Model evaluation and spatial interpolation by Bayesian combination of observations with outputs from numerical models

Constructing maps of dry deposition pollution levels is vital for air quality management, and presents statistical problems typical of many environmental and spatial applications. Ideally, such maps would be based on a dense network of monitoring stations, but this does not exist. Instead, there are two main sources of information for dry deposition levels in the United States: one is pollution measurements at a sparse set of about 50 monitoring stations called CASTNet, and the other is the output of the regional scale air quality models, called Models-3. A related problem is the evaluation of these numerical models for air quality applications, which is crucial for control strategy selection. We develop formal methods for combining sources of information with different spatial resolutions and for the evaluation of numerical models. We specify a simple model for both the Models-3 output and the CASTNet observations in terms of the unobserved ground truth, and we estimate the model in a Bayesian way. This provides improved spatial prediction via the posterior distribution of the ground truth, allows us to validate Models-3 via the posterior predictive distribution of the CASTNet observations, and enables us to remove the bias in the Models-3 output. We apply our methods to data on SO2 concentrations, and we obtain high-resolution SO2 distributions by combining observed data with model output. We also conclude that the numerical models perform worse in areas closer to power plants, where the SO2 values are overestimated by the models.     

Informing management decisions for ecological networks,using dynamic models calibrated to noisy time‐series data

Well‐intentioned environmental management can backfire, causing unforeseen damage. To avoid this, managers and ecologists seek accurate predictions of the ecosystem‐wide impacts of interventions, given small and imprecise datasets, which is an incredibly difficult task. We generated and analysed thousands of ecosystem population time series to investigate whether fitted models can aid decision‐makers to select interventions. Using these time‐series data (sparse and noisy datasets drawn from deterministic Lotka‐Volterra systems with two to nine species, of known network structure), dynamic model forecasts of whether a species’ future population will be positively or negatively affected by rapid eradication of another species were correct > 70% of the time. Although 70% correct classifications is only slightly better than an uninformative prediction (50%), this classification accuracy can be feasibly improved by increasing monitoring accuracy and frequency. Our findings suggest that models may not need to produce well‐constrained predictions before they can inform decisions that improve environmental outcomes.     

A scalable deep learning system for monitoring and forecasting pollutant concentration levels on UK highways

The construction of intercity highways by the government has resulted in a progressive increase in vehicle emissions and pollution from noise, dust, and vibrations despite its recognition of the air pollution menace. Efforts that have targeted roadside pollution still do not accurately monitor deadly pollutants such as nitrogen oxides and particulate matter. Reports on regional highways across the country are based on a limited number of fixed monitoring stations that are sometimes located far from the highway. These periodic and coarse-grained measurements cause inefficient highway air quality reporting, leading to inaccurate air quality forecasts. This paper, therefore, proposes and validates a scalable deep learning framework for efficiently capturing fine-grained highway data and forecasting future concentration levels. Highways in four different UK regions - Newport, Lewisham, Southwark, and Chepstow were used as case studies to develop a REVIS system and validate the proposed framework. REVIS examined the framework's ability to capture granular pollution data, scale up its storage facility to rapid data growth and translate high-level user queries to structured query language (SQL) required for exploratory data analysis. Finally, the framework's suitability for predictive analytics was tested using fastai's library for tabular data, and automated hyperparameter tuning was implemented using bayesian optimisation. The results of our experiments demonstrate the suitability of the proposed framework in building end-to-end systems for extensive monitoring and forecasting of pollutant concentration levels on highways. The study serves as a background for future related research looking to improve the overall performance of roadside and highway air quality forecasting models.     

Dynamic graph convolution neural network based on spatial-temporal correlation for air quality prediction

Air pollution is a serious threat to both the ecological environment and the physical health of individuals. Therefore, accurate air quality prediction is urgent and necessary for pollution mitigation and residents’ travel. However, few existing models are established based on the dynamic spatiotemporal correlation of air pollutants to predict air quality. In this paper, a novel deep learning model combining the dynamic graph convolutional network and the multi-channel temporal convolutional network (DGC-MTCN) is proposed for air quality prediction. To efficiently represent the time-varying spatial dependencies, a new spatiotemporal dynamic correlation calculation method based on gray relation analysis is proposed to construct dynamic adjacency matrices. Then, the spatiotemporal features are sufficiently extracted by the graph convolutional network and the multi-channel temporal convolutional network. Two real-world air quality datasets collected from Beijing and Fushun are applied to verify the performance of our proposed model. The experimental results show that compared with other baselines, the DGC-MTCN model has excellent prediction accuracy. Especially for the prediction of multi-step and different stations, our model performs better temporal stability and generalization ability.     

