Unbiased Classification Research Articles

Comprehensive transcriptomic analysis identifies three distinct subtypes of pituitary adenomas: insights into tumor behavior, prognosis, and stem cell characteristics

BackgroundPituitary adenomas (PAs) are the second most common intracranial tumor. While current diagnostic practices rely primarily on histological testing, they often fail to capture the molecular complexities of pituitary adenomas, underscoring the need for a molecular-based classification to refine therapeutic strategies and prognostic assessments. This study aims to provide a molecularly unbiased classification of pituitary adenomas and explore their unique gene expression patterns and clinical features.MethodsWe performed unsupervised hierarchical clustering of the gene expression profiles of 117 PA samples to identify three distinct molecular subtypes. Subsequently, we analyzed the compiled transcriptomic profiles of each individual subtype for pathway enrichment. We also validated the new classification with a validation set containing 158 PAs and 24 pituitary adenoma stem cells (PASCs).ResultsConsensus clustering of transcriptomic data from 117 pituitary adenoma (PA) samples identified three distinct molecular subtypes, each showing unique gene expression patterns and associated biological processes: Group I is enriched in signaling pathways, such as the cAMP signaling pathway and the calcium signaling pathway. Group II is primarily related to metabolic processes, including nitrogen metabolism and arginine biosynthesis in cancer. Group III predominantly shows enrichment in immune responses and potential malignant transformation of the disease, especially through cancer-related pathways such as the JAK–STAT signaling pathway and the PI3K–Akt signaling pathway. The immune profiling revealed distinct patterns for each subtype: Group I had higher dendritic cells and fewer CD8+ T cells, Group II had more monocytes and macrophages, and Group III had elevated levels of T cells. Additionally, there were differences in clinical characteristics and prognosis among the subtypes, with Group III having a worse prognosis, despite the smaller tumor size compared to other groups. Notably, differences in PASCs correlated with the molecular subtypes, with Group III stem cells being enriched in tumorigenesis pathways, PI3K–Akt signaling pathway and Ras signaling pathway.ConclusionOur study introduces a novel molecular classification for pituitary adenomas, independent of traditional histological methods. Each subtype features distinct genetic, molecular, and immunological profiles. We have isolated pituitary adenoma stem-like cells (PASCs), pairing them with tumor tissues for detailed transcriptomic analysis. These PASCs exhibit diverse molecular traits consistent with the new classification.

Role of inter- and extra-lesion tissue, transfer learning, and fine-tuning in the robust classification of breast lesions

Accurate and unbiased classification of breast lesions is pivotal for early diagnosis and treatment, and a deep learning approach can effectively represent and utilize the digital content of images for more precise medical image analysis. Breast ultrasound imaging is useful for detecting and distinguishing benign masses from malignant masses. Based on the different ways in which benign and malignant tumors affect neighboring tissues, i.e., the pattern of growth and border irregularities, the penetration degree of the adjacent tissue, and tissue-level changes, we investigated the relationship between breast cancer imaging features and the roles of inter- and extra-lesional tissues and their impact on refining the performance of deep learning classification. The novelty of the proposed approach lies in considering the features extracted from the tissue inside the tumor (by performing an erosion operation) and from the lesion and surrounding tissue (by performing a dilation operation) for classification. This study uses these new features and three pre-trained deep neuronal networks to address the challenge of breast lesion classification in ultrasound images. To improve the classification accuracy and interpretability of the model, the proposed model leverages transfer learning to accelerate the training process. Three modern pre-trained CNN architectures (MobileNetV2, VGG16, and EfficientNetB7) are used for transfer learning and fine-tuning for optimization. There are concerns related to the neuronal networks producing erroneous outputs in the presence of noisy images, variations in input data, or adversarial attacks; thus, the proposed system uses the BUS-BRA database (two classes/benign and malignant) for training and testing and the unseen BUSI database (two classes/benign and malignant) for testing. Extensive experiments have recorded accuracy and AUC as performance parameters. The results indicate that the proposed system outperforms the existing breast cancer detection algorithms reported in the literature. AUC values of 1.00 are calculated for VGG16 and EfficientNet-B7 in the dilation cases. The proposed approach will facilitate this challenging and time-consuming classification task.

