Cost-sensitive ordinal classification methods to predict SARS-CoV-2 pneumonia severity.

OBJECTIVE: To study the suitability of costsensitive ordinal artificial intelligence-machine learning (AIML) strategies in the prognosis of SARS-CoV-2 pneumonia severity. MATERIALS & METHODS: Observational, retrospective, longitudinal, cohort study in 4 hospitals in Spain. Information regarding demographic and clinical status was supplemented by socioeconomic data and air pollution exposures. We proposed AI-ML algorithms for ordinal classification via ordinal decomposition and for cost-sensitive learning via resampling techniques. For performance-based model selection, we defined a custom score including per-class sensitivities and asymmetric misprognosis costs. 260 distinct AI-ML models were evaluated via 10 repetitions of 5×5 nested cross-validation with hyperparameter tuning. Model selection was followed by the calibration of predicted probabilities. Final overall performance was compared against five well-established clinical severity scores and against a 'standard' (non-cost sensitive, non-ordinal) AI-ML baseline. In our best model, we also evaluated its explainability with respect to each of the input variables. RESULTS: The study enrolled n = 1548 patients: 712 experienced low, 238 medium, and 598 high clinical severity. d = 131 variables were collected, becoming d ' = 148 features after categorical encoding. Model selection resulted in our best-performing AI-ML pipeline having: a) no imputation of missing data, b) no feature selection (i.e. using the full set of d ' features), c) 'Ordered Partitions' ordinal decomposition, d) cost-based reimbalance, and e) a Histogram-based Gradient Boosting classifier. This best model (calibrated) obtained a median accuracy of 68.1% [67.3%, 68.8%] (95% confidence interval), a balanced accuracy of 57.0% [55.6%, 57.9%], and an overall area under the curve (AUC) 0.802 [0.795, 0.808]. In our dataset, it outperformed all five clinical severity scores and the 'standard' AI-ML baseline. DISCUSSION & CONCLUSION: We conducted an exhaustive exploration of AI-ML methods designed for both ordinal and cost-sensitive classification, motivated by a real-world application domain (clinical severity prognosis) in which these topics arise naturally. Our model with the best classification performance exploited successfully the ordering information of ground truth classes, coping with imbalance and asymmetric costs. However, these ordinal and cost-sensitive aspects are seldom explored in the literature.

Extracting relevant predictive variables for COVID-19 severity prognosis: An exhaustive comparison of feature selection techniques.

With the COVID-19 pandemic having caused unprecedented numbers of infections and deaths, large research efforts have been undertaken to increase our understanding of the disease and the factors which determine diverse clinical evolutions. Here we focused on a fully data-driven exploration regarding which factors (clinical or otherwise) were most informative for SARS-CoV-2 pneumonia severity prediction via machine learning (ML). In particular, feature selection techniques (FS), designed to reduce the dimensionality of data, allowed us to characterize which of our variables were the most useful for ML prognosis. We conducted a multi-centre clinical study, enrolling n = 1548 patients hospitalized due to SARS-CoV-2 pneumonia: where 792, 238, and 598 patients experienced low, medium and high-severity evolutions, respectively. Up to 106 patient-specific clinical variables were collected at admission, although 14 of them had to be discarded for containing ⩾60% missing values. Alongside 7 socioeconomic attributes and 32 exposures to air pollution (chronic and acute), these became d = 148 features after variable encoding. We addressed this ordinal classification problem both as a ML classification and regression task. Two imputation techniques for missing data were explored, along with a total of 166 unique FS algorithm configurations: 46 filters, 100 wrappers and 20 embeddeds. Of these, 21 setups achieved satisfactory bootstrap stability (⩾0.70) with reasonable computation times: 16 filters, 2 wrappers, and 3 embeddeds. The subsets of features selected by each technique showed modest Jaccard similarities across them. However, they consistently pointed out the importance of certain explanatory variables. Namely: patient's C-reactive protein (CRP), pneumonia severity index (PSI), respiratory rate (RR) and oxygen levels -saturation Sp O2, quotients Sp O2/RR and arterial Sat O2/Fi O2-, the neutrophil-to-lymphocyte ratio (NLR) -to certain extent, also neutrophil and lymphocyte counts separately-, lactate dehydrogenase (LDH), and procalcitonin (PCT) levels in blood. A remarkable agreement has been found a posteriori between our strategy and independent clinical research works investigating risk factors for COVID-19 severity. Hence, these findings stress the suitability of this type of fully data-driven approaches for knowledge extraction, as a complementary to clinical perspectives.

Modelling seabirds biodiversity through Bayesian Spatial Beta regression models: A proxy to inform marine protected areas in the Mediterranean Sea.

