A common procedure for studying the microbiome is binning the sequenced contigs into metagenome-assembled genomes. State-of-the-art binning methods use coabundance and sequence-based motifs such as tetranucleotide frequencies, whereas taxonomic labels derived from alignment based classification have not been widely used. Here we propose TaxVAMB, a metagenome binning tool based on semisupervised bimodal variational autoencoders, combining tetranucleotide frequencies and contig coabundances with taxonomic information. TaxVAMB outperformed all other binners on CAMI2 human microbiome datasets, returning on average 29% more high-quality assemblies than the next best binner, and performed on par with the best binners on short-read datasets. On a human gut long-read dataset, TaxVAMB recovered 29% more high-quality bins. In a typical single-sample setup, TaxVAMB on average returns 83% more high-quality bins compared to VAMB. Lastly, TaxVAMB binned incomplete genomes better than any other tool, returning on average 300% more high-quality bins of incomplete genomes than the next best binner.

Spectral imaging data acquired via multispectral and hyperspectral cameras can have hundreds of channels, where each channel records the reflectance at a specific wavelength and bandwidth. Time and resource constraints limit our ability to collect large spectral datasets, making it difficult to build and train predictive models from scratch. In the RGB domain, we can often alleviate some of the limitations of smaller datasets by using pretrained foundational models as a starting point. However, most existing foundation models are pretrained on large datasets of 3-channel RGB images, severely limiting their effectiveness when used with spectral imaging data. The few spectral foundation models that do exist usually have one of two limitations: (1) they are built and trained only on remote sensing data limiting their application in proximal spectral imaging, (2) they utilize the more widely available multispectral imaging datasets with less than 15 channels restricting their use with hundred-channel hyperspectral images. To alleviate these issues, we propose a large-scale foundational model and dataset built upon the masked autoencoder architecture that takes advantage of spectral channel encoding, spatial-spectral masking and ImageNet pretraining for an adaptable and robust model for downstream spectral imaging tasks.

A common procedure for studying the microbiome is binning the sequenced contigs into metagenome-assembled genomes. Currently, unsupervised and self-supervised deep learning based methods using co-abundance and sequence based motifs such as tetranucleotide frequencies are state-of-the-art for metagenome binning. Taxonomic labels derived from alignment based classification have not been widely used. Here, we propose TaxVAMB, a metagenome binning tool based on semi-supervised bi-modal variational autoencoders, combining tetranucleotide frequencies and contig co-abundances with contig annotations returned by any taxonomic classifier on any taxonomic rank. TaxVAMB outperforms all other binners on CAMI2 human microbiome datasets, returning on average 40% more near-complete assemblies than the next best binner. On real long-read datasets TaxVAMB recovers on average 13% more near-complete bins and 14% more species. When used in a single-sample setup, TaxVAMB on average returns 83% more high quality bins than VAMB. TaxVAMB bins incomplete genomes drastically better than any other tool, returning 255% more high quality bins of incomplete genomes than the next best binner. Our method has immediate research and industrial applications, as well as methodological novelty which can be translated to other biological problems with semi-supervised multimodal datasets. ### Competing Interest Statement J.N.N. is the author of the VAMB binning tool, which has been developed using a prototype of BinBencher, which is used to calculate some of the benchmarking metrics in this paper. S.R. is the founder and owner of the Danish company BioAI and have performed consulting for Sidera Bio ApS. Other authors declare no competing interests.

Hyperspectral imaging is a widely used method in remote sensing, particularly for use in airborne and satellite-based land surveillance. Its versatility is, however, much larger and has also seen usage in everything ranging from food processing and surveillance to astronomy and waste sorting. It is also gaining inroads with agricultural research. With most available datasets focusing on per-pixel classification, there is, however, a potential for hyperspectral whole-image analysis, but there is a severe lack of datasets for whole-image analysis. To help fill this gap and facilitate methodological development in whole-image hyperspectral image analysis, we introduce the Hy-perLeaf2024 dataset. The dataset consists of 2410 hyper-spectral images of wheat leaves, along with associated classification and regression targets at both the leaf level and the plot level. In addition to the dataset, we also provide experiments showing the importance of pretraining and highlighting the future research direction in whole-image hyper-spectral image analysis.

