Extensive Annotation Research Articles

PIPdb: a comprehensive plasmid sequence resource for tracking the horizontal transfer of pathogenic factors and antimicrobial resistance genes.

Plasmids, as independent genetic elements, carrying resistance or virulence genes and transfer them among different pathogens, posing a significant threat to human health. Under the 'One Health' approach, it is crucial to control the spread of plasmids carrying such genes. To achieve this, a comprehensive characterization of plasmids in pathogens is essential. Here we present the Plasmids in Pathogens Database (PIPdb), a pioneering resource that includes 792964 plasmid segment clusters (PSCs) derived from 1 009571 assembled genomes across 450 pathogenic species from 110 genera. To our knowledge, PIPdb is the first database specifically dedicated to plasmids in pathogenic bacteria, offering detailed multi-dimensional metadata such as collection date, geographical origin, ecosystem, host taxonomy, and habitat. PIPdb also provides extensive functional annotations, including plasmid type, insertion sequences, integron, oriT, relaxase, T4CP, virulence factors genes, heavy metal resistance genesand antibiotic resistance genes. The database features a user-friendly interface that facilitates studies on plasmids across diverse host taxa, habitats, and ecosystems, with a focus on those carrying antimicrobial resistance genes (ARGs). We have integrated online tools for plasmid identification and annotation from assembled genomes. Additionally, PIPdb includes a risk-scoring system for identifying potentially high-risk plasmids. The PIPdb web interface is accessible at https://nmdc.cn/pipdb.

RPS 2.0: an updated database of RNAs involved in liquid-liquid phase separation.

Liquid-liquid phase separation (LLPS) is a crucial process for the formation of biomolecular condensates such as coacervate droplets, P-bodies and stress granules, which play critical roles in many physiological and pathological processes. Increasing studies have shown that not only proteins but also RNAs play a critical role in LLPS. To host LLPS-associated RNAs, we previously developed a database named 'RPS' in 2021. In this study, we present an updated version RPS 2.0 (https://rps.renlab.cn/) to incorporate the newly generated data and to host new LLPS-associated RNAs driven by post-transcriptional regulatory mechanisms. Currently, RPS 2.0 hosts 171 301 entries of LLPS-associated RNAs in 24 different biomolecular condensates with four evidence types, including 'Reviewed', 'High-throughput (LLPS enrichment)', 'High-throughput (LLPS perturbation)' and 'Predicted', and five event types, including 'Expression', 'APA', 'AS', 'A-to-I' and 'Modification'. Additionally, extensive annotations of LLPS-associated RNAs are provided in RPS 2.0, including RNA sequence and structure features, RNA-protein/RNA-RNA interactions, RNA modifications, as well as diseases related annotations. We expect that RPS 2.0 will further promote research of LLPS-associated RNAs and deepen our understanding of the biological functions and regulatory mechanisms of LLPS.

LncPepAtlas: a comprehensive resource for exploring the translational landscape of long non-coding RNAs.

Long non-coding RNAs were commonly viewed as non-coding elements. However, they are increasingly recognized for their ability to be translated into proteins, thereby playing a significant role in various cellular processes and diseases. With developments in biotechnology and computational algorithms, a range of novel approaches are being applied to investigate the translation of long non-coding RNA (lncRNAs). Herein, we developed the LncPepAtlas database (http://www.cnitbiotool.net/LncPepAtlas/), which aims to compile multiple evidences for the translation of lncRNAs and annotations for the upstream regulation of lncRNAs across various species. LncPepAtlas integrated compelling evidence from nine distinct sources for the translation of lncRNAs. These include a dataset comprising 2631 publicly available Ribo-seq samples from nine species, which has been collected and analysed. LncPepAtlas offers extensive annotation for lncRNA upstream regulation and expression profiles across various cancers, tissues or cell lines at transcriptional and translational levels. Importantly, it enables novel antigen predictions for lncRNA-encoded peptides. By identifying numerous peptide candidates that could potentially bind to major histocompatibility complex class I and II molecules, this work may provide new insights into cancer immunotherapy. The function of peptides were inferred by aligning them with experimentally detected proteins. LncPepAtlas aims to become a convenient resource for exploring translatable lncRNAs.

