Modeling vegetation greenness and its climate sensitivity with deep‐learning technology

Climate sensitivity of vegetation has long been explored using statistical or process‐based models. However, great uncertainties still remain due to the methodologies’ deficiency in capturing the complex interactions between climate and vegetation. Here, we developed global gridded climate–vegetation models based on long short‐term memory (LSTM) network, which is a powerful deep‐learning algorithm for long‐time series modeling, to achieve accurate vegetation monitoring and investigate the complex relationship between climate and vegetation. We selected the normalized difference vegetation index (NDVI) that represents vegetation greenness as model outputs. The climate data (monthly temperature and precipitation) were used as inputs. We trained the networks with data from 1982 to 2003, and the data from 2004 to 2015 were used to validate the models. Error analysis and sensitivity analysis were performed to assess the model errors and investigate the sensitivity of global vegetation to climate change. Results show that models based on deep learning are very effective in simulating and predicting the vegetation greenness dynamics. For models training, the root mean square error (RMSE) is <0.01. Model validation also assure the accuracy of our models. Furthermore, sensitivity analysis of models revealed a spatial pattern of global vegetation to climate, which provides us a new way to investigate the climate sensitivity of vegetation. Our study suggests that it is a good way to integrate deep‐learning method to monitor the vegetation change under global change. In the future, we can explore more complex climatic and ecological systems with deep learning and coupling with certain physical process to better understand the nature.     

Automated location invariant animal detection in camera trap images using publicly available data sources

1. A time‐consuming challenge faced by camera trap practitioners is the extraction of meaningful data from images to inform ecological management. An increasingly popular solution is automated image classification software. However, most solutions are not sufficiently robust to be deployed on a large scale due to lack of location invariance when transferring models between sites. This prevents optimal use of ecological data resulting in significant expenditure of time and resources to annotate and retrain deep learning models.
2. We present a method ecologists can use to develop optimized location invariant camera trap object detectors by (a) evaluating publicly available image datasets characterized by high intradataset variability in training deep learning models for camera trap object detection and (b) using small subsets of camera trap images to optimize models for high accuracy domain‐specific applications.
3. We collected and annotated three datasets of images of striped hyena, rhinoceros, and pigs, from the image‐sharing websites FlickR and iNaturalist (FiN), to train three object detection models. We compared the performance of these models to that of three models trained on the Wildlife Conservation Society and Camera CATalogue datasets, when tested on out‐of‐sample Snapshot Serengeti datasets. We then increased FiN model robustness by infusing small subsets of camera trap images into training.
4. In all experiments, the mean Average Precision (mAP) of the FiN trained models was significantly higher (82.33%–88.59%) than that achieved by the models trained only on camera trap datasets (38.5%–66.74%). Infusion further improved mAP by 1.78%–32.08%.
5. Ecologists can use FiN images for training deep learning object detection solutions for camera trap image processing to develop location invariant, robust, out‐of‐the‐box software. Models can be further optimized by infusion of 5%–10% camera trap images into training data. This would allow AI technologies to be deployed on a large scale in ecological applications. Datasets and code related to this study are open source and available on this repository: https://doi.org/10.5061/dryad.1c59zw3tx.

     

Estimating leaf area index using unmanned aerial vehicle data: shallow vs. deep machine learning algorithms

Measuring leaf area index (LAI) is essential for evaluating crop growth and estimating yield, thereby facilitating high-throughput phenotyping of maize (Zea mays). LAI estimation models use multi-source data from unmanned aerial vehicles (UAVs), but using multimodal data to estimate maize LAI, and the effect of tassels and soil background, remain understudied. Our research aims to (1) determine how multimodal data contribute to LAI and propose a framework for estimating LAI based on remote-sensing data, (2) evaluate the robustness and adaptability of an LAI estimation model that uses multimodal data fusion and deep neural networks (DNNs) in single- and whole growth stages, and (3) explore how soil background and maize tasseling affect LAI estimation. To construct multimodal datasets, our UAV collected red–green–blue, multispectral, and thermal infrared images. We then developed partial least square regression (PLSR), support vector regression, and random forest regression models to estimate LAI. We also developed a deep learning model with three hidden layers. This multimodal data structure accurately estimated maize LAI. The DNN model provided the best estimate (coefficient of determination [R²] = 0.89, relative root mean square error [rRMSE] = 12.92%) for a single growth period, and the PLSR model provided the best estimate (R² = 0.70, rRMSE = 12.78%) for a whole growth period. Tassels reduced the accuracy of LAI estimation, but the soil background provided additional image feature information, improving accuracy. These results indicate that multimodal data fusion using low-cost UAVs and DNNs can accurately and reliably estimate LAI for crops, which is valuable for high-throughput phenotyping and high-spatial precision farmland management.