Hyperspectral imaging is central for remote sensing, and much research has been done on analysis methods for land surveillance using space- and air-borne imaging systems. Proximal hyperspectral imaging is also widely used in plant and agriculture science. It allows the remote capturing of leaf reflectance information in order to determine and classify plant health and disease. With the high information density in hyperspectral images, it becomes increasingly important to apply sophisticated feature extraction in order to reduce image dimensionality while retaining useful information. Autoencoders are one of the primary methods for deep learning-based feature extraction in hyperspectral images. We investigate different setups of autoencoders to encode the spatial and spectral dimensions in different orders and ways. To our surprise, the best turns out to be a 3D CNN, where the spectral dimension is treated in the same way as the spatial dimensions.

Effective detection of potato late blight (PLB) is an essential aspect of potato cultivation. However, it is a challenge to detect late blight in asymptomatic biotrophic phase in fields with conventional imaging approaches because of the lack of visual symptoms in the canopy. Hyperspectral imaging can capture spectral signals from a wide range of wavelengths also outside the visual wavelengths. Here, we propose a deep learning classification architecture for hyperspectral images by combining 2D convolutional neural network (2D-CNN) and 3D-CNN with deep cooperative attention networks (PLB-2D-3D-A). First, 2D-CNN and 3D-CNN are used to extract rich spectral space features, and then the attention mechanism AttentionBlock and SE-ResNet are used to emphasize the salient features in the feature maps and increase the generalization ability of the model. The dataset is built with 15,360 images (64x64x204), cropped from 240 raw images captured in an experimental field with over 20 potato genotypes. The accuracy in the test dataset of 2000 images reached 0.739 in the full band and 0.790 in the specific bands (492 nm, 519 nm, 560 nm, 592 nm, 717 nm and 765 nm). This study shows an encouraging result for classification of the asymptomatic biotrophic phase of PLB disease with deep learning and proximal hyperspectral imaging.

Leaf rust caused by Puccinia triticina Erikss. can have devastating effects on wheat (Triticum aestivum L.), causing severe economic losses. This comprehensive study serves to facilitate our understanding of the impact of carbohydrate and antioxidant metabolism in association with sensor-based phenotyping and leaf rust stress responses in wheat seedlings. After 24 h of inoculation (hai) very susceptible variety to leaf rust (Ficko) increased cell-wall invertase (cwInv; EC 3.2.1.26), compared to other varieties that significantly increased cwInv later. This could mean that the Ficko variety cannot defend itself from leaf rust infections once symptoms have started to develop. Also, Ficko had significantly decreased amounts of cytoplasmic invertase (cytInv; EC 3.2.1.26) at 8 hai. The downregulation of cytInv in susceptible plants may facilitate the maintenance of elevated apoplastic sucrose availability favoring the pathogen. The significant role of vacuolar invertase (vacInv; EC 3.2.1.26) in moderately resistant varieties was recorded. Also, a significant decrease of glucose-6-phosphate dehydrogenase (G6PDH; EC 1.1.1.49) and UDP-glucose pyrophosphorylase (UGPase; EC 2.7.7.9) in moderately resistant varieties might restrict normal development of leaf rust due to reduced sugar. During plant-pathogen interaction, when the invader spreads systemically throughout the plant, the main role of ascorbate peroxidase (APX; EC 1.11.1.11) activity in one moderately resistant variety (Olimpija) and catalase (CAT; EC 1.11.1.6) activity in another moderately resistant variety (Alka) is to protect the plant against oxidative damage in the early stages of infection. Non-invasive phenotyping with a sensor-based technique could be used as a rapid method for pre-symptomatic determination of wheat leaf rust resistance or susceptibility.

Interannual and local fluctuations in wheat crop yield are mostly explained by abiotic constraints. Heatwaves and drought, which are among the top stressors, commonly co-occur, and their frequency is increasing with global climate change. High-throughput methods were optimized to phenotype wheat plants under controlled water deficit and high temperature, with the aim to identify phenotypic traits conferring adaptative stress responses. Wheat plants of 10 genotypes were grown in a fully automated plant facility under 25/18 °C day/night for 30 d, and then the temperature was increased for 7 d (38/31 °C day/night) while maintaining half of the plants well irrigated and half at 30% field capacity. Thermal and multispectral images and pot weights were registered twice daily. At the end of the experiment, key metabolites and enzyme activities from carbohydrate and antioxidant metabolism were quantified. Regression machine learning models were successfully established to predict plant biomass using image-extracted parameters. Evapotranspiration traits expressed significant genotype-environment interactions (G×E) when acclimatization to stress was continuously monitored. Consequently, transpiration efficiency was essential to maintain the balance between water-saving strategies and biomass production in wheat under water deficit and high temperature. Stress tolerance included changes in carbohydrate metabolism, particularly in the sucrolytic and glycolytic pathways, and in antioxidant metabolism. The observed genetic differences in sensitivity to high temperature and water deficit can be exploited in breeding programmes to improve wheat resilience to climate change.