Augmenting biomedical named entity recognition with general-domain resources

ObjectiveTraining a neural network-based biomedical named entity recognition (BioNER) model usually requires extensive and costly human annotations. While several studies have employed multi-task learning with multiple BioNER datasets to reduce human effort, this approach does not consistently yield performance improvements and may introduce label ambiguity in different biomedical corpora. We aim to tackle those challenges through transfer learning from easily accessible resources with fewer concept overlaps with biomedical datasets. MethodsWe proposed GERBERA, a simple-yet-effective method that utilized general-domain NER datasets for training. We performed multi-task learning to train a pre-trained biomedical language model with both the target BioNER dataset and the general-domain dataset. Subsequently, we fine-tuned the models specifically for the BioNER dataset. ResultsWe systematically evaluated GERBERA on five datasets of eight entity types, collectively consisting of 81,410 instances. Despite using fewer biomedical resources, our models demonstrated superior performance compared to baseline models trained with additional BioNER datasets. Specifically, our models consistently outperformed the baseline models in six out of eight entity types, achieving an average improvement of 0.9% over the best baseline performance across eight entities. Our method was especially effective in amplifying performance on BioNER datasets characterized by limited data, with a 4.7% improvement in F1 scores on the JNLPBA-RNA dataset. ConclusionThis study introduces a new training method that leverages cost-effective general-domain NER datasets to augment BioNER models. This approach significantly improves BioNER model performance, making it a valuable asset for scenarios with scarce or costly biomedical datasets. We make data, codes, and models publicly available via https://github.com/qingyu-qc/bioner_gerbera.

Beyond low-dimensional features: Enhancing semi-supervised medical image semantic segmentation with advanced consistency learning techniques

In medical imaging, semantic segmentation is crucial for accurate diagnosis. However, it is constrained by the scarcity of labeled data. To reduce the dependency on extensive annotations and improve the model’s learning efficiency and generalization capability, semi-supervised learning merges a small amount of labeled data with a large pool of unlabeled data. Unfortunately, current mainstream methods have not adapted to the subtle variations in low-dimensional features that are characteristic of medical images. Our method proposes a new consistency learning approach, specifically tailored to capture subtle nuances in low-dimensional features, such as the spatial arrangement of structures typical in medical images. It utilises a two-strong and two-weak perturbation strategy to compute multiple sets of losses. The introduction of a Multi-Head Feature Decoder module compels the model to learn beyond just colour and texture changes, enabling accurate predictions in areas with subtle variations. In the scenario where the proportion of labeled data is at its lowest, our method outperforms the current best methods on three datasets in terms of DICE scores by 2.49%, 2.57%, and 3.77%, respectively, demonstrating its significant superiority.

MedLSAM: Localize and segment anything model for 3D CT images

Recent advancements in foundation models have shown significant potential in medical image analysis. However, there is still a gap in models specifically designed for medical image localization. To address this, we introduce MedLAM, a 3D medical foundation localization model that accurately identifies any anatomical part within the body using only a few template scans. MedLAM employs two self-supervision tasks: unified anatomical mapping (UAM) and multi-scale similarity (MSS) across a comprehensive dataset of 14,012 CT scans. Furthermore, we developed MedLSAM by integrating MedLAM with the Segment Anything Model (SAM). This innovative framework requires extreme point annotations across three directions on several templates to enable MedLAM to locate the target anatomical structure in the image, with SAM performing the segmentation. It significantly reduces the amount of manual annotation required by SAM in 3D medical imaging scenarios. We conducted extensive experiments on two 3D datasets covering 38 distinct organs. Our findings are twofold: (1) MedLAM can directly localize anatomical structures using just a few template scans, achieving performance comparable to fully supervised models; (2) MedLSAM closely matches the performance of SAM and its specialized medical adaptations with manual prompts, while minimizing the need for extensive point annotations across the entire dataset. Moreover, MedLAM has the potential to be seamlessly integrated with future 3D SAM models, paving the way for enhanced segmentation performance. Our code is public at https://github.com/openmedlab/MedLSAM.

