The usefulness of genomic prediction in crop and livestock breeding programs has prompted efforts to develop new and improved genomic prediction algorithms, such as artificial neural networks and gradient tree boosting. However, the performance of these algorithms has not been compared in a systematic manner using a wide range of datasets and models. Using data of 18 traits across six plant species with different marker densities and training population sizes, we compared the performance of six linear and six non-linear algorithms. First, we found that hyperparameter selection was necessary for all non-linear algorithms and that feature selection prior to model training was critical for artificial neural networks when the markers greatly outnumbered the number of training lines. Across all species and trait combinations, no one algorithm performed best, however predictions based on a combination of results from multiple algorithms (i.e., ensemble predictions) performed consistently well. While linear and non-linear algorithms performed best for a similar number of traits, the performance of non-linear algorithms vary more between traits. Although artificial neural networks did not perform best for any trait, we identified strategies (i.e., feature selection, seeded starting weights) that boosted their performance to near the level of other algorithms. Our results highlight the importance of algorithm selection for the prediction of trait values.

The Expanded Disability Status Scale (EDSS) is the current 'gold standard' for monitoring disease severity in multiple sclerosis (MS). The EDSS is a physician-based assessment. A patient-related surrogate for the EDSS may be useful in remotely capturing information. Eighty-one patients (EDSS range 0-8) having EDSS as part of clinical trials were recruited. All patients carried out the web-based survey with minimal assistance. Full EDSS scores were available for 78 patients. The EDSS scores were compared to those generated by the online survey using analysis of variance, matched pair test, Pearson's coefficient, weighted kappa coefficient, and the intra-class correlation coefficient. The internet-based EDSS scores showed good correlation with the physician-measured assessment (Pearson's coefficient = 0.85). Weighted kappa for full agreement was 0.647. Full agreement was observed in 20 patients who had EDSS scores ranging from 0 to 6; many of those with 100 % agreement had scores of 5.5-6 (n = 8).The intra-class coefficient was 0.844 overall for all cases. Internet-based FS and EDSS show good agreement with physician-measured scores. Agreement was better in patients with higher scores. Overall patient satisfaction with the web-based assessment was high. An internet-based assessment tool is likely to prove an invaluable tool in the long-term monitoring in MS.

The objectives of this study were to; quantify positional differences in the activity profiles of Gaelic football players and to evaluate decrements in physical performance during a pre-season competition. Global positioning system (GPS) data was recorded from 36 players from 3 teams across 5 games. The relative distance covered in locomotor activities, peak speed, relative PlayerLoad™ (PL.min-1) and heart rate responses were evaluated between playing positions and across match periods using a mixed model analysis. The mean relative distance of 92.4 ± 23.3 m.min-1 covered, comprised 28.4 ± 10.2 m.min-1 of high intensity running (m.min-1 ≥ 4.0 m.s-1) and 9.9 ± 3.9 m.min-1 of very high intensity running (m.min-1 ≥ 5.5 m.s-1). High intensity running and relative PlayerLoad™ (PL.min-1) was significantly higher in half-backs, midfielders and half-forwards compared to the fullbacks , whereas only the half-backs and half-forwards displayed significantly greater values compared to full-forwards. When compared to the first 15 min (P1) of the game, analysis of pooled positional data revealed significant declines in; overall relative distance covered, jogging (≥2.0-< 4.0 m.s-1), running (≥4.0-<5.5 m.s-1), high intensity running and PL.min-1 ,in P2 (20-35 min) and P4 (55-70 min). Significant reductions in average heart rate were also found between the first and second halves and between P1 with both P3 and P4. These results highlight differences in the physical performance requirements of specific positions and provide evidence of reductions in work-rate during games. Coaches can use this information to inform the development of both team and position-specific conditioning programmes.

The objective of this study was to identify performance indicators which differentiated between winning and losing elite Gaelic football teams. Eighty three technical and tactical performance variables were measured in 13 teams during 26 league and championship games throughout 2014-15. Univariate analysis of full-games revealed that winners achieved a significantly higher total score, number of scores, shots, points, points from play and goals, resulting in superior shot efficiency, average attack per score, and scores per 10 possessions. Winners gained significantly more turnovers and completed significantly less unsuccessful hand passes. Winners also performed significantly less kick outs, resulting in fewer successful kick outs and successful dead ball kick passes overall. A principal component analysis, conducted on 18 variables produced 4 components, which explained 81.9% of the variance. Both logistic regression (8.00, χ 2 (1) = 16.00, p < 0.001) and discriminant analysis (Ʌ = 0.53, χ 2 (1) = 13.77, p < 0.001) revealed that 1 component; defensive counterattacking, significantly contributed to outcome and differentiated winners from losers with a cross-validation accuracy of 87.5%. Coaches can use this information to organise their defensive system to generate opposition turnovers and also incorporate sufficient flexibility to facilitate effective transitions to exploit their own offensive counterattacking opportunities.

The usefulness of Genomic Prediction (GP) in crop and livestock breeding programs has led to efforts to develop new and improved GP approaches including non-linear algorithm, such as artificial neural networks (ANN) (i.e. deep learning) and gradient tree boosting. However, the performance of these algorithms has not been compared in a systematic manner using a wide range of GP datasets and models. Using data of 18 traits across six plant species with different marker densities and training population sizes, we compared the performance of six linear and five non-linear algorithms, including ANNs. First, we found that hyperparameter selection was critical for all non-linear algorithms and that feature selection prior to model training was necessary for ANNs when the markers greatly outnumbered the number of training lines. Across all species and trait combinations, no one algorithm performed best, however predictions based on a combination of results from multiple GP algorithms (i.e. ensemble predictions) performed consistently well. While linear and non-linear algorithms performed best for a similar number of traits, the performance of non-linear algorithms vary more between traits than that of linear algorithms. Although ANNs did not perform best for any trait, we identified strategies (i.e. feature selection, seeded starting weights) that boosted their performance near the level of other algorithms. These results, together with the fact that even small improvements in GP performance could accumulate into large genetic gains over the course of a breeding program, highlights the importance of algorithm selection for the prediction of trait values.