Prediction of beta-turns in proteins using neural networks

The use of neural networks to improve empirical secondary structure prediction is explored with regard to the identification of the position and conformational class of beta-turns, a four-residue chain reversal. Recently an algorithm was developed for beta-turn predictions based on the empirical approach of Chou and Fasman using different parameters for three classes (I, II and non-specific) of beta-turns. In this paper, using the same data, an alternative approach to derive an empirical prediction method is used based on neural networks which is a general learning algorithm extensively used in artificial intelligence. Thus the results of the two approaches can be compared. The most severe test of prediction accuracy is the percentage of turn predictions that are correct and the neural network gives an overall improvement from 20.6% to 26.0%. The proportion of correctly predicted residues is 71%, compared to a chance level of about 58%. Thus neural networks provide a method of obtaining more accurate predictions from empirical data than a simpler method of deriving propensities.     

Artificial neural networks as a useful tool to predict the risk level of Betula pollen in the air

An increasing percentage of the European population suffers from allergies to pollen. The study of the evolution of air pollen concentration supplies prior knowledge of the levels of pollen in the air, which can be useful for the prevention and treatment of allergic symptoms, and the management of medical resources. The symptoms of Betula pollinosis can be associated with certain levels of pollen in the air. The aim of this study was to predict the risk of the concentration of pollen exceeding a given level, using previous pollen and meteorological information, by applying neural network techniques. Neural networks are a widespread statistical tool useful for the study of problems associated with complex or poorly understood phenomena. The binary response variable associated with each level requires a careful selection of the neural network and the error function associated with the learning algorithm used during the training phase. The performance of the neural network with the validation set showed that the risk of the pollen level exceeding a certain threshold can be successfully forecasted using artificial neural networks. This prediction tool may be implemented to create an automatic system that forecasts the risk of suffering allergic symptoms.     

A community resource benchmarking predictions of peptide binding to MHC-I molecules

Recognition of peptides bound to major histocompatibility complex (MHC) class I molecules by T lymphocytes is an essential part of immune surveillance. Each MHC allele has a characteristic peptide binding preference, which can be captured in prediction algorithms, allowing for the rapid scan of entire pathogen proteomes for peptide likely to bind MHC. Here we make public a large set of 48,828 quantitative peptide-binding affinity measurements relating to 48 different mouse, human, macaque, and chimpanzee MHC class I alleles. We use this data to establish a set of benchmark predictions with one neural network method and two matrix-based prediction methods extensively utilized in our groups. In general, the neural network outperforms the matrix-based predictions mainly due to its ability to generalize even on a small amount of data. We also retrieved predictions from tools publicly available on the internet. While differences in the data used to generate these predictions hamper direct comparisons, we do conclude that tools based on combinatorial peptide libraries perform remarkably well. The transparent prediction evaluation on this dataset provides tool developers with a benchmark for comparison of newly developed prediction methods. In addition, to generate and evaluate our own prediction methods, we have established an easily extensible web-based prediction framework that allows automated side-by-side comparisons of prediction methods implemented by experts. This is an advance over the current practice of tool developers having to generate reference predictions themselves, which can lead to underestimating the performance of prediction methods they are not as familiar with as their own. The overall goal of this effort is to provide a transparent prediction evaluation allowing bioinformaticians to identify promising features of prediction methods and providing guidance to immunologists regarding the reliability of prediction tools.     

Time series predictions with neural nets: Application to airborne pollen forecasting

Pollen allergy is a common disease causing rhinoconjunctivitis (hay fever) in 5–10% of the population. Medical studies have indicated that pollen related diseases could be highly reduced if future pollen contents in the air could be predicted. In this work we have developed a new forecasting method that applies the ability of neural nets to predict the future behaviour of chaotic systems in order to make accurate predictions of the airborne pollen concentration. The method requires that the neural net be fed with non-zero values, which restricts the method predictions to the period following the start of pollen flight. The operational method outlined here constitutes a different point of view with respect to the more generally used forecasts of time series analysis, which require input of many meteorological parameters. Excellent forecasts were obtained training a neural net by using only the time series pollen concentration values.     