A large structural variant collection in Holstein cattle and associated database for variant discovery, characterization, and application

BackgroundStructural variants (SVs) such as deletions, duplications, and insertions are known to contribute to phenotypic variation but remain challenging to identify and genotype. A more complete, accessible, and assessable collection of SVs will assist efforts to study SV function in cattle and to incorporate SV genotyping into animal evaluation.ResultsIn this work we produced a large and deeply characterized collection of SVs in Holstein cattle using two popular SV callers (Manta and Smoove) and publicly available Illumina whole-genome sequence (WGS) read sets from 310 samples (290 male, 20 female, mean 20X coverage). Manta and Smoove identified 31 K and 68 K SVs, respectively. In total the SVs cover 5% (Manta) and 6% (Smoove) of the reference genome, in contrast to the 1% impacted by SNPs and indels. SV genotypes from each caller were confirmed to accurately recapitulate animal relationships estimated using WGS SNP genotypes from the same dataset, with Manta genotypes outperforming Smoove, and deletions outperforming duplications. To support efforts to link the SVs to phenotypic variation, overlapping and tag SNPs were identified for each SV, using genotype sets extracted from the WGS results corresponding to two bovine SNP chips (BovineSNP50 and BovineHD). 9% (Manta) and 11% (Smoove) of the SVs were found to have overlapping BovineHD panel SNPs, while 21% (Manta) and 9% (Smoove) have BovineHD panel tag SNPs. A custom interactive database (https://svdb-dc.pslab.ca) containing the identified sequence variants with extensive annotations, gene feature information, and BAM file content for all SVs was created to enable the evaluation and prioritization of SVs for further study. Illustrative examples involving the genes POPDC3, ORM1, G2E3, FANCI, TFB1M, FOXC2, N4BP2, GSTA3, and COPA show how this resource can be used to find well-supported genic SVs, determine SV breakpoints, design genotyping approaches, and identify processed pseudogenes masquerading as deletions.ConclusionsThe resources developed through this study can be used to explore sequence variation in Holstein cattle and to develop strategies for studying SVs of interest. The lack of overlapping and tag SNPs from commonly used SNP chips for most of the SVs suggests that other genotyping approaches will be needed (for example direct genotyping) to understand their potential contributions to phenotype. The included SV genotype assessments point to challenges in characterizing SVs, especially duplications, using short-read data and support ongoing efforts to better characterize cattle genomes through long-read sequencing. Lastly, the identification of previously known functional SVs and additional CDS-overlapping SVs supports the phenotypic relevance of this dataset.

Generalizable Error Modeling for Human Data Annotation: Evidence From an Industry-Scale Search Data Annotation Program

Machine learning (ML) and artificial intelligence (AI) systems rely heavily on human-annotated data for training and evaluation. A major challenge in this context is the occurrence of annotation errors, as their effects can degrade model performance. This paper presents a predictive error model trained to detect potential errors in search relevance annotation tasks for three industry-scale ML applications (music streaming, video streaming, and mobile apps). Drawing on data from an extensive search relevance annotation program, we demonstrate that errors can be predicted with moderate model performance (AUC=0.65-0.75) and that model performance generalizes well across applications (i.e., a global, task-agnostic model performs on par with task-specific models). In contrast to past research, which has often focused on predicting annotation labels from task-specific features, our model is trained to predict errors directly from a combination of task features and behavioral features derived from the annotation process, in order to achieve a high degree of generalizability. We demonstrate the usefulness of the model in the context of auditing, where prioritizing tasks with high predicted error probabilities considerably increases the amount of corrected annotation errors (e.g., 40% efficiency gains for the music streaming application). These results highlight that behavioral error detection models can yield considerable improvements in the efficiency and quality of data annotation processes. Our findings reveal critical insights into effective error management in the data annotation process, thereby contributing to the broader field of human-in-the-loop ML.

Abstract C003: Multimodal and multiplexed spatial regulation of the breast collagen proteome: Comparative analysis in Black and White women from normal breast tissue and triple-negative breast cancer

