Multinomial Mixture Research Articles

Lung allograft dysbiosis associates with immune response and primary graft dysfunction

RationaleLower airway enrichment with oral commensals has been previously associated with grade 3 severe primary graft dysfunction (PGD) after lung transplantation (LT). We aimed to determine whether this dysbiotic signature is present across all PGD severity grades, including milder forms, and whether it is associated with a distinct host inflammatory endotype. MethodsLower airway samples from 96 LT recipients with varying degrees of PGD were used to evaluate the lung allograft microbiota via 16S rRNA gene sequencing. Bronchoalveolar lavage (BAL) cytokine concentrations and cell differential percentages were compared across PGD grades. In a subset of samples, we evaluated the lower airway host transcriptome using RNA sequencing methods. ResultsDifferential analyses demonstrated lower airway enrichment with supraglottic-predominant taxa (SPT) in both moderate and severe PGD. Dirichlet Multinomial Mixtures (DMM) modeling identified two distinct microbial clusters. A greater percentage of subjects with moderate-severe PGD were identified within the dysbiotic cluster (C-SPT) than within the no PGD group (48 and 29%, respectively) though this difference did not reach statistical significance (p=0.06). PGD severity associated with increased BAL neutrophil concentration (p=0.03) and correlated with BAL concentrations of MCP-1/CCL2, IP-10/CXCL10, IL-10, and TNF-α (p<0.05). Furthermore, microbial signatures of dysbiosis correlated with neutrophils, MCP-1/CCL-2, IL-10, and TNF-α (p<0.05). C-SPT exhibited differential expression of TNF, SERPINE1 (PAI-1), MPO, and MMP1 genes and upregulation of MAPK pathways, suggesting that dysbiosis regulates host signaling to promote neutrophilic inflammation. ConclusionsLower airway dysbiosis within the lung allograft is associated with a neutrophilic inflammatory endotype, an immune profile commonly recognized as the hallmark for PGD pathogenesis. This data highlights a putative role for lower airway microbial dysbiosis in the pathogenesis of this syndrome.

Dynamics of Gut Microbiota after Fecal Microbiota Transplantation in Ulcerative Colitis: Success Linked to Control of Prevotellaceae.

Fecal microbiota transplantation (FMT) is an experimental treatment for ulcerative colitis (UC). We aimed to study microbial families associated with FMT treatment success. We analyzed stools from 24 UC patients treated with four FMTs weekly after randomization for pretreatment during three weeks with budesonide (n = 12) or placebo (n = 12). Stool samples were collected nine times pre-, during, and post FMT. Clinical and endoscopic response was assessed 14 weeks after initiation of the study using the full Mayo score. Early withdrawal due to worsening of UC symptoms was classified as non-response. Nine patients (38%) reached remission at week 14, and 15 patients had a partial response or non-response at or before week 14. With a Dirichlet Multinomial Mixture model we identified five distinct clusters based on the microbiota composition of 180 longitudinally collected patient samples and 27 donor samples. A Prevotellaceae-dominant cluster was associated with poor response to FMT treatment. Conversely, the families Ruminococcaceae and Lachnospiraceae were associated with a successful clinical response. These associations were already visible at the start of the treatment for a subgroup of patients and were retained in repeated measures analyses of family-specific abundance over time. Responders were also characterized by a significantly lower Simpson dominance compared to non-responders. The success of FMT treatment of UC patients appears to be associated with specific gut microbiota families, such as control of Prevotellaceae. Monitoring the dynamics of these microbial families could potentially be used to inform treatment success early during FMT.

Reproductive Tract Microbial Transitions from Late Gestation to Early Postpartum Using 16S rRNA Metagenetic Profiling in First-Pregnancy Heifers.