Seabirds are bioindicators of marine ecosystems health and one of the world's most endangered avian groups. The creation of marine protected areas plays an important role in the conservation of marine environment and its biodiversity. The distributions of top predators, as seabirds, have been commonly used for the management and creation of these figures of protection. The main objective of this study is to investigate seabirds biodiversity distribution in the Mediterranean Sea through the use of Bayesian spatial Beta regression models. We used an extensive historical database of at-sea locations of 19 different seabird species as well as geophysical, climatology variables and cumulative anthropogenic threats to model species biodiversity. We found negative associations between seabirds biodiversity and distance to the coast as well as concavity of the seabed, and positive with chlorophyll and slope. Further, a positive association was found between seabirds biodiversity and coastal impact. In this study we define as hot spot of seabird biodiversity those areas with a posterior predictive mean over 0.50. We found potential hot spots in the Mediterranean Sea which do not overlap with the existing MPASs and marine IBAs. Specifically, our hot spots areas do not overlap with the 52.04% and 16.87% of the current MPAs and marine IBAs, respectively. Overall, our study highlights the need for the extension of spatial prioritization of conservation areas to seabirds biodiversity, addressing the challenges of establishing transboundary governance.

A mesoscale analysis of relations between fish species richness and environmental and anthropogenic pressures in the Mediterranean Sea.

Although there is a great knowledge about individual anthropogenic threats to different fish species in the Mediterranean Sea, little is known about how these threats accumulate and interact to affect fish species richness in conjunction with environmental dynamics. This study assesses the role of these threats in the fish richness component and identifies the main areas where the interaction between fish species richness and threats is highest. Our results show that fish richness seems to be higher in saltier and colder areas where the chlorophyll-a and phosphate concentrations are lower. Among the anthropogenic threats analyzed, the costal impact and the fishing effort seems to be the more relevant ones. Overall areas with high fish richness are mainly located along the western and northern shores, with lower values in the south-eastern regions. Areas of potential high cumulative threats are widespread in both the western and eastern basins, with fewer areas located in the south-eastern region. By describing the spatial patterns of the fish richness and which drivers explain these patterns we can also identify which anthropogenic activities can be managed more effectively to maintain and restore marine fish biodiversity in the basin.

Modeling Inoculum Availability of Plurivorosphaerella nawae in Persimmon Leaf Litter with Bayesian Beta Regression.

Circular leaf spot (CLS), caused by Plurivorosphaerella nawae, is a serious disease affecting persimmon (Diospyros kaki) that is characterized by necrotic lesions on leaves, defoliation, and fruit drop. Under Mediterranean conditions, P. nawae forms pseudothecia in the leaf litter in winter, and ascospores are released in spring, infecting susceptible leaves. Persimmon growers are advised to apply fungicides for CLS control during the period of inoculum availability, which was previously defined based on ascospore counts under the microscope. A model of inoculum availability of P. nawae was developed and evaluated as an alternative to ascospore counts. Leaf litter samples were collected weekly in L'Alcúdia (Spain) from 2010 to 2015. Leaves were soaked and placed in a wind tunnel, and the released ascospores of P. nawae were counted. Hierarchical Bayesian beta regression methods were used to model the dynamics of ascospore production in the leaf litter. The selected model included accumulated degree-days (ADDs) and ADDs taking into account the vapor pressure deficit (ADDvpd) as fixed effects and year as random effect. This model had a mean absolute error of 0.042 and a root mean square error of 0.062. The beta regression model was evaluated in four orchards from 2010 to 2015. Higher accuracy was obtained at the beginning and the end of the ascospore production period, which are the events of interest to schedule fungicide sprays for CLS control in Spain. This same modeling framework can be extended to other fungal plant pathogens whose inoculum dynamics are expressed as proportion data.[Formula: see text] Copyright © 2021 The Author(s). This is an open access article distributed under the CC BY 4.0 International license.

Spatial Bayesian Modeling Applied to the Surveys of Xylella fastidiosa in Alicante (Spain) and Apulia (Italy).

The plant-pathogenic bacterium Xylella fastidiosa was first reported in Europe in 2013, in the province of Lecce, Italy, where extensive areas were affected by the olive quick decline syndrome, caused by the subsp. pauca. In Alicante, Spain, almond leaf scorch, caused by X. fastidiosa subsp. multiplex, was detected in 2017. The effects of climatic and spatial factors on the geographic distribution of X. fastidiosa in these two infested regions in Europe were studied. The presence/absence data of X. fastidiosa in the official surveys were analyzed using Bayesian hierarchical models through the integrated nested Laplace approximation (INLA) methodology. Climatic covariates were obtained from the WorldClim v.2 database. A categorical variable was also included according to Purcell's minimum winter temperature thresholds for the risk of occurrence of Pierce's disease of grapevine, caused by X. fastidiosa subsp. fastidiosa. In Alicante, data were presented aggregated on a 1 km grid (lattice data), where the spatial effect was included in the model through a conditional autoregressive structure. In Lecce, data were observed at continuous locations occurring within a defined spatial domain (geostatistical data). Therefore, the spatial effect was included via the stochastic partial differential equation approach. In Alicante, the pathogen was detected in all four of Purcell's categories, illustrating the environmental plasticity of the subsp. multiplex. Here, none of the climatic covariates were retained in the selected model. Only two of Purcell's categories were represented in Lecce. The mean diurnal range (bio2) and the mean temperature of the wettest quarter (bio8) were retained in the selected model, with a negative relationship with the presence of the pathogen. However, this may be due to the heterogeneous sampling distribution having a confounding effect with the climatic covariates. In both regions, the spatial structure had a strong influence on the models, but not the climatic covariates. Therefore, pathogen distribution was largely defined by the spatial relationship between geographic locations. This substantial contribution of the spatial effect in the models might indicate that the current extent of X. fastidiosa in the study regions had arisen from a single focus or from several foci, which have been coalesced.

