Model selection for inferential models with high dimensional data: synthesis and graphical representation of multiple techniques

A Comparative Analysis of XGBoost and Neural Network Models for Predicting Some Tomato Fruit Quality Traits from Environmental and Meteorological Data

The tomato as a raw material for processing is globally important and is pivotal in dietary and agronomic research due to its nutritional, economic, and health significance. This study explored the potential of machine learning (ML) for predicting tomato quality, utilizing data from 48 cultivars and 28 locations in Hungary over 5 seasons. It focused on °Brix, lycopene content, and colour (a/b ratio) using extreme gradient boosting (XGBoost) and artificial neural network (ANN) models. The results revealed that XGBoost consistently outperformed ANN, achieving high accuracy in predicting °Brix (R² = 0.98, RMSE = 0.07) and lycopene content (R² = 0.87, RMSE = 0.61), and excelling in colour prediction (a/b ratio) with a R² of 0.93 and RMSE of 0.03. ANN lagged behind particularly in colour prediction, showing a negative R² value of -0.35. Shapley additive explanation's (SHAP) summary plot analysis indicated that both models are effective in predicting °Brix and lycopene content in tomatoes, highlighting different aspects of the data. SHAP analysis highlighted the models' efficiency (especially in °Brix and lycopene predictions) and underscored the significant influence of cultivar choice and environmental factors like climate and soil. These findings emphasize the importance of selecting and fine-tuning the appropriate ML model for enhancing precision agriculture, underlining XGBoost's superiority in handling complex agronomic data for quality assessment.

Predicting lameness in dairy cattle using untargeted liquid chromatography-mass spectrometry-based metabolomics and machine learning

Lameness in dairy cattle is a highly prevalent condition that impacts on the health and welfare of dairy cows. Prompt detection and implementation of effective treatment is important for managing lameness. However, major limitations are associated with visual assessment of lameness, which is the most commonly used method to detect lameness. The aims of this study were to investigate the use of metabolomics and machine learning to develop novel methods to detect lameness. Untargeted metabolomics using liquid chromatography-mass spectrometry (LC-MS) alongside machine learning models and a stability selection method were utilized to evaluate the predictive accuracy of differences in the metabolomics profile of first-lactation dairy cows before (during the transition period) and at the time of lameness (based on visual assessment using the 0-3 scale of the Agriculture and Horticulture Development Board). Urine samples were collected from 2 cohorts of dairy heifers and stored at -86°C before analysis using LC-MS. Cohort 1 (n = 90) cows were recruited as current first-lactation cows with weekly mobility scores recorded over a 4-mo timeframe, from which newly lame and nonlame cows were identified. Cohort 2 (n = 30) cows were recruited within 3 wk before calving, and lameness events (based on mobility score) were recorded through lactation until a minimum of 70 d in milk (DIM). All cows were matched paired by DIM ± 14 d. The median DIM at lameness identification was 187.5 and 28.5 for cohort 1 and 2, respectively. The best performing machine learning models predicted lameness at the time of lameness with an accuracy of between 81 and 82%. Using stability selection, the prediction accuracy at the time of lameness was 80 to 81%. For samples collected before and after calving, the best performing machine learning model predicted lameness with an accuracy of 71 and 75%, respectively. The findings from this study demonstrate that untargeted LC-MS profiling combined with machine learning methods can be used to predict lameness as early as before calving and before observable changes in gait in first-lactation dairy cows. The methods also provide accuracies for detecting lameness at the time of observable changes in gait of up to 82%. The findings demonstrate that these methods could provide substantial advancements in the early prediction and prevention of lameness risk. Further external validation work is required to confirm these findings are generalizable; however, this study provides the basis from which future work can be conducted.