Deep orthogonal multi-wavelength fusion for tomogram-free diagnosis in diffuse optical imaging

Novel portable diffuse optical tomography (DOT) devices for breast cancer lesions hold great promise for non-invasive, non-ionizing breast cancer screening. Critical to this capability is not just the identification of lesions but rather the complex problem of discriminating between malignant and benign lesions. To accurately reconstruct the highly heterogeneous tissue of a cancer lesion in healthy breast tissue using DOT, multiple wavelengths can be leveraged to maximize signal penetration while minimizing sensitivity to noise. However, these wavelength responses can overlap, capture common information, and correlate, potentially confounding reconstruction and downstream end tasks. We show that an orthogonal fusion loss regularizes multi-wavelength DOT leading to improved reconstruction and accuracy of end-to-end discrimination of malignant versus benign lesions. We further show that our raw-to-task model significantly reduces computational complexity without sacrificing accuracy, making it ideal for real-time throughput, desired in medical settings where handheld devices have severely restricted power budgets. Furthermore, our results indicate that image reconstruction is not necessary for unbiased classification of lesions with a balanced accuracy of 77% and 66% on the synthetic dataset and clinical dataset, respectively, using the raw-to-task model. Code is available at https://github.com/sfu-mial/FuseNet.

Abstract 1152: Pan-cancer single-cell and subcellular characterization of lymphoid aggregates and tertiary lymphoid structures using high-plex spatial molecular imaging

Abstract Tertiary lymphoid structures (TLSs), clusters of immune cells that form in non-lymphoid tissues, are frequently identified in a wide range of solid tumors. They function as immunological “hubs”, driving anti-tumor responses. Previous studies have demonstrated that the presence of TLSs is often associated with improved responses to immunotherapy and superior clinical outcomes. Despite these insights, our understanding of their phenotypic diversity, spatial organization, intercellular communication, and interactions with cancer cells and the tumor microenvironment remains incomplete. Existing methods lack the resolution, depth, and scale required for a thorough understanding of these complex structures. We have obtained in-house and public spatial multi-omics data generated on various tissues across 9 different cancer types using the cutting-edge single-cell spatial transcriptomics and high-plex spatial molecular imaging platforms. We have developed innovative computational approaches for the detailed characterization of TLSs at single-cell and subcellular resolution and carried out an extensive analysis of lymphoid aggregates (LAs) and TLSs in tissue sections across these different solid cancer types. We defined the phenotypic state of each individual cell within TLSs including B/plasma cells, follicular helper/regulatory/stressed/exhausted T cells, follicular dendritic cells, fibroblasts and high endothelial venules. Then, we performed unbiased phenotyping and classification of TLSs based on their cellular states, compositions and organization patterns across various types of tissues. Additionally, we performed spatial neighborhood analysis, characterized the neighboring cells of each TLS and analyzed their cell-cell communication networks. Our approach allowed an unprecedented examination of the spatial phenotypes of TLSs in large, complex tissues at super resolution, in ways that were not previously described. We comprehensively characterized TLSs across different solid tumor types. Our analysis revealed a high degree of spatial and phenotypic heterogeneity in LAs and TLSs. We created a comprehensive spatial atlas of LAs and TLSs across solid tumor types, profiled TLSs at various maturation states with distinct cellular composition and organization patterns, and discovered new TLS states. Notably, we observed distinct communication networks associated with TLS phenotypic states. Our innovative methods and analytical approaches offer a more comprehensive way to study LAs and TLSs, enhancing detection capabilities and providing deeper insights into their spatial interaction and functional phenotypes. This research holds the potential to significantly advance our understanding of TLSs and their roles in immunobiology and immunotherapy responses across a broad range of cancers and diseases. Citation Format: Yunhe Liu, Yuanyuan Zhang, Guangchun Han, Guangsheng Pei, Kyung Serk Cho, Enyu Dai, Yanshuo Chu, Jiahui Jiang, Ansam Sinjab, Luisa M. Soto, Alejandra Serrano, Jianjun Gao, Humam Kadara, Catherine Fridman, Wolf Herve Fridman, Linghua Wang. Pan-cancer single-cell and subcellular characterization of lymphoid aggregates and tertiary lymphoid structures using high-plex spatial molecular imaging [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 1152.

Transcriptomic profiling of immune cells in murine polymicrobial sepsis.