Abstract Non-Hispanic Black women (BW) are diagnosed with more aggressive breast cancer subtypes at younger ages and have higher mortality rates compared to non-Hispanic White women (WW). Within the breast microenvironment emergent data shows that stromal collagen plays a significant role in aggressiveness of breast cancer linked to ancestral traits. However, the molecular composition of collagen stroma has yet to be linked to ancestry and aggressiveness of breast cancer. The purpose of this study was to evaluate the microenvironment within ancestry-defined normal breast and triple-negative breast cancer (TNBC) to understand spatial regulation of collagen stroma that may link to ancestral differences in cancer. Multimodal, multiplexed spatial analyses were performed. Techniques included Second Harmonic Generation (SHG) to measure collagen fibers, mass spectrometry imaging to measure the collagen proteome, and mass-tagged antibody panels to investigate immune signatures in TNBC. Normal breast tissues were obtained from the Susan G. Komen Tissue Bank, African American (AA; n=20) and European American (EA; n=20) samples were annotated with risk scores, age, Body Mass Index (BMI), Breast Imaging-Reporting and Data System (BI-RADS), continental ancestry, and family history. Triple-negative breast cancer (TNBC, n=80) tissues from BW with extensive clinical annotations including age, BMI, stage, metastasis, and survival were evaluated in comparison to 20+ TNBC tissues from WW. In the normal breast cohort, evaluation of BMI (overweight and obese) within ancestry groups showed significant but opposing differences in fiber width measurements. As BMI increases, AA showed statistically thicker fibers while EA showed statistically thinner fibers. At the protein level, two types of collagens were significantly higher in EA and one type is higher in AA. Within the TNBC cohort, collagen fiber length and width measurements by SHG were significantly shorter and thinner in stage II compared to stage I but no other stage comparisons were significant. This may indicate that ECM remodeling occurs at earlier stages in TNBC. Initial analyses showed that collagen peptide intensities significantly vary by TNBC stage (75 peptides, ANOVA test FDR p-value < 0.05). In the cohort of women who survived, 90 ECM peptides significantly correlated with collagen fiber widths, indicating a relationship between collagen physical and molecular properties within a defined region. Seven immune markers defined differences in survival when grouped by high vs low expression within tumor regions. These markers include CD20 (Hazard Ratio HR = 7.99, p-value = 0.002) and PD-1 (HR = 3.68, p-value = 0.048), which have both been studied in breast cancer. Current work focuses on collagen peptide post-translational modifications that may differ in TNBC patients by vital status and are detected in normal breast at risk of cancer. These findings support that molecular composition in breast stroma contributes to aggressive ancestral traits of breast cancer risk and triple-negative breast cancer. Citation Format: Jaclyn B. Dunne, Heather Jensen-Smith, Laura Spruill, Taylor S. Hulahan, Jade K. Macdonald, George E. Sandusky, M. A. Hollingsworth, Anand S. Mehta, Richard R. Drake, Jeffrey R. Marks, Harikrishna Nakshatri, Marvella E. Ford, Peggi M. Angel. Multimodal and multiplexed spatial regulation of the breast collagen proteome: Comparative analysis in Black and White women from normal breast tissue and triple-negative breast cancer [abstract]. In: Proceedings of the 17th AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2024 Sep 21-24; Los Angeles, CA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2024;33(9 Suppl):Abstract nr C003.

Data-centric AI approach for automated wildflower monitoring.

We present the Eindhoven Wildflower Dataset (EWD) as well as a PyTorch object detection model that is able to classify and count wildflowers. EWD, collected over two entire flowering seasons and expert annotated, contains 2,002 top-view images of flowering plants captured 'in the wild' in five different landscape types (roadsides, urban green spaces, cropland, weed-rich grassland, marshland). It holds a total of 65,571 annotations for 160 species belonging to 31 different families of flowering plants and serves as a reference dataset for automating wildflower monitoring and object detection in general. To ensure consistent annotations, we define species-specific floral count units and provide extensive annotation guidelines. With a 0.82 mAP (@IoU > 0.50) score the presented baseline model, trained on a balanced subset of EWD, is to the best of our knowledge superior in its class. Our approach empowers automated quantification of wildflower richness and abundance, which helps understanding and assessing natural capital, and encourages the development of standards for AI-based wildflower monitoring. The annotated EWD dataset and the code to train and run the baseline model are publicly available.

Accelerating histopathology workflows with generative AI-based virtually multiplexed tumour profiling

Understanding the spatial heterogeneity of tumours and its links to disease initiation and progression is a cornerstone of cancer biology. Presently, histopathology workflows heavily rely on hematoxylin and eosin and serial immunohistochemistry staining, a cumbersome, tissue-exhaustive process that results in non-aligned tissue images. We propose the VirtualMultiplexer, a generative artificial intelligence toolkit that effectively synthesizes multiplexed immunohistochemistry images for several antibody markers (namely AR, NKX3.1, CD44, CD146, p53 and ERG) from only an input hematoxylin and eosin image. The VirtualMultiplexer captures biologically relevant staining patterns across tissue scales without requiring consecutive tissue sections, image registration or extensive expert annotations. Thorough qualitative and quantitative assessment indicates that the VirtualMultiplexer achieves rapid, robust and precise generation of virtually multiplexed imaging datasets of high staining quality that are indistinguishable from the real ones. The VirtualMultiplexer is successfully transferred across tissue scales and patient cohorts with no need for model fine-tuning. Crucially, the virtually multiplexed images enabled training a graph transformer that simultaneously learns from the joint spatial distribution of several proteins to predict clinically relevant endpoints. We observe that this multiplexed learning scheme was able to greatly improve clinical prediction, as corroborated across several downstream tasks, independent patient cohorts and cancer types. Our results showcase the clinical relevance of artificial intelligence-assisted multiplexed tumour imaging, accelerating histopathology workflows and cancer biology.