Incorporating heterogeneity in farmer disease control behaviour into a livestock disease transmission model

Human behaviour is critical to effective responses to livestock disease outbreaks, especially with respect to vaccination uptake. Traditionally, mathematical models used to inform this behaviour have not taken heterogeneity in farmer behaviour into account. We address this by exploring how heterogeneity in farmers vaccination behaviour can be incorporated to inform mathematical models. We developed and used a graphical user interface to elicit farmers (n = 60) vaccination decisions to an unfolding fast-spreading epidemic and linked this to their psychosocial and behavioural profiles. We identified, via cluster analysis, robust patterns of heterogeneity in vaccination behaviour. By incorporating these vaccination behavioural groupings into a mathematical model for a fast-spreading livestock infection, using computational simulation we explored how the inclusion of heterogeneity in farmer disease control behaviour may impact epidemiological and economic focused outcomes. When assuming homogeneity in farmer behaviour versus configurations informed by the psychosocial profile cluster estimates, the modelled scenarios revealed a disconnect in projected distributions and threshold statistics across outbreak size, outbreak duration and economic metrics.

Serum 1H nuclear magnetic resonance-based metabolomics of sole lesion development in Holstein cows

Sole hemorrhage and sole ulcers, referred to as sole lesions, are important causes of lameness in dairy cattle. We aimed to compare the serum metabolome of dairy cows that developed sole lesions in early lactation with that of cows that remained unaffected. We prospectively enrolled a cohort of 1,169 Holstein dairy cows from a single dairy herd and assessed animals at 4 time points: before calving, immediately after calving, early lactation, and late lactation. Sole lesions were recorded by veterinary surgeons at each time point, and serum samples were collected at the first 3 time points. Cases were defined by the presence of sole lesions in early lactation and further subdivided by whether sole lesions had been previously recorded; unaffected controls were randomly selected to match cases. Serum samples from a case-control subset of 228 animals were analyzed with proton nuclear magnetic resonance spectroscopy. Spectral signals, corresponding to 34 provisionally annotated metabolites and 51 unlabeled metabolites, were analyzed in subsets relating to time point, parity cohort, and sole lesion outcome. We used 3 analytic methods (partial least squares discriminant analysis, least absolute shrinkage and selection operator regression, and random forest) to determine the predictive capacity of the serum metabolome and identify informative metabolites. We applied bootstrapped selection stability, triangulation, and permutation to support the inference of variable selection. The average balanced accuracy of class prediction ranged from 50 to 62% depending on the subset. Across all 17 subsets, 20 variables had a high probability of being informative; those with the strongest evidence of being associated with sole lesions corresponded to phenylalanine and 4 unlabeled metabolites. We conclude that the serum metabolome, as characterized by proton nuclear magnetic resonance spectroscopy, does not appear able to predict sole lesion presence or future development of lesions. A small number of metabolites may be associated with sole lesions although, given the poor prediction accuracies, these metabolites are likely to explain only a small proportion of the differences between affected and unaffected animals. Future metabolomic studies may reveal underlying metabolic mechanisms of sole lesion etiopathogenesis in dairy cows; however, the experimental design and analysis need to effectively control for interanimal and extraneous sources of spectral variation.