Various immune cell types play critical roles in sepsis with numerous distinct subsets exhibiting unique phenotypes even within the same cell population. Single-cell RNA sequencing (scRNA-seq) enables comprehensive transcriptome profiling and unbiased cell classification. In this study, we have unveiled the transcriptomic landscape of immune cells in sepsis through scRNA-seq analysis. We induced sepsis in mice by cecal ligation and puncture. 20 h after the surgery, the spleen and peritoneal lavage were collected. Single-cell suspensions were processed using a 10× Genomics pipeline and sequenced on an Illumina platform. Count matrices were generated using the Cell Ranger pipeline, which maps reads to the mouse reference transcriptome, GRCm38/mm10. Subsequent scRNA-seq analysis was performed using the R package Seurat. After quality control, we subjected the entire data set to unsupervised classification. Four major clusters were identified as neutrophils, macrophages, B cells, and T cells according to their putative markers. Based on the differentially expressed genes, we identified activated pathways in sepsis for each cell type. In neutrophils, pathways related to inflammatory signaling, such as NF-κB and responses to pathogen-associated molecular patterns (PAMPs), cytokines, and hypoxia were activated. In macrophages, activated pathways were the ones related to cell aging, inflammatory signaling, and responses to PAMPs. In B cells, pathways related to endoplasmic reticulum stress were activated. In T cells, activated pathways were the ones related to inflammatory signaling, responses to PAMPs, and acute lung injury. Next, we further classified each cell type into subsets. Neutrophils consisted of four clusters. Some subsets were activated in inflammatory signaling or cell metabolism, whereas others possessed immunoregulatory or aging properties. Macrophages consisted of four clusters, namely, the ones with enhanced aging, lymphocyte activation, extracellular matrix organization, or cytokine activity. B cells consisted of four clusters, including the ones possessing the phenotype of cell maturation or aging. T cells consisted of six clusters, whose phenotypes include molecular translocation or cell activation. Transcriptomic analysis by scRNA-seq has unveiled a comprehensive spectrum of immune cell responses and distinct subsets in the context of sepsis. These findings are poised to enhance our understanding of sepsis pathophysiology, offering avenues for targeting novel molecules, cells, and pathways to combat infectious diseases.

Feature learning based on connectivity estimation for unbiased mammography mass classification

Breast cancer is the most commonly diagnosed female malignancy worldwide. Recent developments in deep convolutional neural networks have shown promising performance for breast cancer detection and classification. However, due to variations in appearance and small datasets, biased features can be learned by the networks in distinguishing malignant and benign instances. To investigate these aspects, we trained a densely connected convolutional network (DenseNet) to obtain representative features of breast tissue, selecting texture features representing different physical morphological representations as the network’s inputs. Connectivity estimation, represented by a connection matrix, is proposed for feature learning. To make the network provide an unbiased prediction, we used k-nearest neighbors to find k training samples whose connection matrices are closest to the test case. When evaluated on OMI-DB we achieved improved diagnostic accuracy 73.89±2.89% compared with 71.35±2.66% for the initial CNN model, which showed a statistically significant difference (p=0.00036). The k training samples can provide visual explanations which are useful in understanding the model predictions and failures of the model.

From images to features: unbiased morphology classification via variational auto-encoders and domain adaptation

ABSTRACT We present a novel approach for the dimensionality reduction of galaxy images by leveraging a combination of variational auto-encoders (VAEs) and domain adaptation (DA). We demonstrate the effectiveness of this approach using a sample of low-redshift galaxies with detailed morphological type labels from the Galaxy Zoo Dark Energy Camera Legacy Survey (DECaLS) project. We show that 40-dimensional latent variables can effectively reproduce most morphological features in galaxy images. To further validate the effectiveness of our approach, we utilized a classical random forest classifier on the 40-dimensional latent variables to make detailed morphology feature classifications. This approach performs similar to a direct neural network application on galaxy images. We further enhance our model by tuning the VAE network via DA using galaxies in the overlapping footprint of DECaLS and Beijing-Arizona Sky Survey + Mayall z-band Legacy Survey, enabling the unbiased application of our model to galaxy images in both surveys. We observed that DA led to even better morphological feature extraction and classification performance. Overall, this combination of VAE and DA can be applied to achieve image dimensionality reduction, defect image identification, and morphology classification in large optical surveys.