Enhancing pavement crack segmentation via semantic diffusion synthesis model for strategic road assessment

Computer-aided deep learning has significantly advanced road crack segmentation. However, supervised models face challenges due to limited annotated images. There is also a lack of emphasis on deriving pavement condition indices from predicted masks. This article introduces a novel semantic diffusion synthesis model that creates synthetic crack images from segmentation masks. The model is optimized in terms of architectural complexity, noise schedules, and condition scaling. The optimal architecture outperforms state-of-the-art semantic synthesis models across multiple benchmark datasets, demonstrating superior image quality assessment metrics. The synthetic frames augment these datasets, resulting in segmentation models with significantly improved efficiency. This approach enhances results without extensive data collection or annotation, addressing a key challenge in engineering. Finally, a refined pavement condition index has been developed for automated end-to-end defect detection systems, promoting more effective maintenance planning.

The N6-methyladenosine Epitranscriptomic Landscape of Lung Adenocarcinoma.

Comprehensive m6A epitranscriptome profiling of primary tumors remains largely uncharted. Here, we profiled the m6A epitranscriptome of 10 non-neoplastic lung (NL) tissues and 51 lung adenocarcinoma (LUAD) tumors, integrating the corresponding transcriptome, proteome and extensive clinical annotations. We identified distinct clusters and genes that were exclusively linked to disease progression through m6A modifications. In comparison with NL tissues, we identified 430 transcripts to be hypo-methylated and 222 to be hyper-methylated in tumors. Among these genes, EML4 emerged as a novel metastatic driver, displaying significant hyper-methylation in tumors. m6A modification promoted the translation of EML4, leading to its widespread overexpression in primary tumors. Functionally, EML4 modulated cytoskeleton dynamics through interacting with ARPC1A, enhancing lamellipodia formation, cellular motility, local invasion, and metastasis. Clinically, high EML4 protein abundance correlated with features of metastasis. METTL3 small molecule inhibitor markedly diminished both EML4 m6A and protein abundance, and efficiently suppressed lung metastases in vivo.

DynaSeg: A deep dynamic fusion method for unsupervised image segmentation incorporating feature similarity and spatial continuity

Our work tackles the fundamental challenge of image segmentation in computer vision, which is crucial for diverse applications. While supervised methods demonstrate proficiency, their reliance on extensive pixel-level annotations limits scalability. We introduce DynaSeg, an innovative unsupervised image segmentation approach that overcomes the challenge of balancing feature similarity and spatial continuity without relying on extensive hyperparameter tuning. Unlike traditional methods, DynaSeg employs a dynamic weighting scheme that automates parameter tuning, adapts flexibly to image characteristics, and facilitates easy integration with other segmentation networks. By incorporating a Silhouette Score Phase, DynaSeg prevents undersegmentation failures where the number of predicted clusters might converge to one. DynaSeg uses CNN-based and pre-trained ResNet feature extraction, making it computationally efficient and more straightforward than other complex models. Experimental results showcase state-of-the-art performance, achieving a 12.2% and 14.12% mIOU improvement over current unsupervised segmentation approaches on COCO-All and COCO-Stuff datasets, respectively. We provide qualitative and quantitative results on five benchmark datasets, demonstrating the efficacy of the proposed approach. Code available at \\url{https://github.com/RyersonMultimediaLab/DynaSeg}

Weakly supervised semantic segmentation of leukocyte images based on class activation maps.