Studies in recent years indicate that reproductive tract microbial communities are crucial for shaping mammals' health and reproductive outcomes. Following parturition, uterine bacterial contamination often occurs due to the open cervix, which may lead to postpartum uterine inflammatory diseases, especially in primiparous individuals. However, investigations into spatio-temporal microbial transitions in the reproductive tract of primigravid females remain limited. Our objective was to describe and compare the microbial community compositions in the vagina at late gestation and in the vagina and uterus at early postpartum in first-pregnancy heifers. Three swab samples were collected from 33 first-pregnancy Holstein Friesian heifers: one vaginal sample at gestation day 258 ± 4, and vaginal and uterine samples at postpartum day 7 ± 2. Each sample underwent 16S rRNA V4 region metagenetic analysis via Illumina MiSeq, with bioinformatics following Mothur MiSeq SOP. The reproductive tract bacterial communities were assigned to 1255 genus-level OTUs across 30 phyla. Dominant phyla, accounting for approximately 90% of the communities, included Proteobacteria, Firmicutes, Actinobacteria, Bacteroidetes, and Fusobacteria. However, the results revealed distinct shifts in microbial composition between the prepartum vagina (Vag-pre), postpartum vagina (Vag-post), and postpartum uterus (Utr-post). The Vag-pre and Utr-post microbial profiles were the most distinct. The Utr-post group had lower relative abundances of Proteobacteria but higher abundances of Bacteroidetes, Fusobacteria, and Tenericutes compared to Vag-pre, while Vag-post displayed intermediate values for these phyla, suggesting a transitional profile. Additionally, the Utr-post group exhibited lower bacterial richness and diversity compared to both Vag-pre and Vag-post. The unsupervised probabilistic Dirichlet Multinomial Mixtures model identified two distinct community types: most Vag-pre samples clustered into one type and Utr-post samples into another, while Vag-post samples were distributed evenly between the two. LEfSe analysis revealed distinct microbial profiles at the genus level. Overall, specific microbial markers were associated with anatomical and temporal transitions, revealing a dynamic microbial landscape during the first pregnancy and parturition. These differences highlight the complexity of these ecosystems and open new avenues for research in reproductive biology and microbial ecology.

Population stratification based on healthcare trajectories: A method for encouraging adaptive learning at meso level

This paper proposes a method to support population management by evaluating population needs using population stratification based on healthcare trajectories.Reimbursed healthcare consumption data for the first semester of 2017 contained within the inter-mutualist database were analysed to create healthcare trajectories for a subset of the population aged between 60 and 79 (N = 22,832) to identify (1) the nature of health events, (2) key transitions between lines of care, (3) the relative duration of different events, and (4) the hierarchy of events. These factors were classified using a K-mers approach followed by multinomial mixture modelling.Five population groups were identified using this healthcare trajectory approach: “low users”, “high intensity of nursing care”, “transitional care & nursing care”, “transitional care”, and “long time in hospital”.This method could be used by loco-regional governing bodies to learn reflectively from the place where care is provided, taking a systems perspective rather than a disease perspective, and avoiding the one-size-fits-all definition. It invites decision makers to make better use of routinely collected data to guide continuous learning and adaptive management of population health needs.

Using machine learning to identify environmental factors that collectively determine microbial community structure of activated sludge

Activated sludge (AS) microbial communities are influenced by various environmental variables. However, a comprehensive analysis of how these variables jointly and nonlinearly shape the AS microbial community remains challenging. In this study, we employed advanced machine learning techniques to elucidate the collective effects of environmental variables on the structure and function of AS microbial communities. Applying Dirichlet multinomial mixtures analysis to 311 global AS samples, we identified four distinct microbial community types (AS-types), each characterized by unique microbial compositions and metabolic profiles. We used 14 classical linear and nonlinear machine learning methods to select a baseline model. The extremely randomized trees demonstrated optimal performance in learning the relationship between environmental factors and AS types (with an accuracy of 71.43%). Feature selection identified critical environmental factors and their importance rankings, including latitude (Lat), longitude (Long), precipitation during sampling (Precip), solids retention time (SRT), effluent total nitrogen (Effluent TN), average temperature during sampling month (Avg Temp), mixed liquor temperature (Mixed Temp), influent biochemical oxygen demand (Influent BOD), and annual precipitation (Annual Precip). Significantly, Lat, Long, Precip, Avg Temp, and Annual Precip, influenced metabolic variations among AS types. These findings emphasize the pivotal role of environmental variables in shaping microbial community structures and enhancing metabolic pathways within activated sludge. Our study encourages the application of machine learning techniques to design artificial activated sludge microbial communities for specific environmental purposes.