The value of harmful algal bloom predictions to the nearshore commercial shellfish fishery in the Gulf of Maine

In this study, we develop a framework for measuring the value of harmful algal bloom (HAB) predictions. The framework captures the effects of both private and public responses to HABs. Using data from the New England nearshore commercial shellfish fishery and impact estimates for a large-scale HAB event in 2005, we illustrate how the potential value of HAB forecasts may be estimated. The results of our study suggest that the long-term value of a HAB prediction and tracking system for the Gulf of Maine is sensitive to the frequency of HAB events, the accuracy of predictions, the choice of HAB impact measures, and the effectiveness of public and private responses.     

A HAB warning system for shellfish harvesting in Portugal

The development of sustainable shellfish aquaculture is highly dependent on the provision of reliable monitoring and predictive information on the occurrence of harmful algal blooms (HABs). The Portuguese HAB early warning system and shellfish closures presented here is a prototype, developed in the ASIMUTH project. It relies on weekly monitoring data composed of observations of HAB species and toxin concentrations within shellfish, and ocean circulation forecasts generated by an operational oceanographic model. The shellfish harvesting areas comprise coastal areas, estuaries + rías and coastal lagoons. The weekly bulletin characterizes the current shellfish closure situation and next week's forecasts for potentially impacted areas. The period analyzed ranged from 27 July 2013 to 17 March 2014, and describes the first skill assessment of the warning system. The forecast accuracy was evaluated, considering the number of forecasts that were verified to be correct the following week (85%) as well as the number of events not forecasted (false negatives, 12%) and those expected but did not occur (false positives, 3%). Variations were most visible in the first weeks of bulletin implementation and during autumn–winter months. The complementary use of field data, remote sensing and operational models led to more accurate predictions of blooms and range of the event.     

Prediction of β-Turns

Kohonen's self-organization model, a neural network model, is applied to predict the β-turns in proteins. There are 455 β-turn tetrapeptides and 3807 non-β-turn tetrapeptides in the training database. The rates of correct prediction for the 110 β-turn tetrapeptides and 30,229 non-β-turn tetrapeptides in the testing database are 81.8% and 90.7%, respectively. The high quality of prediction of neural network model implies that the residue-coupled effect along a polypeptide chain is important for the formation of reversal turns, such as β-turns, during the process of protein folding.     

Probabilistic seasonal dengue forecasting in Vietnam: A modelling study using superensembles

BackgroundWith enough advanced notice, dengue outbreaks can be mitigated. As a climate-sensitive disease, environmental conditions and past patterns of dengue can be used to make predictions about future outbreak risk. These predictions improve public health planning and decision-making to ultimately reduce the burden of disease. Past approaches to dengue forecasting have used seasonal climate forecasts, but the predictive ability of a system using different lead times in a year-round prediction system has been seldom explored. Moreover, the transition from theoretical to operational systems integrated with disease control activities is rare.Methods and findingsWe introduce an operational seasonal dengue forecasting system for Vietnam where Earth observations, seasonal climate forecasts, and lagged dengue cases are used to drive a superensemble of probabilistic dengue models to predict dengue risk up to 6 months ahead. Bayesian spatiotemporal models were fit to 19 years (2002–2020) of dengue data at the province level across Vietnam. A superensemble of these models then makes probabilistic predictions of dengue incidence at various future time points aligned with key Vietnamese decision and planning deadlines. We demonstrate that the superensemble generates more accurate predictions of dengue incidence than the individual models it incorporates across a suite of time horizons and transmission settings. Using historical data, the superensemble made slightly more accurate predictions (continuous rank probability score [CRPS] = 66.8, 95% CI 60.6–148.0) than a baseline model which forecasts the same incidence rate every month (CRPS = 79.4, 95% CI 78.5–80.5) at lead times of 1 to 3 months, albeit with larger uncertainty. The outbreak detection capability of the superensemble was considerably larger (69%) than that of the baseline model (54.5%). Predictions were most accurate in southern Vietnam, an area that experiences semi-regular seasonal dengue transmission. The system also demonstrated added value across multiple areas compared to previous practice of not using a forecast. We use the system to make a prospective prediction for dengue incidence in Vietnam for the period May to October 2020. Prospective predictions made with the superensemble were slightly more accurate (CRPS = 110, 95% CI 102–575) than those made with the baseline model (CRPS = 125, 95% CI 120–168) but had larger uncertainty. Finally, we propose a framework for the evaluation of probabilistic predictions. Despite the demonstrated value of our forecasting system, the approach is limited by the consistency of the dengue case data, as well as the lack of publicly available, continuous, and long-term data sets on mosquito control efforts and serotype-specific case data.ConclusionsThis study shows that by combining detailed Earth observation data, seasonal climate forecasts, and state-of-the-art models, dengue outbreaks can be predicted across a broad range of settings, with enough lead time to meaningfully inform dengue control. While our system omits some important variables not currently available at a subnational scale, the majority of past outbreaks could be predicted up to 3 months ahead. Over the next 2 years, the system will be prospectively evaluated and, if successful, potentially extended to other areas and other climate-sensitive disease systems.     

