Evaluating the performance of machine learning methods and variable selection methods for predicting difficult-to-measure traits in Holstein dairy cattle using milk infrared spectral data

Fourier-transform infrared (FTIR) spectroscopy is a powerful high-throughput phenotyping tool for predicting traits that are expensive and difficult to measure in dairy cattle. Calibration equations are often developed using standard methods, such as partial least squares (PLS) regression. Methods that employ penalization, rank-reduction, and variable selection, as well as being able to model the nonlinear relations between phenotype and FTIR, might offer improvements in predictive ability and model robustness. This study aimed to compare the predictive ability of 2 machine learning methods, namely random forest (RF) and gradient boosting machine (GBM), and penalized regression against PLS regression for predicting 3 phenotypes differing in terms of biological meaning and relationships with milk composition (i.e., phenotypes measurable directly and not directly in milk, reflecting different biological processes which can be captured using milk spectra) in Holstein-Friesian cattle under 2 cross-validation scenarios. The data set comprised phenotypic information from 471 Holstein-Friesian cows, and 3 target phenotypes were evaluated: (1) body condition score (BCS), (2) blood β-hydroxybutyrate (BHB, mmol/L), and (3) κ-casein expressed as a percentage of nitrogen (κ-CN, % N). The data set was split considering 2 cross-validation scenarios: samples-out random in which the population was randomly split into 10-folds (8-folds for training and 1-fold for validation and testing); and herd/date-out in which the population was randomly assigned to training (70% herd), validation (10%), and testing (20% herd) based on the herd and date in which the samples were collected. The random grid search was performed using the training subset for the hyperparameter optimization and the validation set was used for the generalization of prediction error. The trained model was then used to assess the final prediction in the testing subset. The grid search for penalized regression evidenced that the elastic net (EN) was the best regularization with increase in predictive ability of 5%. The performance of PLS (standard model) was compared against 2 machine learning techniques and penalized regression using 2 cross-validation scenarios. Machine learning methods showed a greater predictive ability for BCS (0.63 for GBM and 0.61 for RF), BHB (0.80 for GBM and 0.79 for RF), and κ-CN (0.81 for GBM and 0.80 for RF) in samples-out cross-validation. Considering a herd/date-out cross-validation these values were 0.58 (GBM and RF) for BCS, 0.73 (GBM and RF) for BHB, and 0.77 (GBM and RF) for κ-CN. The GBM model tended to outperform other methods in predictive ability around 4%, 1%, and 7% for EN, RF, and PLS, respectively. The prediction accuracies of the GBM and RF models were similar, and differed statistically from the PLS model in samples-out random cross-validation. Although, machine learning techniques outperformed PLS in herd/date-out cross-validation, no significant differences were observed in terms of predictive ability due to the large standard deviation observed for predictions. Overall, GBM achieved the highest accuracy of FTIR-based prediction of the different phenotypic traits across the cross-validation scenarios. These results indicate that GBM is a promising method for obtaining more accurate FTIR-based predictions for different phenotypes in dairy cattle.     

Short communication: Comparison of growth kinetics at different temperatures of Streptococcus macedonicus and Streptococcus thermophilus strains of dairy origin

Within the genus Streptococcus, S. thermophilus and S. macedonicus are the 2 known species related to foods. Streptococci are widely used as starter cultures to rapidly lower milk pH. As S. macedonicus has been introduced quite recently, much less information is available on its technological potential. Because temperature is an important factor in fermented food production, we compared the growth kinetics over 24 h of 8 S. thermophilus and 7 S. macedonicus strains isolated from various dairy environments in Italy, at 4 temperatures, 30°C, 34°C, 37°C and 42°C. We used the Gompertz model to estimate the 3 main growth parameters; namely, lag phase duration (λ), maximum growth rate (µ_max), and maximum cell number at the stationary phase (N_max). Our results showed significant differences in average growth kinetics between the 2 species. Among the strains tested, 37°C appeared to be the optimal temperature for the growth of both species, particularly for S. macedonicus strains, which showed mean shorter lag phases and higher cell numbers compared with S. thermophilus. Overall, the growth curves of S. macedonicus strains were more similar to each other whereas S. thermophilus strains grew very differently. These results help to better define and compare technological characteristics of the 2 species, in view of the potential use of S. macedonicus in place of S. thermophilus in selected technological applications.     