Multimodal data fusion (red–green–blue, multispectral, and thermal infrared) using low-cost unmanned aerial vehicles in a deep neural network and machine learning framework estimates maize leaf area index     

Prediction of Hematopoietic Stem Cell Transplantation Related Mortality- Lessons Learned from the In-Silico Approach: A European Society for Blood and Marrow Transplantation Acute Leukemia Working Party Data Mining Study

Models for prediction of allogeneic hematopoietic stem transplantation (HSCT) related mortality partially account for transplant risk. Improving predictive accuracy requires understating of prediction limiting factors, such as the statistical methodology used, number and quality of features collected, or simply the population size. Using an in-silico approach (i.e., iterative computerized simulations), based on machine learning (ML) algorithms, we set out to analyze these factors. A cohort of 25,923 adult acute leukemia patients from the European Society for Blood and Marrow Transplantation (EBMT) registry was analyzed. Predictive objective was non-relapse mortality (NRM) 100 days following HSCT. Thousands of prediction models were developed under varying conditions: increasing sample size, specific subpopulations and an increasing number of variables, which were selected and ranked by separate feature selection algorithms. Depending on the algorithm, predictive performance plateaued on a population size of 6,611–8,814 patients, reaching a maximal area under the receiver operator characteristic curve (AUC) of 0.67. AUCs’ of models developed on specific subpopulation ranged from 0.59 to 0.67 for patients in second complete remission and receiving reduced intensity conditioning, respectively. Only 3–5 variables were necessary to achieve near maximal AUCs. The top 3 ranking variables, shared by all algorithms were disease stage, donor type, and conditioning regimen. Our findings empirically demonstrate that with regards to NRM prediction, few variables “carry the weight” and that traditional HSCT data has been “worn out”. “Breaking through” the predictive boundaries will likely require additional types of inputs.     

Learning deep representations of enzyme thermal adaptation

Temperature is a fundamental environmental factor that shapes the evolution of organisms. Learning thermal determinants of protein sequences in evolution thus has profound significance for basic biology, drug discovery, and protein engineering. Here, we use a data set of over 3 million BRENDA enzymes labeled with optimal growth temperatures (OGTs) of their source organisms to train a deep neural network model (DeepET). The protein‐temperature representations learned by DeepET provide a temperature‐related statistical summary of protein sequences and capture structural properties that affect thermal stability. For prediction of enzyme optimal catalytic temperatures and protein melting temperatures via a transfer learning approach, our DeepET model outperforms classical regression models trained on rationally designed features and other deep‐learning‐based representations. DeepET thus holds promise for understanding enzyme thermal adaptation and guiding the engineering of thermostable enzymes.     

Deep transformation models for functional outcome prediction after acute ischemic stroke

In many medical applications, interpretable models with high prediction performance are sought. Often, those models are required to handle semistructured data like tabular and image data. We show how to apply deep transformation models (DTMs) for distributional regression that fulfill these requirements. DTMs allow the data analyst to specify (deep) neural networks for different input modalities making them applicable to various research questions. Like statistical models, DTMs can provide interpretable effect estimates while achieving the state-of-the-art prediction performance of deep neural networks. In addition, the construction of ensembles of DTMs that retain model structure and interpretability allows quantifying epistemic and aleatoric uncertainty. In this study, we compare several DTMs, including baseline-adjusted models, trained on a semistructured data set of 407 stroke patients with the aim to predict ordinal functional outcome three months after stroke. We follow statistical principles of model-building to achieve an adequate trade-off between interpretability and flexibility while assessing the relative importance of the involved data modalities. We evaluate the models for an ordinal and dichotomized version of the outcome as used in clinical practice. We show that both tabular clinical and brain imaging data are useful for functional outcome prediction, whereas models based on tabular data only outperform those based on imaging data only. There is no substantial evidence for improved prediction when combining both data modalities. Overall, we highlight that DTMs provide a powerful, interpretable approach to analyzing semistructured data and that they have the potential to support clinical decision-making.     