Neuropathologist-level integrated classification of adult-type diffuse gliomas using deep learning from whole-slide pathological images

Current diagnosis of glioma types requires combining both histological features and molecular characteristics, which is an expensive and time-consuming procedure. Determining the tumor types directly from whole-slide images (WSIs) is of great value for glioma diagnosis. This study presents an integrated diagnosis model for automatic classification of diffuse gliomas from annotation-free standard WSIs. Our model is developed on a training cohort (n = 1362) and a validation cohort (n = 340), and tested on an internal testing cohort (n = 289) and two external cohorts (n = 305 and 328, respectively). The model can learn imaging features containing both pathological morphology and underlying biological clues to achieve the integrated diagnosis. Our model achieves high performance with area under receiver operator curve all above 0.90 in classifying major tumor types, in identifying tumor grades within type, and especially in distinguishing tumor genotypes with shared histological features. This integrated diagnosis model has the potential to be used in clinical scenarios for automated and unbiased classification of adult-type diffuse gliomas.

Unbiased classification of the elderly human brain proteome resolves distinct clinical and pathophysiological subtypes of cognitive impairment

Cognitive impairment in the elderly features complex molecular pathophysiology extending beyond the hallmark pathologies of traditional disease classification. Molecular subtyping using large-scale -omic strategies can help resolve this biological heterogeneity. Using quantitative mass spectrometry, we measured ∼8000 proteins across >600 dorsolateral prefrontal cortex tissues with clinical diagnoses of no cognitive impairment (NCI), mild cognitive impairment (MCI), and Alzheimer's disease (AD) dementia. Unbiased classification of MCI and AD cases based on individual proteomic profiles resolved three classes with expression differences across numerous cell types and biological ontologies. Two classes displayed molecular signatures atypical of AD neurodegeneration, such as elevated synaptic and decreased inflammatory markers. In one class, these atypical proteomic features were associated with clinical and pathological hallmarks of cognitive resilience. We were able to replicate these classes and their clinicopathological phenotypes across two additional tissue cohorts. These results promise to better define the molecular heterogeneity of cognitive impairment and meaningfully impact its diagnostic and therapeutic precision.

Robust and unbiased positive and unlabeled learning

Positive and unlabeled learning (PU learning) is a highly effective classification method that only involves positive and unlabeled samples. It has garnered significant attention in recent years due to its ability to greatly improve classification performance when negative samples are absent. However, numerous contributions in this field remain theoretical and fail to perform well on actual data polluted by noise. How to deal with these challenges is currently a hot research topic. Inspirited by the previous work, we propose a noise-insensitive and unbiased classification method to deal with noise data, i.e., Pinball Loss Factorization and Centroid Smoothing, called Pin-LFCS. For nonlinear situation, we also present an extended kernelized version named Pin-KLFCS. Theoretically, we investigate the property of our proposed methods, including noise insensitivity and intra-class scatter minimization. Moreover, generalization error bounds are derived to ensure their validity. Experimentally, through extensive experiments on 14 benchmark datasets with varying levels of noise, we demonstrate the efficacy and validity of our proposed methods, which outperform the existing advanced methods.

The Management of Hydrocephalus in Midline Posterior Fossa Cystic Collections: Surgical Outcome From a Retrospective Single-Center Case Series of 54 Consecutive Pediatric Patients.

Hydrocephalus frequently occurs with midline posterior fossa cystic collections. The classification of this heterogeneous group of developmental anomalies, including Dandy-Walker malformation, persisting Blake's pouch, retrocerebellar arachnoid cysts, and mega cisterna magna, is subject of debate. The absence of diagnostic criteria is confusing regarding the ideal management of PFCC-related hydrocephalus. To decipher the surgical strategy for the treatment of children with PFCC-related hydrocephalus through a retrospective analysis of the surgical outcome driven by their clinical and radiological presentation. This study enrolled patients operated of symptomatic PFCC-related hydrocephalus. Clinical and MRI features were examined, as well as the surgical outcome. Unbiased subgroup classification of the patients was performed with multiple component analysis as a function of imaging characteristics and hierarchical clustering on principal component. Outcome was assessed with binomial logistic regression and Kaplan-Meier analysis. Fifty-four patients were included between 2007 and 2021. Multiple component analysis suggested that cerebellar and vermian hypoplasia, vermian rotation, basal-tentorial angle, and fastigial angle were strongly correlated. Hierarchical clustering and the distribution of the patients in the bidimensional plot showed the clear segregation of 3 major clusters, which correlated with the radiological diagnosis ( P < .01). Binomial logistic regression and survival analysis showed that endoscopic third ventriculostomy was an effective treatment for patients with persisting Blake's pouch, while failing to control hydrocephalus in most of patients with Dandy-Walker malformation. Preoperative MRI in patients with PFCC-related hydrocephalus is essential to better define the diagnosis. The choice of treatment strategy notably relies on correct radiological diagnosis.