New methods for accurate prediction of protein secondary structure.

A primary and a secondary neural network are applied to secondary structure and structural class prediction for a database of 681 non-homologous protein chains. A new method of decoding the outputs of the secondary structure prediction network is used to produce an estimate of the probability of finding each type of secondary structure at every position in the sequence. In addition to providing a reliable estimate of the accuracy of the predictions, this method gives a more accurate Q3 (74.6%) than the cutoff method which is commonly used. Use of these predictions in jury methods improves the Q3 to 74.8%, the best available at present. On a database of 126 proteins commonly used for comparison of prediction methods, the jury predictions are 76.6% accurate. An estimate of the overall Q3 for a given sequence is made by averaging the estimated accuracy of the prediction over all residues in the sequence. As an example, the analysis is applied to the target beta-cryptogein, which was a difficult target for ab initio predictions in the CASP2 study; it shows that the prediction made with the present method (62% of residues correct) is close to the expected accuracy (66%) for this protein. The larger database and use of a new network training protocol also improve structural class prediction accuracy to 86%, relative to 80% obtained previously. Secondary structure content is predicted with accuracy comparable to that obtained with spectroscopic methods, such as vibrational or electronic circular dichroism and Fourier transform infrared spectroscopy.     

Prediction of the subcellular localization of eukaryotic proteins using sequence signals and composition

A tool called Locfind for the sequence-based prediction of the localization of eukaryotic proteins is introduced. It is based on bidirectional recurrent neural networks trained to read sequentially the amino acid sequence and produce localization information along the sequence. Systematic variation of the network architecture in combination with an efficient learning algorithm lead to a 91% correct localization prediction for novel proteins in fivefold cross-validation. The data and evaluation procedure are the same as the non-plant part of the widely used TargetP tool by Emanuelsson et al. The Locfind system is available on the WWW for predictions (http://www.stepc.gr/~synaptic/locfind.html).     

Advances in the prediction of protein targeting signals

Enlarged sets of reference data and special machine learning approaches have improved the accuracy of the prediction of protein subcellular localization. Recent approaches report over 95% correct predictions with low fractions of false-positives for secretory proteins. A clear trend is to develop specifically tailored organism- and organelle-specific prediction tools rather than using one general method. Focus of the review is on machine learning systems, highlighting four concepts: the artificial neural feed-forward network, the self-organizing map (SOM), the Hidden-Markov-Model (HMM), and the support vector machine (SVM).     

Protein structural motif prediction in multidimensional phi-psi space leads to improved secondary structure prediction.

A significant step towards establishing the structure and function of a protein is the prediction of the local conformation of the polypeptide chain. In this article, we present systems for the prediction of three new alphabets of local structural motifs. The motifs are built by applying multidimensional scaling (MDS) and clustering to pair-wise angular distances for multiple phi-psi angle values collected from high-resolution protein structures. The predictive systems, based on ensembles of bidirectional recurrent neural network architectures, and trained on a large non-redundant set of protein structures, achieve 72%, 66%, and 60% correct motif prediction on an independent test set for di-peptides (six classes), tri-peptides (eight classes) and tetra-peptides (14 classes), respectively, 28-30% above baseline statistical predictors. We then build a further system, based on ensembles of two-layered bidirectional recurrent neural networks, to map structural motif predictions into a traditional 3-class (helix, strand, coil) secondary structure. This system achieves 79.5% correct prediction using the "hard" CASP 3-class assignment, and 81.4% with a more lenient assignment, outperforming a sophisticated state-of-the-art predictor (Porter) trained in the same experimental conditions. The structural motif predictor is publicly available at: http://distill.ucd.ie/porter+/.     