A comparison of genomic selection models across time in interior spruce (Picea engelmannii × glauca) using unordered SNP imputation methods

Genomic selection (GS) potentially offers an unparalleled advantage over traditional pedigree-based selection (TS) methods by reducing the time commitment required to carry out a single cycle of tree improvement. This quality is particularly appealing to tree breeders, where lengthy improvement cycles are the norm. We explored the prospect of implementing GS for interior spruce (Picea engelmannii × glauca) utilizing a genotyped population of 769 trees belonging to 25 open-pollinated families. A series of repeated tree height measurements through ages 3–40 years permitted the testing of GS methods temporally. The genotyping-by-sequencing (GBS) platform was used for single nucleotide polymorphism (SNP) discovery in conjunction with three unordered imputation methods applied to a data set with 60% missing information. Further, three diverse GS models were evaluated based on predictive accuracy (PA), and their marker effects. Moderate levels of PA (0.31–0.55) were observed and were of sufficient capacity to deliver improved selection response over TS. Additionally, PA varied substantially through time accordingly with spatial competition among trees. As expected, temporal PA was well correlated with age-age genetic correlation (r=0.99), and decreased substantially with increasing difference in age between the training and validation populations (0.04–0.47). Moreover, our imputation comparisons indicate that k-nearest neighbor and singular value decomposition yielded a greater number of SNPs and gave higher predictive accuracies than imputing with the mean. Furthermore, the ridge regression (rrBLUP) and BayesCπ (BCπ) models both yielded equal, and better PA than the generalized ridge regression heteroscedastic effect model for the traits evaluated.     

A Flexible and Accurate Genotype Imputation Method for the Next Generation of Genome-Wide Association Studies

Genotype imputation methods are now being widely used in the analysis of genome-wide association studies. Most imputation analyses to date have used the HapMap as a reference dataset, but new reference panels (such as controls genotyped on multiple SNP chips and densely typed samples from the 1,000 Genomes Project) will soon allow a broader range of SNPs to be imputed with higher accuracy, thereby increasing power. We describe a genotype imputation method (IMPUTE version 2) that is designed to address the challenges presented by these new datasets. The main innovation of our approach is a flexible modelling framework that increases accuracy and combines information across multiple reference panels while remaining computationally feasible. We find that IMPUTE v2 attains higher accuracy than other methods when the HapMap provides the sole reference panel, but that the size of the panel constrains the improvements that can be made. We also find that imputation accuracy can be greatly enhanced by expanding the reference panel to contain thousands of chromosomes and that IMPUTE v2 outperforms other methods in this setting at both rare and common SNPs, with overall error rates that are 15%–20% lower than those of the closest competing method. One particularly challenging aspect of next-generation association studies is to integrate information across multiple reference panels genotyped on different sets of SNPs; we show that our approach to this problem has practical advantages over other suggested solutions.     

Missing Data in Randomized Clinical Trials for Weight Loss: Scope of the Problem,State of the Field,and Performance of Statistical Methods

Background

Dropouts and missing data are nearly-ubiquitous in obesity randomized controlled trails, threatening validity and generalizability of conclusions. Herein, we meta-analytically evaluate the extent of missing data, the frequency with which various analytic methods are employed to accommodate dropouts, and the performance of multiple statistical methods.

Methodology/Principal Findings

We searched PubMed and Cochrane databases (2000–2006) for articles published in English and manually searched bibliographic references. Articles of pharmaceutical randomized controlled trials with weight loss or weight gain prevention as major endpoints were included. Two authors independently reviewed each publication for inclusion. 121 articles met the inclusion criteria. Two authors independently extracted treatment, sample size, drop-out rates, study duration, and statistical method used to handle missing data from all articles and resolved disagreements by consensus. In the meta-analysis, drop-out rates were substantial with the survival (non-dropout) rates being approximated by an exponential decay curve (e^−λt) where λ was estimated to be .0088 (95% bootstrap confidence interval: .0076 to .0100) and t represents time in weeks. The estimated drop-out rate at 1 year was 37%. Most studies used last observation carried forward as the primary analytic method to handle missing data. We also obtained 12 raw obesity randomized controlled trial datasets for empirical analyses. Analyses of raw randomized controlled trial data suggested that both mixed models and multiple imputation performed well, but that multiple imputation may be more robust when missing data are extensive.