Soft-Seeded Pomegranate (<Emphasis Type="Italic">Punica granatum</Emphasis> L.) Varieties: Preliminary Characterization and Quality Changes of Minimally Processed Arils During Storage

A preliminary characterization of two new soft-seeded pomegranate varieties (MR-100® and KINGDOM®) based on their main physico-chemical and nutritional parameters was reported. The two varieties showed significant differences (p ≤ 0.05) in polyphenols, anthocyanins and antioxidant activity. Kingdom pomegranate had higher polyphenols (2524.73 mg GAE/L), anthocyanins (752.49 mg C₃gE/L) and antioxidant activity (EC₅₀ 13.58 μL/mL) than MR-100 (1792.74 mg GAE/L, 141.29 mg C₃gE/L and EC₅₀ 47.53 μL/mL, respectively). Moreover, minimally processed arils of the two varieties were packaged in semipermeable and micro-perforated film at 5 °C, and the quality changes that occurred during storage condition (15 days) were investigated. During storage, Kingdom arils exhibited a better performance in terms of antioxidant activity, polyphenols and anthocyanin content with respect to MR-100. Furthermore, the packaging systems did not affect the estimated quality parameters for both varieties. Based on the sensory evaluation and microbial counts, both aril varieties reached, at 15-day storage, suitable values for commercial purpose.     

On-road emissions of light-duty vehicles in europe

For obtaining type approval in the European Union, light-duty vehicles have to comply with emission limits during standardized laboratory emissions testing. Although emission limits have become more stringent in past decades, light-duty vehicles remain an important source of nitrogen oxides and carbon monoxide emissions in Europe. Furthermore, persisting air quality problems in many urban areas suggest that laboratory emissions testing may not accurately capture the on-road emissions of light-duty vehicles. To address this issue, we conduct the first comprehensive on-road emissions test of light-duty vehicles with state-of-the-art Portable Emission Measurement Systems. We find that nitrogen oxides emissions of gasoline vehicles as well as carbon monoxide and total hydrocarbon emissions of both diesel and gasoline vehicles generally remain below the respective emission limits. By contrast, nitrogen oxides emissions of diesel vehicles (0.93 ± 0.39 grams per kilometer [g/km]), including modern Euro 5 diesel vehicles (0.62 ± 0.19 g/km), exceed emission limits by 320 ± 90%. On-road carbon dioxide emissions surpass laboratory emission levels by 21 ± 9%, suggesting that the current laboratory emissions testing fails to accurately capture the on-road emissions of light-duty vehicles. Our findings provide the empirical foundation for the European Commission to establish a complementary emissions test procedure for light-duty vehicles. This procedure could be implemented together with more stringent Euro 6 emission limits in 2014. The envisaged measures should improve urban air quality and provide incentive for innovation in the automotive industry.     

Beer Clarification by Novel Ceramic Hollow‐Fiber Membranes: Effect of Pore Size on Product Quality

In this work, the crossflow microfiltration performance of rough beer samples was assessed using ceramic hollow‐fiber (HF) membrane modules with a nominal pore size ranging from 0.2 to 1.4 μm. Under constant operating conditions (that is, transmembrane pressure difference, TMP = 2.35 bar; feed superficial velocity, v_S = 2.5 m/s; temperature, T = 10 °C), quite small steady‐state permeation fluxes (J^*) of 32 or 37 L/m²/h were achieved using the 0.2‐ or 0.5‐μm symmetric membrane modules. Both permeates exhibited turbidity <1 EBC unit, but a significant reduction in density, viscosity, color, extract, and foam half‐life with respect to their corresponding retentates. The 0.8‐μm asymmetric membrane module might be selected, its corresponding permeate having quite a good turbidity and medium reduction in the aforementioned beer quality parameters. Moreover, it exhibited J^* values of the same order of magnitude of those claimed for the polyethersulfone HF membrane modules currently commercialized. The 1.4‐μm asymmetric membrane module yielded quite a high steady‐state permeation flux (196 ± 38 L/m²/h), and a minimum decline in permeate quality parameters, except for the high levels of turbidity at room temperature and chill haze. In the circumstances, such a membrane module might be regarded as a real valid alternative to conventional powder filters on condition that the resulting permeate were submitted to a final finishing step using 0.45‐ or 0.65‐μm microbially rated membrane cartridges prior to aseptic bottling. A novel combined beer clarification process was thus outlined.     