Hoof lesions in partly housed pasture-based dairy cows

Lameness is a symptom of a painful disorder affecting the limbs, which impacts dairy cow welfare and productivity. Lameness is primarily caused by hoof lesions. The prevalence of different lesion types can differ depending on environmental conditions and farm management practices. The aims of this observational study were to establish the cow-level and herd-level lesion prevalence during both housing and grazing periods in a partly housed, pasture-based system, establish the prevalence of lesions always associated with pain ("alarm" lesion), identify the lesions associated with a higher lameness score, determine relationships between lesions, and identify risk factors for digital dermatitis. On 98 farms during the grazing period and on 74 of the same farms during the housing period, every cow was lameness scored (0-3 lameness scoring scale), and the hind hooves of lame cows (score 2 and 3) were examined (maximum 20 cows per visit) and the prevalence of each lesion type recorded. To gather data on potential predictors for the risk factor analysis, a questionnaire with the farmer was conducted on lameness management practices and infrastructure measurements were taken at each visit. Cow-level data were also collected (e.g., parity, breed, milk yield, and so on). Noninfectious lesions were found to be more prevalent than infectious lesions in this system type. The most prevalent lesion types during both grazing and housing periods were white line separation, sole hemorrhages and overgrown claws; all remaining lesions had a cow-level prevalence of less than 15%. The cow-level prevalence of alarm lesions was 19% during the grazing period and 25% during the housing period; the most prevalent alarm lesion was sole ulcers during both periods. We found significantly more foreign bodies within the hoof sole (grazing = 14%, housing = 7%) and overgrown claws (grazing = 71%, housing = 55%) during the grazing period compared with the housing period. Cows with foul of the foot, sole ulcer, white line abscess, toe necrosis or an amputated claw had higher odds of being more severely lame, compared with mildly lame. The strongest correlation between lesions were between toe necrosis and digital dermatitis (r = 0.40), overgrown claws and corkscrew claws (r = 0.33), and interdigital hyperplasia and digital dermatitis (r = 0.31) at herd level. At the cow level, the strongest correlation was between overgrown claws and corkscrew claws (r = 0.27), and digital dermatitis and heel erosion (r = 0.22). The farmers' perception of the presence of digital dermatitis (and lameness) was significantly correlated with the actual presence of digital dermatitis recorded. Additional risk factors for the presence of digital dermatitis were cow track and verge width near the collecting yard, and stone presence on the cow tracks. Results from this study help further our understanding of the causes of lameness in partly housed, pasture-based dairy cows, and can be used to guide prevention and treatment protocols.

Stability selection for mixed effect models with large numbers of predictor variables: A simulation study

Covariate selection when the number of available variables is large relative to the number of observations is problematic in epidemiology and remains the focus of continued research. Whilst a variety of statistical methods have been developed to attempt to overcome this issue, at present very few methods are available for wide data that include a clustered outcome. The purpose of this research was to make an empirical evaluation of a new method for covariate selection in wide data settings when the dependent variable is clustered. We used 3300 simulated datasets with a variety of defined structures and known sets of true predictor variables to conduct an empirical evaluation of a mixed model stability selection procedure. Comparison was made with an alternative method based on regularisation using the least absolute shrinkage and selection operator (Lasso) penalty. Model performance was assessed using several metrics including the true positive rate (proportion of true covariates selected in a final model) and false discovery rate (proportion of variables selected in a final model that were non-true (false) variables). For stability selection, the false discovery rate was consistently low, generally remaining ≤ 0.02 indicating that on average fewer than 1 in 50 of the variables selected in a final model were false variables. This was in contrast to the Lasso-based method in which the false discovery rate was between 0.59 and 0.72, indicating that generally more than 60% of variables selected in a final model were false variables. In contrast however, the Lasso method attained higher true positive rates than stability selection, although both methods achieved good results. For the Lasso method, true positive rates remained ≥ 0.93 whereas for stability selection the true positive rate was 0.73-0.97. Our results suggest both methods may be of value for covariate selection with high dimensional data with a clustered outcome. When high specificity is needed for identification of true covariates, stability selection appeared to offer the better solution, although with a slight loss of sensitivity. Conversely when high sensitivity is needed, the Lasso approach may be useful, even if accompanied by a substantial loss of specificity. Overall, the results indicated the loss of sensitivity when employing stability selection is relatively small compared to the loss of specificity when using the Lasso and therefore stability selection may provide the better option for the analyst when evaluating data of this type.