Conclusion/Significance

Our analysis offers an equation for predictions of dropout rates useful for future study planning. Our raw data analyses suggests that multiple imputation is better than other methods for handling missing data in obesity randomized controlled trials, followed closely by mixed models. We suggest these methods supplant last observation carried forward as the primary method of analysis.     

Fusing Data Mining,Machine Learning and Traditional Statistics to Detect Biomarkers Associated with Depression

Background

Atheoretical large-scale data mining techniques using machine learning algorithms have promise in the analysis of large epidemiological datasets. This study illustrates the use of a hybrid methodology for variable selection that took account of missing data and complex survey design to identify key biomarkers associated with depression from a large epidemiological study.

Methods

The study used a three-step methodology amalgamating multiple imputation, a machine learning boosted regression algorithm and logistic regression, to identify key biomarkers associated with depression in the National Health and Nutrition Examination Study (2009–2010). Depression was measured using the Patient Health Questionnaire-9 and 67 biomarkers were analysed. Covariates in this study included gender, age, race, smoking, food security, Poverty Income Ratio, Body Mass Index, physical activity, alcohol use, medical conditions and medications. The final imputed weighted multiple logistic regression model included possible confounders and moderators.

Results

After the creation of 20 imputation data sets from multiple chained regression sequences, machine learning boosted regression initially identified 21 biomarkers associated with depression. Using traditional logistic regression methods, including controlling for possible confounders and moderators, a final set of three biomarkers were selected. The final three biomarkers from the novel hybrid variable selection methodology were red cell distribution width (OR 1.15; 95% CI 1.01, 1.30), serum glucose (OR 1.01; 95% CI 1.00, 1.01) and total bilirubin (OR 0.12; 95% CI 0.05, 0.28). Significant interactions were found between total bilirubin with Mexican American/Hispanic group (p = 0.016), and current smokers (p<0.001).

Conclusion

The systematic use of a hybrid methodology for variable selection, fusing data mining techniques using a machine learning algorithm with traditional statistical modelling, accounted for missing data and complex survey sampling methodology and was demonstrated to be a useful tool for detecting three biomarkers associated with depression for future hypothesis generation: red cell distribution width, serum glucose and total bilirubin.     

Artificial intelligent system for multimedia services in smart home environments

Internet of Things (IoT) has introduced new applications and environments. Smart Home provides new ways of communication and service consumption. In addition, Artificial Intelligence (AI) and deep learning have improved different services and tasks by automatizing them. In this field, reinforcement learning (RL) provides an unsupervised way to learn from the environment. In this paper, a new intelligent system based on RL and deep learning is proposed for Smart Home environments to guarantee good levels of QoE, focused on multimedia services. This system is aimed to reduce the impact on user experience when the classifying system achieves a low accuracy. The experiments performed show that the deep learning model proposed achieves better accuracy than the KNN algorithm and that the RL system increases the QoE of the user up to 3.8 on a scale of 10.

     

Benchmarking AlphaFold for protein complex modeling reveals accuracy determinants

High‐resolution experimental structural determination of protein–protein interactions has led to valuable mechanistic insights, yet due to the massive number of interactions and experimental limitations there is a need for computational methods that can accurately model their structures. Here we explore the use of the recently developed deep learning method, AlphaFold, to predict structures of protein complexes from sequence. With a benchmark of 152 diverse heterodimeric protein complexes, multiple implementations and parameters of AlphaFold were tested for accuracy. Remarkably, many cases (43%) had near‐native models (medium or high critical assessment of predicted interactions accuracy) generated as top‐ranked predictions by AlphaFold, greatly surpassing the performance of unbound protein–protein docking (9% success rate for near‐native top‐ranked models), however AlphaFold modeling of antibody–antigen complexes within our set was unsuccessful. We identified sequence and structural features associated with lack of AlphaFold success, and we also investigated the impact of multiple sequence alignment input. Benchmarking of a multimer‐optimized version of AlphaFold (AlphaFold‐Multimer) with a set of recently released antibody–antigen structures confirmed a low rate of success for antibody–antigen complexes (11% success), and we found that T cell receptor–antigen complexes are likewise not accurately modeled by that algorithm, showing that adaptive immune recognition poses a challenge for the current AlphaFold algorithm and model. Overall, our study demonstrates that end‐to‐end deep learning can accurately model many transient protein complexes, and highlights areas of improvement for future developments to reliably model any protein–protein interaction of interest.     