Towards an unbiased shape classification of super-resolution microscopy localizations.

Single molecule localization microscopy (SMLM) captures biological structures at nanoscale spatial resolution and has become an important tool for discovery in cell biology in recent years. However, tools that enable robust quantification of complex biological structures have lagged behind. In this study, we developed a new method to quantitatively describe and classify a wide range of diverse biological structures in SMLM data and have applied it to describe the degradation process of neuronal Tau protein aggregates, an important precursor in neurodegenerative diseases.

Automated Classification of Estuarine Sub‐Depositional Environment Using Sediment Texture

AbstractInterpretation of unconsolidated Quaternary sedimentary core is difficult if key diagnostic features are obscured or not present, therefore traditional facies analysis is challenging. However, sediment texture remains a universal attribute which can be used to interpret sedimentary core. Here we present an automated classification workflow which implements Extreme Gradient Boosting and Bayesian Optimization of hyperparameters to differentiate estuarine sub‐depositional environments. We use 19 textural attributes, measured using laser particle size analysis of surface sediment samples from the Ravenglass Estuary, Cumbria, northwest England, to make unbiased classification of sub‐depositional environment and estuarine zone. Two predictive models created using the automated workflow are presented and evaluated using a suite of evaluation metrics, confusion matrices, and spatial analysis to understand their geological implications. Model 1 keeps all sub‐depositional environments discrete and has an overall accuracy of 68.96%. Model 2 merges related sub‐depositional environments to form inner‐coarse and outer‐estuary zones and has an overall accuracy of 84.14%. Both models have been applied to textural data obtained at 5 cm intervals from a Holocene core drilled through a tidal bar in the Ravenglass estuarine succession, NW England, to classify palaeo sub‐depositional environment. Predictive output of the models suggests that the core consistently experienced inner estuary deposition; all inner estuary environments are represented in the core. The workflow presented here could be applied to datasets from other marginal marine depositional systems to enhance the interpretation of their subsurface deposits. Ultimately, detailed interpretations of ancient, buried deposits could be made using models derived from analogous modern systems.

Molecular taxonomy of nociceptors and pruriceptors.

Molecular taxonomy of nociceptors and pruriceptors.

Early detection of late blight in potato by whole-plant redox imaging.

Late blight caused by the oomycete Phytophthora infestans is a most devastating disease of potatoes (Solanum tuberosum). Its early detection is crucial for suppressing disease spread. Necrotic lesions are normally seen in leaves at 4 days post-inoculation (dpi) when colonized cells are dead, but early detection of the initial biotrophic growth stage, when the pathogen feeds on living cells, is challenging. Here, the biotrophic growth phase of P. infestans was detected by whole-plant redox imaging of potato plants expressing chloroplast-targeted reduction-oxidation sensitivegreen fluorescent protein (chl-roGFP2). Clear spots on potato leaves with a lower chl-roGFP2 oxidation state were detected as early as 2dpi, before any visual symptoms were recorded. These spots were particularly evident during light-to-dark transitions, and reflected the mislocalization of chl-roGFP2 outside the chloroplasts. Image analysis based on machine learning enabled systematic identification and quantification of spots, and unbiased classification of infected and uninfected leaves in inoculated plants. Comparing redox with chlorophyll fluorescence imaging showed that infected leaf areas that exhibit mislocalized chl-roGFP2 also showed reduced non-photochemical quenching and enhanced quantum PSII yield (ΦPSII) compared with the surrounding leaf areas. The data suggest that mislocalization of chloroplast-targeted proteins is an efficient marker of late blight infection, and demonstrate how it can be utilized for non-destructive monitoring of the disease biotrophic stage using whole-plant redox imaging.

Automated Estimation of Crop Yield Using Artificial Intelligence and Remote Sensing Technologies.