Cow- and herd-level risk factors for lameness in partly housed pasture-based dairy cows

Lameness in dairy cows is a major animal welfare concern and has substantial economic impact through reduced production and fertility. Previous risk factor analyses have focused on housed systems, rather than those where cows were grazed for the majority of the year and housed only for the winter period. Therefore, the aim of this observational study was to identify a robust set of cow-level and herd-level risk factors for lameness in a pasture-based system, based on predictors from the housing and grazing periods. Ninety-nine farms were visited during the grazing period (April 2019-September 2019), and 85 farms were revisited during the housing period (October 2019-February 2020). At each visit, all lactating cows were scored for lameness (0 = good mobility, 1 = imperfect mobility, 2 = impaired mobility, 3 = severely impaired mobility), and potential herd-level risk factors were recorded through questionnaires and infrastructure measurements. Routine cow-level management data were also collected. Important risk factors for lameness were derived though triangulation of results from elastic net regression, and from logistic regression model selection using modified Bayesian information criterion. Both selection methods were implemented using bootstrapping. This novel approach has not previously been used in a cow-level or herd-level risk factor analysis in dairy cows, to the authors' knowledge. The binary outcome variable was lameness status, whereby cows with a lameness score of 0 or 1 were classed as non-lame and cows with a score of 2 or 3 were classed as lame. Cow-level risk factors for increased lameness prevalence were age and genetic predicted transmitting ability for lameness. Herd-level risk factors included farm and herd size, stones in paddock gateways, slats on cow tracks near the collecting yard, a sharper turn at the parlor exit, presence of digital dermatitis on the farm, and the farmers' perception of whether lameness was a problem on the farm. This large-scale study identified the most important associations between risk factors and lameness, based on the entire year (grazing and housing periods), providing a focus for future randomized clinical trials.

Multiple model triangulation to identify factors associated with lameness in British sheep flocks

•

Multiple model triangulation identifies variables that are likely true positives.

•

Triangulation increases confidence in which managements to recommend in practice.

•

Effective management of ewes can lower prevalence of lameness in ewes and lambs.

Identification of factors associated with an outcome can be challenging when the number of explanatory variables is large in relation to the number of observations. Multiple model triangulation, where results from several model types are combined, improves the likelihood of identifying true predictor variables. The aim of this study was to use triangulation to identify covariates likely to be truly associated with the prevalence of lameness in sheep flocks in Great Britain.

Data were collected using a questionnaire sent to 3200 sheep farmers in Great Britain in 2018. The useable response rate was 14.1 %. The geometric mean prevalence of lameness was 1.4 % (95 % CI 1.2−1.7) for ewes, and 0.6 % (95 % CI 0.5−0.9) for lambs, however, approximately 60 % flocks had >2% prevalence of lameness in ewes.

Four model types were investigated, two generalised linear models (negative binomial and quasi-Poisson) built using stepwise selection, and two elastic net models (Poisson and Gaussian distributions) refined with selection stability estimation.

Triangulated covariates were those selected in three or all four models – 10 for ewes and 12 for lambs. Higher prevalence of lameness in ewes was associated with 5−100% feet bleeding during routine foot trimming compared with not foot trimming, footbathing the flock to treat severe footrot (SFR) and always using formalin in footbaths, both compared with not footbathing, using FootVax™ for <1 year compared with not using FootVax™, and never quarantining new or returning sheep to the farm for >3 weeks compared with always. Lower prevalence of lameness in ewes was associated with vaccinating with FootVax™ for >5 years compared with not vaccinating, peat soil compared with no peat soil, and having no lame ewes to treat.

Higher prevalence of lameness in lambs was associated with 5−100% feet bleeding during routine foot trimming, always foot trimming ewes with SFR, not knowingly selecting replacement ewes from ewes that were never lame compared with always, replacement sheep purchased and homebred compared with only homebred, treating lambs >3 days after recognition of lameness compared with 0-3 days and footbathing the flock to treat interdigital dermatitis compared with not footbathing at all. Lower prevalence of lameness in lambs was associated with peat soil, flocks in Scotland versus England, an altitude of >230−500 m compared with ≤230 m, never using antibiotic injection to treat lambs with SFR compared with always, and having no lame lambs to treat.

We conclude triangulation identified reliable management practices for farmers to implement to minimise lameness in sheep.