Mudskipper detects combinatorial RNA binding protein interactions in multiplexed CLIP data

The uncovering of protein-RNA interactions enables a deeper understanding of RNA processing. Recent multiplexed crosslinking and immunoprecipitation (CLIP) technologies such as antibody-barcoded eCLIP (ABC) dramatically increase the throughput of mapping RNA binding protein (RBP) binding sites. However, multiplex CLIP datasets are multivariate, and each RBP suffers non-uniform signal-to-noise ratio. To address this, we developed Mudskipper, a versatile computational suite comprising two components: a Dirichlet multinomial mixture model to account for the multivariate nature of ABC datasets and a softmasking approach that identifies and removes non-specific protein-RNA interactions in RBPs with low signal-to-noise ratio. Mudskipper demonstrates superior precision and recall over existing tools on multiplex datasets and supports analysis of repetitive elements and small non-coding RNAs. Our findings unravel splicing outcomes and variant-associated disruptions, enabling higher-throughput investigations into diseases and regulation mediated by RBPs.

Dirichlet compound negative multinomial mixture models and applications

Dirichlet compound negative multinomial mixture models and applications

Hybrid topic modeling method based on dirichlet multinomial mixture and fuzzy match algorithm for short text clustering

Topic modeling methods proved to be effective for inferring latent topics from short texts. Dealing with short texts is challenging yet helpful for many real-world applications, due to the sparse terms in the text and the high dimensionality representation. Most of the topic modeling methods require the number of topics to be defined earlier. Similarly, methods based on Dirichlet Multinomial Mixture (DMM) involve the maximum possible number of topics before execution which is hard to determine due to topic uncertainty, and many noises exist in the dataset. Hence, a new approach called the Topic Clustering algorithm based on Levenshtein Distance (TCLD) is introduced in this paper, TCLD combines DMM models and the Fuzzy matching algorithm to address two key challenges in topic modeling: (a) The outlier problem in topic modeling methods. (b) The problem of determining the optimal number of topics. TCLD uses the initial clustered topics generated by DMM models and then evaluates the semantic relationships between documents using Levenshtein Distance. Subsequently, it determines whether to keep the document in the same cluster, relocate it to another cluster, or mark it as an outlier. The results demonstrate the efficiency of the proposed approach across six English benchmark datasets, in comparison to seven topic modeling approaches, with 83% improvement in purity and 67% enhancement in Normalized Mutual Information (NMI) across all datasets. The proposed method was also applied to a collected Arabic tweet and the results showed that only 12% of the Arabic short texts were incorrectly clustered, according to human inspection.

Impact of COVID-19 Pandemic on Social Determinants of Health Issues of Marginalized Black and Asian Communities: A Social Media Analysis Empowered by Natural Language Processing.