Necrotic and chlorotic symptoms induced during Pyrenophora teres infection in barley leaves indicate a compatible interaction that allows the hemi-biotrophic fungus Pyrenophora teres to colonise the host. However, it is unexplored how this fungus affects the physiological responses of resistant and susceptible cultivars during infection. To assess the degree of resistance in four different cultivars, we quantified visible symptoms and fungal DNA and performed expression analyses of genes involved in plant defence and ROS scavenging. To obtain insight into the interaction between fungus and host, we determined the activity of 19 key enzymes of carbohydrate and antioxidant metabolism. The pathogen impact was also phenotyped non-invasively by sensor-based multireflectance and -fluorescence imaging. Symptoms, regulation of stress-related genes and pathogen DNA content distinguished the cultivar Guld as being resistant. Severity of net blotch symptoms was also strongly correlated with the dynamics of enzyme activities already within the first day of infection. In contrast to the resistant cultivar, the three susceptible cultivars showed a higher reflectance over seven spectral bands and higher fluorescence intensities at specific excitation wavelengths. The combination of semi high-throughput physiological and molecular analyses with non-invasive phenotyping enabled the identification of bio-signatures that discriminates the resistant from susceptible cultivars.

Effective early detection of potato late blight (PLB) is an essential aspect of potato cultivation. However, it is a challenge to detect late blight at an early stage in fields with conventional imaging approaches because of the lack of visual cues displayed at the canopy level. Hyperspectral imaging can, capture spectral signals from a wide range of wavelengths also outside the visual wavelengths. In this context, we propose a deep learning classification architecture for hyperspectral images by combining 2D convolutional neural network (2D-CNN) and 3D-CNN with deep cooperative attention networks (PLB-2D-3D-A). First, 2D-CNN and 3D-CNN are used to extract rich spectral space features, and then the attention mechanism AttentionBlock and SE-ResNet are used to emphasize the salient features in the feature maps and increase the generalization ability of the model. The dataset is built with 15,360 images (64x64x204), cropped from 240 raw images captured in an experimental field with over 20 potato genotypes. The accuracy in the test dataset of 2000 images reached 0.739 in the full band and 0.790 in the specific bands (492nm, 519nm, 560nm, 592nm, 717nm and 765nm). This study shows an encouraging result for early detection of PLB with deep learning and proximal hyperspectral imaging.

The foliar microbiome can extend the host plant phenotype by expanding its genomic and metabolic capabilities. Despite increasing recognition of the importance of the foliar microbiome for plant fitness, stress physiology, and yield, the diversity, function, and contribution of foliar microbiomes to plant phenotypic traits remain largely elusive. The recent adoption of high-throughput technologies is helping to unravel the diversityand spatiotemporal dynamics of foliar microbiomes, but we have yet to resolve their functional importance for plant growth, development, and ecology. Here, we focus on the processes that govern the assembly of the foliar microbiome and the potential mechanisms involved in extended plant phenotypes. We highlight knowledge gaps and provide suggestions for new research directions that can propel the field forward. These efforts will be instrumental in maximizing the functional potential of the foliar microbiome for sustainable crop production.

Field phenotyping of crops has recently gained considerable attention leading to the development of new protocols for recording plant traits of interest. Phenotyping in field conditions can be performed by various cameras, sensors, and imaging platforms. In this chapter, practical aspects as well as advantages and disadvantages of aboveground phenotyping platforms are highlighted with a focus on drone-based imaging and relevant image analysis for field conditions. It includes useful planning tips for experimental design as well as protocols, sources, and tools for image acquisition, preprocessing, feature extraction, and machine learning highlighting the possibilities with computer vision. Several open and free resources are given to speed up data analysis for biologists.This chapter targets professionals and researchers with limited computational background performing or wishing to perform phenotyping of field crops, especially with a drone-based platform. The advice and methods described focus on potato but can mostly be used for field phenotyping of any crops.