Leukocytes are an essential component of the human defense system, accurate segmentation of leukocyte images is a crucial step towards automating detection. Most existing methods for leukocyte images segmentation relied on fully supervised semantic segmentation (FSSS) with extensive pixel-level annotations, which are time-consuming and labor-intensive. To address this issue, this paper proposes a weakly supervised semantic segmentation (WSSS) approach for leukocyte images utilizing improved class activation maps (CAMs). Firstly, to alleviate ambiguous boundary problem between leukocytes and background, preprocessing technique is employed to enhance the image quality. Secondly, attention mechanism is added to refine the CAMs generated by improving the matching of local and global features. Random walks, dense conditional random fields and hole filling were leveraged to obtain final pseudo-segmentation labels. Finally, a fully supervised segmentation network is trained with pseudo-segmentation labels. The method is evaluated on BCCD and TMAMD datasets. Experimental results demonstrate that by employing the pseudo segmentation annotations generated through this method can be utilized to train UNet as close as possible to FSSS. This method effectively reduces manual annotation cost while achieving WSSS of leukocyte images.

Progressive expansion for semi-supervised bi-modal salient object detection

Existing bi-modal salient object detection (SOD) methods primarily rely on fully supervised training strategies that require extensive manual annotation. Undoubtedly, extensive manual annotation is time-consuming and laborious, and the fully supervised strategy is also prone to overfitting on the training set. Therefore, we introduce a semi-supervised learning architecture (SSLA) to alleviate these problems while ensuring detection performance. Considering that the inherent training mode and concise architecture of basic SSLA will limit its ability to effectively explore the learning potential of the model, we further propose two optimization strategies, dynamic adjustment and active expansion. Specifically, we dynamically adjust the supervision scheme for unlabeled samples during training so that the model can continuously utilize the model’s gains (pseudo labels) to supervise and guide the model to further explore the unlabeled samples. Furthermore, the active expansion strategy enables the model to acquire more beneficial supervised information and focuses its attention on difficult-to-segment samples. In summary, an effective progressive expansion network (PENet) architecture for semi-supervised bi-modal SOD is proposed. Extensive experiments indicate that our PENet architecture, while effectively alleviating 90% of annotation burdens, has achieved highly competitive results in RGB-T and RGB-D tasks compared to fully supervised methods. The performance is even more pronounced during cross-dataset testing.

Discovery and generalization of tissue structures from spatial omics data

Tissues are organized into anatomical and functional units at different scales. New technologies for high-dimensional molecular profiling in situ have enabled the characterization of structure-function relationships in increasing molecular detail. However, it remains a challenge to consistently identify key functional units across experiments, tissues, and disease contexts, a task that demands extensive manual annotation. Here, we present spatial cellular graph partitioning (SCGP), a flexible method for the unsupervised annotation of tissue structures. We further present a reference-query extension pipeline, SCGP-Extension, that generalizes reference tissue structure labels to previously unseen samples, performing data integration and tissue structure discovery. Our experiments demonstrate reliable, robust partitioning of spatial data in a wide variety of contexts and best-in-class accuracy in identifying expertly annotated structures. Downstream analysis on SCGP-identified tissue structures reveals disease-relevant insights regarding diabetic kidney disease, skin disorder, and neoplastic diseases, underscoring its potential to drive biological insight and discovery from spatial datasets.

Dirty road extraction from GF-2 images by semi-supervised deep learning method for arid and semiarid regions of southern Mongolia

ABSTRACT The uncontrolled proliferation of natural roads in arid regions has exacerbated regional land degradation and desertification, presenting substantial challenges to their accurate mapping owing to their dynamic and obscure features. Moreover, the high cost of data annotation restricts the availability of comprehensively labelled datasets, which are essential for advanced remote sensing processing and natural road detection. This study dedicated to implement a semi-supervised deep learning method for dirty road extraction in southern Mongolia. A new thematic semantic segmentation dataset of natural roads was established firstly to address scarcity of annotation datasets this region. A semi-supervised UniMatch structure was designed consequently. Operating with high-resolution GaoFen-2 images, this approach minimises the need for extensive manual annotation, achieving an IOU of 73.51% and MIOU of 86.37%. This method significantly reduces labour and time costs associated with manual and fully supervised methods. These observations provide a valuable data source and methodology for addressing natural road expansion in arid regions. They can aid governments in evaluating transportation infrastructure in remote areas, and analysing dirty road traffic impact on environment.