This study aims to understand the impact of the COVID-19 pandemic on social determinants of health (SDOH) of marginalized racial/ethnic US population groups, specifically African Americans and Asians, by leveraging natural language processing (NLP) and machine learning (ML) techniques on race-related spatiotemporal social media text data. Specifically, this study establishes the extent to which Latent Dirichlet Allocation (LDA) and Gibbs Sampling Dirichlet Multinomial Mixture (GSDMM)-based topic modeling determines social determinants of health (SDOH) categories, and how adequately custom named-entity recognition (NER) detects key SDOH factors from a race/ethnicity-related Reddit data corpus. In this study, we collected race/ethnicity-specific data from 5 location subreddits including New York City, NY; Los Angeles, CA; Chicago, IL; Philadelphia, PA; and Houston, TX from March to December 2019 (before COVID-19 pandemic) and from March to December 2020 (during COVID-19 pandemic). Next, we applied methods from natural language processing and machine learning to analyze SDOH issues from extracted Reddit comments and conversation threads using feature engineering, topic modeling, and custom named-entity recognition (NER). Topic modeling identified 35 SDOH-related topics. The SDOH-based custom NER analyses revealed that the COVID-19 pandemic significantly impacted SDOH issues of marginalized Black and Asian communities. On average, the Social and Community Context (SCC) category of SDOH had the highest percent increase (366%) from the pre-pandemic period to the pandemic period across all locations and population groups. Some of the detected SCC issues were racism, protests, arrests, immigration, police brutality, hate crime, white supremacy, and discrimination. Reddit social media platform can be an alternative source to assess the SDOH issues of marginalized Black and Asian communities during the COVID-19 pandemic. By employing NLP/ML techniques such as LDA/GSDMM-based topic modeling and custom NER on a race/ethnicity-specific Reddit corpus, we uncovered various SDOH issues affecting marginalized Black and Asian communities that were significantly worsened during the COVID-19 pandemic. As a result of conducting this research, we recommend that researchers, healthcare providers, and governments utilize social media and collaboratively formulate responses and policies that will address SDOH issues during public health crises.

Decoding mood of the Twitterverse on ESG investing: opinion mining and key themes using machine learning

PurposeGrounded in the stakeholder theory and signaling theory, this study aims to broaden the research agenda on environmental, social and governance (ESG) investing by uncovering public sentiments and key themes using Twitter data spanning from 2009 to 2022.Design/methodology/approachUsing various machine learning models for text tonality analysis and topic modeling, this research scrutinizes 1,842,985 Twitter texts to extract prevalent ESG investing trends and gauge their sentiment.FindingsGibbs Sampling Dirichlet Multinomial Mixture emerges as the optimal topic modeling method, unveiling significant topics such as “Physical risk of climate change,” “Employee Health, Safety and well-being” and “Water management and Scarcity.” RoBERTa, an attention-based model, outperforms other machine learning models in sentiment analysis, revealing a predominantly positive shift in public sentiment toward ESG investing over the past five years.Research limitations/implicationsThis study establishes a framework for sentiment analysis and topic modeling on alternative data, offering a foundation for future research. Prospective studies can enhance insights by incorporating data from additional social media platforms like LinkedIn and Facebook.Practical implicationsLeveraging unstructured data on ESG from platforms like Twitter provides a novel avenue to capture company-related information, supplementing traditional self-reported sustainability disclosures. This approach opens new possibilities for understanding a company’s ESG standing.Social implicationsBy shedding light on public perceptions of ESG investing, this research uncovers influential factors that often elude traditional corporate reporting. The findings empower both investors and the general public, aiding managers in refining ESG and management strategies.Originality/valueThis study marks a groundbreaking contribution to scholarly exploration, to the best of the authors’ knowledge, by being the first to analyze unstructured Twitter data in the context of ESG investing, offering unique insights and advancing the understanding of this emerging field.

Abstract 2803: Identification of microbiome and predictive immune biomarkers in patients with advanced non-small lung cancer treated with pembrolizumab with or without chemotherapy