A Decision Support System Based on Degree-Days to Initiate Fungicide Spray Programs for Peach Powdery Mildew in Catalonia, Spain.

The incidence of peach powdery mildew (PPM) on fruit was monitored in commercial peach orchards to i) describe the disease progress in relation to several environmental parameters and ii) establish an operating threshold to initiate a fungicide spray program based on accumulated degree-day (ADD) data. A beta-regression model for disease incidence showed a substantial contribution of the random effects orchard and year, whereas relevant fixed effects corresponded to ADD, wetness duration, and ADD considering vapor pressure deficit and rain. When beta-regression models were fitted for each orchard and year considering only ADD, disease onset was observed at 242 ± 13 ADD and symptoms did not develop further after 484 ± 42 ADD. An operating threshold to initiate fungicide applications was established at 220 ADD, coinciding with a PPM incidence in fruit around 0.05. A validation was further conducted by comparing PPM incidence in i) a standard, calendar-based program, ii) a program with applications initiated at 220 ADD, and iii) a nontreated control. A statistically relevant reduction in disease incidence in fruit was obtained with both fungicide programs, from 0.244 recorded in the control to 0.073 with the 220-ADD alert program, and 0.049 with the standard program. The 220-ADD alert program resulted in 33% reduction in fungicide applications.

Ecological, genetic and evolutionary drivers of regional genetic differentiation in Arabidopsis thaliana.

BACKGROUND: Disentangling the drivers of genetic differentiation is one of the cornerstones in evolution. This is because genetic diversity, and the way in which it is partitioned within and among populations across space, is an important asset for the ability of populations to adapt and persist in changing environments. We tested three major hypotheses accounting for genetic differentiation-isolation-by-distance (IBD), isolation-by-environment (IBE) and isolation-by-resistance (IBR)-in the annual plant Arabidopsis thaliana across the Iberian Peninsula, the region with the largest genomic diversity. To that end, we sampled, genotyped with genome-wide SNPs, and analyzed 1772 individuals from 278 populations distributed across the Iberian Peninsula. RESULTS: IBD, and to a lesser extent IBE, were the most important drivers of genetic differentiation in A. thaliana. In other words, dispersal limitation, genetic drift, and to a lesser extent local adaptation to environmental gradients, accounted for the within- and among-population distribution of genetic diversity. Analyses applied to the four Iberian genetic clusters, which represent the joint outcome of the long demographic and adaptive history of the species in the region, showed similar results except for one cluster, in which IBR (a function of landscape heterogeneity) was the most important driver of genetic differentiation. Using spatial hierarchical Bayesian models, we found that precipitation seasonality and topsoil pH chiefly accounted for the geographic distribution of genetic diversity in Iberian A. thaliana. CONCLUSIONS: Overall, the interplay between the influence of precipitation seasonality on genetic diversity and the effect of restricted dispersal and genetic drift on genetic differentiation emerges as the major forces underlying the evolutionary trajectory of Iberian A. thaliana.

A hierarchical Bayesian Beta regression approach to study the effects of geographical genetic structure and spatial autocorrelation on species distribution range shifts.

Global climate change (GCC) may be causing distribution range shifts in many organisms worldwide. Multiple efforts are currently focused on the development of models to better predict distribution range shifts due to GCC. We addressed this issue by including intraspecific genetic structure and spatial autocorrelation (SAC) of data in distribution range models. Both factors reflect the joint effect of ecoevolutionary processes on the geographical heterogeneity of populations. We used a collection of 301 georeferenced accessions of the annual plant Arabidopsis thaliana in its Iberian Peninsula range, where the species shows strong geographical genetic structure. We developed spatial and nonspatial hierarchical Bayesian models (HBMs) to depict current and future distribution ranges for the four genetic clusters detected. We also compared the performance of HBMs with Maxent (a presence-only model). Maxent and nonspatial HBMs presented some shortcomings, such as the loss of accessions with high genetic admixture in the case of Maxent and the presence of residual SAC for both. As spatial HBMs removed residual SAC, these models showed higher accuracy than nonspatial HBMs and handled the spatial effect on model outcomes. The ease of modelling and the consistency among model outputs for each genetic cluster was conditioned by the sparseness of the populations across the distribution range. Our HBMs enrich the toolbox of software available to evaluate GCC-induced distribution range shifts by considering both genetic heterogeneity and SAC, two inherent properties of any organism that should not be overlooked.