Identification and characterization of four Bacillus species from the intestine of hybrid grouper (Epinephelus fuscoguttatus♀ × E. lanceolatus♂), their antagonistic role on common pathogenic bacteria, and effects on intestinal health

As an alternative to the criticized antibiotics, probiotics have been adopted for their eco-friendly nature and ability to enhance host growth and immunity. Nevertheless, reports suggest ineffectiveness in commercially available probiotics since most are from non-fish sources; thus, this study was envisaged to isolate and characterize new Bacillus spp. from the gut of hybrid grouper (Epinephelus fuscoguttatus♀ × Epinephelus lanceolatus♂) which could serve as potential probiotics. The isolation and characterization were performed based on their morphological and biochemical properties, and 16S rRNA sequencing homology analysis. A subsequent 30-day in vivo biosafety feeding trial was conducted to ascertain isolates' non-pathogenicity, as well as their effects on fish growth, and intestinal mucosal microvilli via scanning electron microscopy (SEM) analysis. Four Bacillus spp. strains, namely, B. velezensis strain PGSAK01 (accession number OQ726606), B. stercoris strain PGSAK05 (accession number OQ726607), B. velezensis strain PGSAK17 (accession number OQ726601), and B. subtilis strain PGSAK19 (accession number OQ726605), were identified and characterized in the current study. The strains showed promising probiotic properties such higher adhesion capability, higher thermotolerance, displaying higher survivability to 0.5 % bile, lower pH tolerance, γ-haemolytic activity, and multispecies characteristics. Among the 24 antibiotics tested, while all isolates showed susceptibility to 21, the PGSAK01 strain showed resistance to furazolidone antibiotics. None of the isolates showed possession of i) virulence factor genes encoding enterotoxigenic (hblA, hblC, hblD, nheA, nheB, and entFM) and emetic (cereulide synthetase gene, ces) genes, and ii) streptomycin resistance gene (vat c), ampicillin-resistant genes (mecA and bla), and vancomycin-resistant gene (van B). Nevertheless, the PGSAK01 and PGSAK17 strains showed possession of tek K, cat, and ant(4′)-Ia (adenylyltransferase) (except the PGSAK01) resistant genes. All isolates displayed better antimicrobial effects against pathogenic bacteria Streptococcus agalactiae, S. iniae, Vibrio harveyi, and V. alginolyticus. The in vivo biosafety trial involved hybrid grouper fish being grouped into five (average weight 32 ± 0.94 g), namely, the group fed the basal diet void of isolate's supplementation (control), and the remaining four groups fed the basal diet with 1 × 108 CFU/g diet of individual strain PGSAK01, PGSAK05, PGSAK17, and PGSAK19 supplementation. At the end of the study, a significantly higher WGR, K (except the PGSAK01 group), VSI; lysozyme (except PGSAK01 group), total antioxidant activity, alkaline phosphatase, superoxide dismutase enzyme activities; highly dense intestinal mucosal villi (based on the scanning electron microscopy analysis); and significantly lower malondialdehyde levels were witnessed in the isolated treated groups compared to the control, supporting the results obtained in the auto-aggregation and cell-surface hydrophobicity test. This work's results have provided thought-provoking targets; thus, studies involving extensive genome sequencing and functional annotation analysis will be explored to offer unfathomable insights into their mechanisms of action and potential health benefits, further establishing the four Bacillus strains' (PGSAK01, PGSAK05, PGSAK17, and PGSAK19) potential role in probiotic fields and functional foods.

Annotation-efficient deep learning for breast cancer whole-slide image classification using tumour infiltrating lymphocytes and slide-level labels

Tumour-Infiltrating Lymphocytes (TILs) are pivotal in the immune response against cancer cells. Existing deep learning methods for TIL analysis in whole-slide images (WSIs) demand extensive patch-level annotations, often requiring labour-intensive specialist input. To address this, we propose a framework named annotation-efficient segmentation and attention-based classifier (ANSAC). ANSAC requires only slide-level labels to classify WSIs as having high vs. low TIL scores, with the binary classes divided by an expert-defined threshold. ANSAC automatically segments tumour and stroma regions relevant to TIL assessment, eliminating extensive manual annotations. Furthermore, it uses an attention model to generate a map that highlights the most pertinent regions for classification. Evaluating ANSAC on four breast cancer datasets, we demonstrate substantial improvements over three baseline methods in identifying TIL-relevant regions, with up to 8% classification improvement on a held-out test dataset. Additionally, we propose a pre-processing modification to a well-known method, enhancing its performance up to 6%.