Abstract Background: Recent studies have shown that gut microbiota modulates antitumor immunity, and affect clinical efficacy of immune checkpoint inhibitors, including anti-programmed death-1 (PD-1) and programmed death-ligand 1 (PD-L1) inhibitors. We aimed to explore the metagenomic profiles of previously untreated, treatment naïve, stage IV non-small cell lung cancer (NSCLC) treated with pembrolizumab, an anti-PD1 agent with or without platinum-based chemotherapy and identify distinct ecological niche between responder and non-responder groups. Methods: We performed metagenome sequencing, whole exome sequencing (WES), and RNA sequencing (RNA-seq) on 50 patients with stage IV NSCLC, including adenocarcinoma (N=31) and squamous cell carcinoma (N=22). We employed Dirichlet Multinomial Mixtures (DMM) community typing to cluster microbial community profiling data at the genus level and analyzed the changes in microbial composition and diversity between enterotype 1 (E1, N=27) and enterotype 2 (E2, N=23). Mutation landscape including genomic alterations in relation to immune response was analyzed between responders and non-responders using WES data. Differentially expressed genes (DEGs), Gene Set Enrichment Analysis (GSEA), and immune deconvolution analysis were carried out using RNA-seq data to analyze molecular profiling of two enterotypes between the responder (N=15) and non-responder (N=35) groups. Results: The two enterotypes exhibited distinct microbial abundances and alpha diversity indices. Abundances of members of Bacteroidaceae (P=0.011) and Oscillospiraceae (P&amp;lt;0.001) were higher in E2 members, and enrichment of Lactobacillaceae (P=0.006) and Bifidobacteriaceae (P=0.041) were higher in E1. Compared to E1, the number of observed bacterial species (P&amp;lt;0.001) and Shannon index (P&amp;lt;0.001) was significantly higher in E2, indicating a healthier gut environment. The proportion of responders was higher in E2 than in E1 cluster. Comparison between responders and non-responder groups within each enterotype revealed that Lactobacillaceae showed an association with the response in E2 cluster. Mutation landscape revealed TP53 mutation as the most common mutation, more frequently observed in the E2 (100%) than in the E1 (56%) cluster. Conclusion: We explored the association between the gut microbiome and clinical response to pembrolizumab with or without combination with chemotherapy and found that E2 cluster enriched with Lactobacillaceae is associated with treatment response. Further analysis with larger cohort is ongoing. Citation Format: Jii Bum (Joy) Lee, Su-Jin Choi, Gang-Taik Lee, Junwon Yang, Hyun-Seok Oh, Arim Min, Bo-Eun Kwon, Seong-san Kang, Sujeong Baek, Dong Kwon Kim, Jae Hwan Kim, Sun Min Lim, Min Hee Hong, Kyoung-Ho Pyo, Byoung Chul Cho. Identification of microbiome and predictive immune biomarkers in patients with advanced non-small lung cancer treated with pembrolizumab with or without chemotherapy [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 2803.

Abstract 757: Integrating microbiome analysis and multi-modal data to identify high-risk population for esophageal adenocarcinoma

Abstract INTRODUCTION: Esophageal Adenocarcinoma (EAC) represents a pressing public health concern with a rising incidence and a staggering 5-year survival rate of less than 20%, emphasizing the critical need for early detection. Barrett's Esophagus (BE) serves as a well-established precursor to EAC, presenting an opportunity for proactive screening. The salivary microbiome provides a novel opportunity to identify a high-risk population with early-stage esophageal diseases. Here, we assessed whether the oral microbiome, in combination with lifestyle and clinical data, is associated with esophageal disease progression. METHODS: The study included n=23 esophageal adenocarcinoma patients, n=51 with Barrett's Esophagus, and n=32 healthy controls. Saliva samples were collected along with comprehensive lifestyle, dietary intake, and clinical data. Shotgun Metagenomics Sequencing was employed to investigate the microbiome composition of participants. Comprehensive microbiome analyses were performed to evaluate taxonomic profiles, including microbial abundance, microbiome diversity, and functional/metabolic insights. A multi-modal data integration strategy was used to incorporate data from the Diet Health Questionnaire (DHQ) and lifestyle and health records, aiming to utilize these multimodal features to predict esophageal disease formation and progression. RESULTS: A significant difference was observed in the salivary microbiome beta diversity between healthy controls and esophagus disease groups (p= 0.03). Clustering by microbiome taxonomic composition using the Dirichlet Multinomial Mixtures (DMM) framework indicates that some microbial signatures are associated with esophageal diseases (p=0.03). Healthy Eating Index Scores, total score (p=0.01); vegetables (p=0.003); greens, and beans score (p=0.009) were significantly different between study groups. Microbial pathways associated with diet, microbial taxa linked to lifestyle and medications, and supervised classification results reveal group characteristics and diversity variations based on key covariates such as B12 and sugar uptakes. CONCLUSION: The study presents an innovative analytic approach to identify high-risk populations for esophageal diseases by integrating microbiome analysis and multi-modal data, providing insights into early detection and prevention. The potential of this research lies in its ability to revolutionize screening strategies, offering a non-invasive and cost-effective means to identify individuals at risk of EAC at earlier, more treatable stages. Citation Format: Duygu Altinok Dindar, Madeline Krieger, Amy Palma, John Cheney, Fouad Otaki, Jessica Yu, Stephanie Wood, Thomas J. Sharpton, James McGann, Lisa Karstens, Gurkan G. Yardimci, Zhenzhen Zhang. Integrating microbiome analysis and multi-modal data to identify high-risk population for esophageal adenocarcinoma [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 757.