The plant pathogen Phytophthora infestans causes the severe disease late blight in potato, which can result in huge yield loss for potato production. Automatic and accurate disease lesion segmentation enables fast evaluation of disease severity and assessment of disease progress. In tasks requiring computer vision, deep learning has recently gained tremendous success for image classification, object detection and semantic segmentation. To test whether we could extract late blight lesions from unstructured field environments based on high-resolution visual field images and deep learning algorithms, we collected ∼500 field RGB images in a set of diverse potato genotypes with different disease severity (0%–70%), resulting in 2100 cropped images. 1600 of these cropped images were used as the dataset for training deep neural networks and 250 cropped images were randomly selected as the validation dataset. Finally, the developed model was tested on the remaining 250 cropped images. The results show that the values for intersection over union (IoU) of the classes background (leaf and soil) and disease lesion in the test dataset were 0.996 and 0.386, respectively. Furthermore, we established a linear relationship (R2=0.655) between manual visual scores of late blight and the number of lesions detected by deep learning at the canopy level. We also showed that imbalance weights of lesion and background classes improved segmentation performance, and that fused masks based on the majority voting of the multiple masks enhanced the correlation with the visual disease scores. This study demonstrates the feasibility of using deep learning algorithms for disease lesion segmentation and severity evaluation based on proximal imagery, which could aid breeding for crop resistance in field environments, and also benefit precision farming.

Increasing agricultural losses due to biotic and abiotic stresses caused by climate change challenge food security worldwide. A promising strategy to sustain crop productivity under conditions of limited water availability is the use of plant growth promoting rhizobacteria (PGPR). Here, the effects of spore forming Bacillus licheniformis (FMCH001) on growth and physiology of maize (Zea mays L. cv. Ronaldinho) under well-watered and drought stressed conditions were investigated. Pot experiments were conducted in the automated high-throughput phenotyping platform PhenoLab and under greenhouse conditions. Results of the PhenoLab experiments showed that plants inoculated with B. licheniformis FMCH001 exhibited increased root dry weight (DW) and plant water use efficiency (WUE) compared to uninoculated plants. In greenhouse experiments, root and shoot DW significantly increased by more than 15% in inoculated plants compared to uninoculated control plants. Also, the WUE increased in FMCH001 plants up to 46% in both well-watered and drought stressed plants. Root and shoot activities of 11 carbohydrate and eight antioxidative enzymes were characterized in response to FMCH001 treatments. This showed a higher antioxidant activity of catalase (CAT) in roots of FMCH001 treated plants compared to uninoculated plants. The higher CAT activity was observed irrespective of the water regime. These findings show that seed coating with Gram positive spore forming B. licheniformis could be used as biostimulants for enhancing plant WUE under both normal and drought stress conditions.

Red fescue (Festuca rubra) is used in seed mixtures for lawns and pastures. It is prone to lodge at flowering, and plant growth regulators (PGRs) are used to prevent lodging, ensuring sufficient pollination. Seed yield and lodging were studied over three years in a red fescue field established with four seeding rates (2, 4, 6 and 8 kg ha−1) and sprayed each year with three doses of the PGR trinexapac-ethyl (250 g L−1) (0, 0.3, 0.6 and 1.2 L ha−1). Half of each plot was sprayed with the PGR and the other half was left unsprayed as control. The degree of lodging was assessed by analysing drone images in the second year of the experiment and using a 10-point scale for scoring lodging at the ground. Generally, application of PGR increased the seed yield but the effect varied between years. There was no interaction between the PGR dosage and seeding rate. We found a positive correlation between the blue intensity of the images and lodging. PGR dosage significantly affected lodging evaluated by visual ranking and the blue intensity of the images, while the seeding rates did not affect lodging. Lodging affected seed yield negatively.

In agricultural research involving unmanned aerial vehicle (UAV) imagery, the quality of measurements is often benchmarked against ground measurements. This study adopted a supplementary approach by introducing repeatability and reproducibility analyses. Repeatability was defined as the variation associated with the measurement itself, and reproducibility was defined as the impact of different ways of measuring or different measuring conditions. In this study, different ways of measuring were represented by different cameras, and different light conditions and flight altitudes represented different measuring conditions. The aim was to investigate the repeatability and reproducibility of measurements based on consumer-grade colour cameras mounted on UAVs without calibration of incoming light. Three field-plot experiments with different genotypes of winter cereals were used to investigate how repeatability and reproducibility affected comparisons of genotypes based on the excess green vegetation index (ExG). Four different statistical methods were used to analyse data. The results showed little variation between repeated measurements based on UAV imagery, and that the variation was smaller than ground measurements of crop coverage (CC) based on sampling. Reproducibility was mainly influenced by the cameras. When incoming light was stable during image capture, light conditions and flight altitude had little or no impact on reproducibility. Shifting light conditions during image capture reduced it. Different cameras resulted in different ExG levels for the genotypes, whereas flight altitude and light conditions were found to be of minor importance for genotype comparisons. The extent to which the camera posed a problem was discussed and potential solutions proposed.