The Unbearable Shallow Understanding of Deep Learning

This paper analyzes the rapid and unexpected rise of deep learning within Artificial Intelligence and its applications. It tackles the possible reasons for this remarkable success, providing candidate paths towards a satisfactory explanation of why it works so well, at least in some domains. A historical account is given for the ups and downs, which have characterized neural networks research and its evolution from “shallow” to “deep” learning architectures. A precise account of “success” is given, in order to sieve out aspects pertaining to marketing or sociology of research, and the remaining aspects seem to certify a genuine value of deep learning, calling for explanation. The alleged two main propelling factors for deep learning, namely computing hardware performance and neuroscience findings, are scrutinized, and evaluated as relevant but insufficient for a comprehensive explanation. We review various attempts that have been made to provide mathematical foundations able to justify the efficiency of deep learning, and we deem this is the most promising road to follow, even if the current achievements are too scattered and relevant for very limited classes of deep neural models. The authors’ take is that most of what can explain the very nature of why deep learning works at all and even very well across so many domains of application is still to be understood and further research, which addresses the theoretical foundation of artificial learning, is still very much needed.

     

Malignant Transformation of Giant Cell Tumour of Bone: A Review of Literature and the Experience of a Referral Centre

Giant cell tumour of bone (GCTB) is a benign, locally aggressive primary bone neoplasm that represents 5% of all bone tumours. The principal treatment approach is surgery. Although generally GCTB is considered only a locally aggressive disease, it can metastasise, and lung metastases occur in 1–9% of patients. To date, only the use of denosumab has been approved as medical treatment for GCTB. Even more rarely, GCTB undergoes sarcomatous transformation into a malignant tumour (4% of all GCTB), but history of this malignant transformation is unclear and unpredictable. Considering the rarity of the event, the data in the literature are few. In this review, we summarise published data of GCTB malignant transformation and we analyse three cases of malignant transformation of GCTB, evaluating histopathology, genetics, and radiological aspects. Despite the rarity of this event, we conclude that a strict follow up is recommended to detect early malignant transformation.     

Accuracy of FIB-4 to Detect Elevated Liver Stiffness Measurements in Patients with Non-Alcoholic Fatty Liver Disease: A Cross-Sectional Study in Referral Centers

The identification of advanced fibrosis by applying noninvasive tests is still a key component of the diagnostic algorithm of NAFLD. The aim of this study is to assess the concordance between the FIB-4 and liver stiffness measurement (LSM) in patients referred to two liver centers for the ultrasound-based diagnosis of NAFLD. Fibrosis 4 Index for Liver Fibrosis (FIB-4) and LSM were assessed in 1338 patients. A total of 428 (32%) had an LSM ≥ 8 kPa, whereas 699 (52%) and 113 (9%) patients had an FIB-4 < 1.3 and >3.25, respectively. Among 699 patients with an FIB-4 < 1.3, 118 (17%) had an LSM ≥ 8 kPa (false-negative FIB-4). This proportion was higher in patients ≥60 years, with diabetes mellitus (DM), arterial hypertension or a body mass index (BMI) ≥ 27 kg/m². In multiple adjusted models, age ≥ 60 years (odds ratio (OR) = 1.96, 95% confidence interval (CI) 1.19–3.23)), DM (OR = 2.59, 95% CI 1.63–4.13), body mass index (BMI) ≥ 27 kg/m² (OR = 2.17, 95% CI 1.33–3.56) and gamma-glutamyltransferase ≥ 25 UI/L (OR = 2.68, 95% CI 1.49–4.84) were associated with false-negative FIB-4. The proportion of false-negative FIB-4 was 6% in patients with none or one of these risk factors and increased to 16, 31 and 46% among those with two, three and four concomitant risk factors, respectively. FIB-4 is suboptimal to identify patients to refer to liver centers, because about one-fifth may be false negative at FIB-4, having instead an LSM ≥ 8 KPa.