Agriculture is the backbone of any country, and plays a viable role in the total gross domestic product (GDP). Healthy and fruitful crops are of immense importance for a government to fulfill the food requirements of its inhabitants. Because of land diversities, weather conditions, geographical locations, defensive measures against diseases, and natural disasters, monitoring crops with human intervention becomes quite challenging. Conventional crop classification and yield estimation methods are ineffective under unfavorable circumstances. This research exploits modern precision agriculture tools for enhanced remote crop yield estimation, and types classification by proposing a fuzzy hybrid ensembled classification and estimation method using remote sensory data. The architecture enhances the pooled images with fuzzy neighborhood spatial filtering, scaling, flipping, shearing, and zooming. The study identifies the optimal weights of the strongest candidate classifiers for the ensembled classification method adopting the bagging strategy. We augmented the imagery datasets to achieve an unbiased classification between different crop types, including jute, maize, rice, sugarcane, and wheat. Further, we considered flaxseed, lentils, rice, sugarcane, and wheat for yield estimation on publicly available datasets provided by the Food and Agriculture Organization (FAO) of the United Nations and the Word Bank DataBank. The ensemble method outperformed the individual classification methods for crop type classification on an average of 13% and 24% compared to the highest gradient boosting and lowest decision tree methods, respectively. Similarly, we observed that the gradient boosting predictor outperformed the multivariate regressor, random forest, and decision tree regressor, with a comparatively lower mean square error value on yield years 2017 to 2021. Further, the proposed architecture supports embedded devices, where remote devices can adopt a lightweight classification algorithm, such as MobilenetV2. This can significantly reduce the processing time and overhead of a large set of pooled images.

Integrated multi-omics profiling yields a clinically relevant molecular classification for esophageal squamous cell carcinoma.

Integrated molecular analysis of human cancer has yielded molecular classification for precise management of cancer patients. Here, we analyzed the whole genomic, epigenomic, transcriptomic, and proteomic data of 155 esophageal squamous cell carcinomas (ESCCs). Multi-omics analysis led to the classification of ESCCs into four subtypes: cell cycle pathway activation, NRF2 oncogenic activation, immune suppression (IS), and immune modulation (IM). IS and IM cases were highly immune infiltrated but differed in the type and distribution of immune cells. IM cases showed better response to immune checkpoint blockade therapy than other subtypes in a clinical trial. We further developed a classifier with 28 features to identify the IM subtype, which predicted anti-PD-1 therapy response with 85.7% sensitivity and 90% specificity. These results emphasize the clinical value of unbiased molecular classification based on multi-omics data and have the potential to further improve the understanding and treatment of ESCC.

A Data-Enhanced High Impedance Fault Detection Method Under Imbalanced Sample Scenarios in Distribution Networks

It is still a huge challenge for data-driven methods to detect high impedance fault (HIF) within limited field fault data. Even more, classification of imbalanced data has encountered a significant degradation in predictive performance since most classifiers were trained with the premise of a relatively balanced distribution. In this paper, a data-driven methodology combined with generative adversarial network (GAN) for HIF detection with imbalanced sample scenarios is proposed. By extending the structure and specializing loss function, the improved GAN (IGAN) can be established to learn the intrinsical probability distribution of raw data and sample new HIF data easily. Adding the high-quality dummy fault data into the original imbalanced training data set, equitable predictive accuracy can be achieved and unbiased classification models are available. Experimental results revealed that the proposed method can generate data subject to target distribution outperformed existing methods. Moreover, the validation results on filed data revealed that the proposed method can detect HIF at around 60 <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ms</i> with fairly higher accuracy on imbalanced samples compared to mainstream data-driven algorithms.

Spatial Transcriptomics of Tissue-Based Immune Response to Acute Graft-Versus-Host Disease of the Lower Gastrointestinal Tract