Neural network modeling in optimisation of continuous fermentation processes

The capability of self-recurrent neural networks in dynamic modeling of continuous fermentation is investigated in this simulation study. In the past, feedforward neural networks have been successfully used as one-step-ahead predictors. However, in steady-state optimisation of continuous fermentations the neural network model has to be iterated to predict many time steps ahead into the future in order to get steady-state values of the variables involved in objective cost function, and this iteration may result in increasing errors. Therefore, as an alternative to classical feedforward neural network trained by using backpropagation method, self-recurrent multilayer neural net trained by backpropagation through time method was chosen in order to improve accuracy of long-term predictions. Prediction capabilities of the resulting neural network model is tested by implementing this into the Integrated System Optimisation and Parameter Estimation (ISOPE) optimisation algorithm. Maximisation of cellular productivity of the baker's yeast continuous fermentation was used as the goal of the proposed optimising control problem. The training and prediction results of proposed neural network and performances of resulting optimisation structure are demonstrated.     

Predictive modelling and spatial mapping of freshwater fish and decapod assemblages using GIS and neural networks

1. We used stream fish and decapod spatial occurrence data extracted from a national database and recent surveys with geospatial landuse data, geomorphologic, climatic, and spatial data in a geographical information system (GIS) to model fish and decapod occurrence in the Wellington Region, New Zealand. 2. To predict the occurrence of each species at a site from a common set of predictor variables we used a multi‐response, artificial neural network (ANN), to produce a single model that predicted the entire fish and decapod assemblage in one procedure. 3. The predictions from the ANN using this landscape scale data proved very accurate based on evaluation metrics that are independent of species abundance or probability thresholds. The important variables contributing to the predictions included the latitudinal and elevational position of the site reach, catchment area, average air temperature, the vegetation type, landuse proportions of the catchment, and catchment geology. 4. Geospatial data available for the entire regional river network were then used to create a habitat‐suitability map for all 14 species over the regional river network using a GIS. This prediction map has many potential uses including: monitoring and predicting temporal changes in fish communities caused by human activities and shifts in climate, identifying areas in need of protection, biodiversity hotspots, and areas suitable for the reintroduction of endangered or rare species.     

Improved performance in protein secondary structure prediction by combining multiple predictions

In this paper(1) we present a novel framework for protein secondary structure prediction. In this prediction framework, firstly we propose a novel parameterized semi-probability profile, which combines single sequence with evolutionary information effectively. Secondly, different semi-probability profiles are respectively applied as network input to predict protein secondary structure. Then a comparison among these different predictions is discussed in this article. Finally, na?ve Bayes approaches are used to combine these predictions in order to obtain a better prediction performance than individual prediction. The experimental results show that our proposed framework can indeed improve the prediction accuracy.     

A convolutional neural network to recognize Chagas disease vectors using mobile phone images

There are several identification tools that can assist researchers, technicians and the community in the recognition of Chagas vector insects (triatomines), from other insects with similar morphologies. They involve using dichotomous keys, field guides, expert knowledge or, in more recent approaches, through the classification by a neural network of high quality photographs taken in standardized conditions. The aim of this research was to develop a deep neural network to recognize triatomines (insects associated with vectorial transmission of Chagas disease) directly from photos taken with any commonly available mobile device, without any other specialized equipment. To overcome the shortcomings of taking images using specific instruments and a controlled environment an innovative machine-learning approach was used: Fastai with Pytorch, a combination of open-source software for deep learning. The Convolutional Neural Network (CNN) was trained with triatomine photos, reaching a correct identification in 94.3% of the cases. Results were validated using photos sent by citizen scientists from the GeoVin project, resulting in 91.4% of correct identification of triatomines. The CNN provides a lightweight, robust method that even works with blurred images, poor lighting and even with the presence of other subjects and objects in the same frame. Future steps include the inclusion of the CNN into the framework of the GeoVin science project, which will also allow to further train the network using the photos sent by the citizen scientists. This would allow the participation of the community in the identification and monitoring of the vector insects, particularly in regions where government-led monitoring programmes are not frequent due to their low accessibility and high costs.