses. This decision is aided with two, world-wide used, motility parameters that come available through computer assisted semen analysis (CASA). This decision process, however, still results in AI doses with variable and sometimes suboptimal fertility outcomes (e.g., small litter size). The hypothesis was that the decision which ejaculates to process into AI doses can be improved by adding more data from CASA systems, and data from other sources, in combination with a data-driven model. Available data consisted of ejaculates that passed the initial decision, and thus, were processed into AI doses and used to inseminate sows. Data were divided into a training set (6793 records) and a validation set (1191 records) for model development, and an independent test set (1434 records) for performance assessment. Gradient Boosting Machine (GBM) models were developed to predict four fertility phenotypes of interest (gestation length, total number born, number born alive, and number of stillborn piglets). Each fertility phenotype was considered as a numeric and as a binary outcome parameter, totaling to eight different fertility phenotypes. Data used to further improve the decision process originated from four sources: 1) CASA information, 2) boar ejaculate information, 3) breeding value estimations, and 4) weather information. These data were used to create seven prediction sets, where each new set added parameters to the ones included in the previous set. The GBM models predicted fertility phenotypes with low correlations (for numeric phenotypes) and area under the curve values (for binary phenotypes) on the test data. Hence, results demonstrated that a combination of more data and GBM did not enable further improvement of the AI dose quality checks, resulting in the rejection of our hypothesis. However, our study revealed parameters affecting boar ejaculate fertility which were not used in today's decision process. These parameters (listed in the top 10 in at least four GBM models) included one parameter associated with boar ejaculate information, two with breeding value estimations, five with CASA information, and one with weather information. These parameters, therefore, should be further investigated for their potential value when assessing the quality of boar ejaculates in daily routine AI doses processing.

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.

Shorter sampling periods and accurate estimates of milk volume and components are possible for pasture based dairy herds milked with automated milking systems.

Dairy cows grazing pasture and milked using automated milking systems (AMS) have lower milking frequencies than indoor fed cows milked using AMS. Therefore, milk recording intervals used for herd testing indoor fed cows may not be suitable for cows on pasture based farms. We hypothesised that accurate standardised 24 h estimates could be determined for AMS herds with milk recording intervals of less than the Gold Standard (48 hs), but that the optimum milk recording interval would depend on the herd average for milking frequency. The Gold Standard protocol was applied on five commercial dairy farms with AMS, between December 2011 and February 2013. From 12 milk recording test periods, involving 2211 cow-test days and 8049 cow milkings, standardised 24 h estimates for milk volume and milk composition were calculated for the Gold Standard protocol and compared with those collected during nine alternative sampling scenarios, including six shorter sampling periods and three in which a fixed number of milk samples per cow were collected. Results infer a 48 h milk recording protocol is unnecessarily long for collecting accurate estimates during milk recording on pasture based AMS farms. Collection of two milk samples only per cow was optimal in terms of high concordance correlation coefficients for milk volume and components and a low proportion of missed cow-test days. Further research is required to determine the effects of diurnal variations in milk composition on standardised 24 h estimates for milk volume and components, before a protocol based on a fixed number of samples could be considered. Based on the results of this study New Zealand have adopted a split protocol for herd testing based on the average milking frequency for the herd (NZ Herd Test Standard 8100:2015).

Sensors and clinical mastitis--the quest for the perfect alert.

When cows on dairy farms are milked with an automatic milking system or in high capacity milking parlors, clinical mastitis (CM) cannot be adequately detected without sensors. The objective of this paper is to describe the performance demands of sensor systems to detect CM and evaluats the current performance of these sensor systems. Several detection models based on different sensors were studied in the past. When evaluating these models, three factors are important: performance (in terms of sensitivity and specificity), the time window and the similarity of the study data with real farm data. A CM detection system should offer at least a sensitivity of 80% and a specificity of 99%. The time window should not be longer than 48 hours and study circumstances should be as similar to practical farm circumstances as possible. The study design should comprise more than one farm for data collection. Since 1992, 16 peer-reviewed papers have been published with a description and evaluation of CM detection models. There is a large variation in the use of sensors and algorithms. All this makes these results not very comparable. There is a also large difference in performance between the detection models and also a large variation in time windows used and little similarity between study data. Therefore, it is difficult to compare the overall performance of the different CM detection models. The sensitivity and specificity found in the different studies could, for a large part, be explained in differences in the used time window. None of the described studies satisfied the demands for CM detection models.

Fish consumption, n-3 fatty acids, and colorectal cancer: a meta-analysis of prospective cohort studies.

Animal studies show favorable effects of n-3 fatty acids on inflammation and cancer, but results from epidemiologic studies appear to be inconsistent. The authors conducted meta-analyses of prospective cohort studies that evaluated the association between fish consumption or n-3 fatty acids and colorectal cancer incidence or mortality. Random-effects models were used, and heterogeneity between study results was explored through stratified analyses. The pooled relative risks for the highest compared with the lowest fish consumption category were 0.88 (95% confidence interval: 0.78, 1.00) for colorectal cancer incidence (14 studies) and 1.02 (95% confidence interval: 0.90, 1.16) for colorectal cancer mortality (four studies). The pooled relative risks for colorectal cancer incidence were 0.96 (95% confidence interval: 0.92, 1.00) for each extra occurrence of fish consumption per week (seven studies) and 0.97 (95% confidence interval: 0.92, 1.03) for each extra 100 g of fish consumed per week (four studies). Stratified analysis showed that the pooled relative risk for colorectal cancer incidence was more pronounced for women and in studies with a large exposure contrast. In cohort studies, fish consumption was shown to slightly reduce colorectal cancer risk. Existing evidence that n-3 fatty acids inhibit colorectal carcinogenesis is in line with these results, but few data are available addressing this association.