Integrating semantic similarity with Dirichlet multinomial mixture model for enhanced web service clustering

Integrating semantic similarity with Dirichlet multinomial mixture model for enhanced web service clustering

Urinary microbiome community types associated with urinary incontinence severity in women

BackgroundUrinary microbiome (urobiome) studies have previously reported on specific taxa and community differences in women with mixed urinary incontinence compared to controls. Therefore, we hypothesized that higher urinary and vaginal microbiome diversity would be associated with increased urinary incontinence severity. ObjectivesTo test if specific urinary or vaginal microbiome community-types are associated with urinary incontinence severity in a population of women with mixed urinary incontinence. Study DesignThis planned secondary, cross-sectional analysis evaluated associations between the urinary/vaginal microbiomes and urinary incontinence severity in a subset of ESTEEM trial participants with urinary incontinence. Incontinence severity was measured using bladder diaries and Urinary Distress Inventory (UDI) questionnaires collected at baseline. Catheterized urine samples and vaginal swabs were concurrently collected prior to treatment at baseline to assess the urinary and vaginal microbiomes.16S rRNA V4-V6 variable regions were sequenced, characterizing bacterial taxa to the genus level using the DADA2 pipeline and SILVA database. Using Dirichlet multinomial mixture methods, samples were clustered into community-types based on core taxa. Associations between community-types and severity measures (UDI total, UDI subscale scores, and the number of urinary incontinence episodes (total, urgency, and stress) from bladder diary) were evaluated using linear regression models adjusted for age and body mass index. Alpha diversity measures for richness (total taxa numbers) and evenness (proportional distribution of taxa abundance) were also analyzed for associations with urinary incontinence episodes and community-type. ResultsSix urinary microbiome community-types were identified, characterized by varying levels of common genera (Lactobacillus, Gardnerella, Prevotella, Tepidomonas, Acidovorax, Escherichia, and others). Analysis of urinary incontinence severity in 126 participants with MUI identified a Lactobacillus-dominated reference group with the highest abundance of Lactobacillus (76% mean relative abundance). A community characterized by fewer Lactobacilli (19% mean relative abundance) and greater alpha diversity, was associated with higher total urinary incontinence episodes (2.67 daily leaks, 95% CI: 0.76 - 4.59, p=0.007) and urgency urinary incontinence episodes (1.75 daily leaks, 95% CI: 0.24 - 3.27, p=0.02) compared to the reference group. No significant association was observed between community-type and stress urinary incontinence episodes or UDI total or subscores. The composition of vaginal community-types and urinary community-types were similar but comprised of slightly different bacterial taxa. Vaginal community-types were not associated with urinary incontinence severity, as measured by bladder diary or UDI total and subscale scores. Alpha diversity indicated that greater sample richness was associated with more incontinence episodes (observed genera p=0.01) in urine. Measures of evenness (Shannon/Pielou) were not associated with incontinence severity in the urinary or vaginal microbiomes. ConclusionsIn the urobiome of women with mixed urinary incontinence, a community-type with fewer Lactobacilli and more diverse bacteria was associated with more severe urinary incontinence episodes (total and urgency) compared to a community-type with high predominance of a single genus, Lactobacillus. Whether mixed urinary incontinence severity is due to lesser predominance of Lactobacillus, greater presence of other non-Lactobacillus genera, or the complement of bacteria comprising urobiome community-types, remains to be determined.