Imputing missing data in plant traits: A guide to improve gap-filling

Aim

Globally distributed plant trait data are increasingly used to understand relationships between biodiversity and ecosystem processes. However, global trait databases are sparse because they are compiled from many, mostly small databases. This sparsity in both trait space completeness and geographical distribution limits the potential for both multivariate and global analyses. Thus, ‘gap-filling’ approaches are often used to impute missing trait data. Recent methods, like Bayesian hierarchical probabilistic matrix factorization (BHPMF), can impute large and sparse data sets using side information. We investigate whether BHPMF imputation leads to biases in trait space and identify aspects influencing bias to provide guidance for its usage.

Innovation

We use a fully observed trait data set from which entries are randomly removed, along with extensive but sparse additional data. We use BHPMF for imputation and evaluate bias by: (1) accuracy (residuals, RMSE, trait means), (2) correlations (bi- and multivariate) and (3) taxonomic and functional clustering (valuewise, uni- and multivariate). BHPMF preserves general patterns of trait distributions but induces taxonomic clustering. Data set–external trait data had little effect on induced taxonomic clustering and stabilized trait–trait correlations.

Main Conclusions

Our study extends the criteria for the evaluation of gap-filling beyond RMSE, providing insight into statistical data structure and allowing better informed use of imputed trait data, with improved practice for imputation. We expect our findings to be valuable beyond applications in plant ecology, for any study using hierarchical side information for imputation.     

HaploBlocks: Efficient Detection of Positive Selection in Large Population Genomic Datasets

Genomic regions under positive selection harbor variation linked for example to adaptation. Most tools for detecting positively selected variants have computational resource requirements rendering them impractical on population genomic datasets with hundreds of thousands of individuals or more. We have developed and implemented an efficient haplotype-based approach able to scan large datasets and accurately detect positive selection. We achieve this by combining a pattern matching approach based on the positional Burrows–Wheeler transform with model-based inference which only requires the evaluation of closed-form expressions. We evaluate our approach with simulations, and find it to be both sensitive and specific. The computational resource requirements quantified using UK Biobank data indicate that our implementation is scalable to population genomic datasets with millions of individuals. Our approach may serve as an algorithmic blueprint for the era of “big data” genomics: a combinatorial core coupled with statistical inference in closed form.     

Evaluation of missing data imputation methods for human osteometric measurements

It is not uncommon for biological anthropologists to analyze incomplete bioarcheological or forensic skeleton specimens. As many quantitative multivariate analyses cannot handle incomplete data, missing data imputation or estimation is a common preprocessing practice for such data. Using William W. Howells' Craniometric Data Set and the Goldman Osteometric Data Set, we evaluated the performance of multiple popular statistical methods for imputing missing metric measurements. Results indicated that multiple imputation methods outperformed single imputation methods, such as Bayesian principal component analysis (BPCA). Multiple imputation with Bayesian linear regression implemented in the R package norm2, the Expectation–Maximization (EM) with Bootstrapping algorithm implemented in Amelia, and the Predictive Mean Matching (PMM) method and several of the derivative linear regression models implemented in mice, perform well regarding accuracy, robustness, and speed. Based on the findings of this study, we suggest a practical procedure for choosing appropriate imputation methods.     

Macroinvertebrate-based multimetric predictive models for evaluating the human impact on biotic condition of Bolivian streams

We developed and validated a single multimetric index based on predictive models that could evaluate anthropogenic disturbances in streams of three disparate ecoregions of Bolivia. To do so, we examined 45 candidate metrics reflecting different aspects of macroinvertebrate assemblage structure and function gleaned from available literature and for their potential to indicate degradation. More importantly, we integrated functional trait metrics to improve the sensitivity of our index. To quantify possible deviation from reference conditions, we first established and validated statistical models describing metric responses to natural environmental differences in the absence of any significant anthropogenic disturbance. We considered that the residual distributions of these models described the response range of each metric, independently of natural environmental influence. After testing the sensitivity of these residuals to a gradient of anthropogenic disturbance, we retained eight metrics that were used in the final assemblage index, four metrics based on richness and composition and four metrics based on biological traits. Our index performed well in discriminating between reference and disturbed sites, giving a significant negative linear response to a gradient of physical and chemical anthropogenic disturbances. After employing a probability survey design and sampling a relatively small number of sites throughout all major ecoregions of Bolivia, we believe our methodology can be used to develop a monitoring tool to evaluate status and trends in biological condition for streams of the entire country despite its complex and heterogeneous geology and climate.