Background: Corticosteroid resistant acute graft-versus-host disease (GVHD) of the lower gastrointestinal (GI) tract is a leading cause of morbidity and mortality after allogeneic hematopoietic cell transplantation. We hypothesized that distinct gene expression profiles exist within affected GI tissue at GVHD diagnosis and may be associated with treatment outcomes. Methods: Unstained tissue sections from 33 patients with biopsy-confirmed acute lower GI GVHD were used for spatial transcriptomics using the NanoString GeoMxTM Digital Spatial Profiler (DSP) platform. Fluorescent morphology markers (PanCK, CD45 and CD31) were used to visualize the epithelial, immune and endothelial compartments and guide capture of RNA expression profiles using the GeoMxTM Human Whole Transcriptome Atlas (18,676 genes). A total of 739 areas of interest (AOIs) were sampled across the 33 patients with a total of 57,960 immune, 37,610 epithelial, and 2298 vascular endothelial cells analyzed (average of 1756.4 immune, 1139.7 epithelial and 69.6 endothelial cells per case). The gene specific oligos were quantified using Illumina's i5 x i7 dual-indexing system and sequenced on the Illumina NextSeq 500 using a 2 x 75 MidOutput v2.5 flow cell. The FASTQ files were processed to generate target probe-specific count files and saved in DCC format. Gene expression files were normalized per NanoString protocol. All unsupervised clustering and differential expression were done in Bioconductor in the R statistical computing space. Results: The gene expression data for each AOI of immune, epithelial, and endothelial cells were normalized and the top 500 most variable genes were used for unsupervised hierarchical clustering to obtain an unbiased classification of different subtypes. There were two clear major AOI clusters for each of the cellular compartments with immune clustering shown in Figure 1A. The left immune cluster (IC1) and the right immune cluster (IC2) were then evaluated for association with clinical GVHD outcomes. IC1 had an over-representation of patients with steroid sensitive GVHD (Fisher exact p value = <0.0001) and AOIs with low grade pathology (Fisher exact p value = 0.031). To obtain gene level differences between IC1 and IC2, differential expression was performed using a Wilcoxon Rank Sum test and volcano plots were generated to illustrate the most differentially expressed genes across the visual AOI-based clusters (Figure 1B). Genes with an adjusted p-value of 0.05 for association with the particular arm of the volcano plot and at least a 2-fold change over the other cluster were considered significant. Gene set enrichment analysis (GSEA) using curated datasets from MSigDB was performed and identified that IC1 genes were enriched for IFNγ response (q value = 2.99e-5), TNFα signaling (q value = 2.99e-5) and inflammatory response (q value = 0.0063). These IC1 differentially expressed genes included CD74 (835 vs. 361.6 mean normalized gene transcripts), LYZ, β2-microglobulin, C1QB and CD68. IC2, on the other hand, had a prominent exhausted/immunosuppressive signature significantly enriched for Th17 cell genes (q value = 0.027) by GSEA. Notable genes that were statistically significant were FCF1 (152.3 vs. 74 mean normalized transcripts) and USP17L11 (78.3 vs. 36.4 mean normalized transcripts). Extension of this analysis to the epithelial (EC) and vascular endothelial (VC) cell compartments also identified distinct clustering with both having an inflamed (EC1 and VC1) and a basal-like state (EC2 and VC2). Interestingly, the inflamed epithelial cells (EC2) associated with immunosuppressive IC2 immune cells (Fisher exact p-value = 0.006), but inflamed vascular endothelium (VC1) associated with the pro-inflammatory IC1 immune cells (Fisher exact p-value = 0.0005). Conclusion: Spatial transcriptomics provides high resolution gene expression profiles in distinct cellular compartments in human acute GI GVHD samples and provides a better understanding of the relationship of the immune response with epithelial and vascular endothelial cell phenotypes. Initial analysis demonstrates that an inflammatory immune signature is associated with steroid responsive acute GI GVHD and an exhausted phenotype with steroid resistance. Future studies are needed to confirm the predictive value of specific gene expression profiles as biomarkers of corticosteroid response at diagnostic biopsy. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Multiple Instance Learning With Random Forest for Event Logs Analysis and Predictive Maintenance in Ship Electric Propulsion System

In this article, a novel weakly supervised machine learning approach is proposed for intelligent predictive maintenance (IPdM). It employs balanced random forest and multiple instance learning based on event logs from ships’ electric propulsion systems. The objectives are predicting failure likelihood, time to failure, and explainable predictions to ensure timely crew intervention. The IPdM approach uncovers, then learns, and classifies <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">sequences of</i> events that represent early causes or symptoms to forecast imminent failures. In particular, this article contributes effective solutions to irregular, imbalanced, and unlabeled data issues where conventional methods become obsolete. First, the events occur at irregular intervals; they include alarms, warnings, and operational information collected across multiple units and control systems. Second, the datasets exhibit extreme imbalance due to few failures and multiple failure modes; this entails biased predictions. Third, the training datasets are weakly labeled; only the failure timestamp is known without any expert input on prior causes or early symptoms. Temporal random indexing is proposed to transform textual log messages into a numerical lower dimensional space where timeseries analyses are applicable. Balanced random-forest models are developed for unbiased classification and regression. The overall approach learns recursively the ungiven data labels while training the base learners. The IPdM approach is validated through millions of events of multithousand types collected from two years of seagoing vessels. It successfully forecasts actual propulsion failures and performs better when compared with contemporary methods.