Exploring topic modelling for generalising design requirements in complex design

As the redesign process progresses in product lifecycle management, effectively managing engineering changes becomes increasingly challenging, often leading to catastrophic and costly project failures. In response, the study provides a framework for generalising design requirements documents into topics that engineers can use to understand complex designs. Based on previous work, this study employs and compares four different models, including latent Dirichlet allocation (LDA), the collapsed Gibbs sampling algorithm for the Dirichlet multinomial mixtures model (GSDMM), LDA-BERT, and GSDMM-BERT to determine the appropriate representation of requirements documents. Both heatmaps and UMAPs are used to illustrate the correlation between topics and words. The results indicate that the combined vector representation of topic modelling and the sentence-BERT model outperforms single topic modelling. This combined model leverages the additional knowledge from a pre-trained sentence-BERT model, thereby improving model performance and word distribution in all three industrial projects. Through this proposed framework, engineers can potentially generalise high-quality requirements topics for large requirements documents.

Categorising the world into local climate zones: towards quantifying labelling uncertainty for machine learning models

Abstract Image classification is often prone to labelling uncertainty. To generate suitable training data, images are labelled according to evaluations of human experts. This can result in ambiguities, which will affect subsequent models. In this work, we aim to model the labelling uncertainty in the context of remote sensing and the classification of satellite images. We construct a multinomial mixture model given the evaluations of multiple experts. This is based on the assumption that there is no ambiguity of the image class, but apparently in the experts’ opinion about it. The model parameters can be estimated by a stochastic expectation maximisation algorithm. Analysing the estimates gives insights into sources of label uncertainty. Here, we focus on the general class ambiguity, the heterogeneity of experts, and the origin city of the images. The results are relevant for all machine learning applications where image classification is pursued and labelling is subject to humans.

Meta-analysis reveals the predictable dynamic development of the gut microbiota in commercial pigs.

Understanding the mechanisms of microbiome assembly during host development is crucial for successful modulation of the gut microbiome to improve host health and growth. However, results from previous microbial intervention studies to improve swine growth have largely been inconsistent due to the constantly changing nature of the gut microbiota and limited longitudinal sampling. Detailed characterization of the swine gut microbiota through meta-analysis allows us to understand the dynamics of microbial community succession, as well as the transient and natural variations between time points and animals. A total of 3,313 fecal microbial communities from over 349 pigs covering 60 time points (from birth to market age) from 14 publications were included in this meta-analysis. Despite differences in animal breeds and management, generalizable patterns of community assembly were identified. Alpha diversity continuously increased during early stages of animal growth and more slowly in the later stages. Beta regression analysis revealed that more microbial taxa were enriched, while fewer were excluded by the gut microbiota. The microbial community structure also changed significantly between days at early ages and became more similar as pigs aged, as revealed by dissimilarity and distance metrics. Dirichlet multinomial mixtures analysis supported gradient microbial clustering in longitudinal pig fecal samples, and we found that the early samples spread to more clusters than those from older pigs. Random forest regression identified 30 operational taxonomic units as potential microbial biomarkers for modeling swine gut microbiota development. External validation showed this model could be generalized to future microbiome studies conducted in suckling and weaning pigs and adds to the toolkit to quantify community succession. IMPORTANCE The swine gut microbiome undergoes an age-dependent assembly pattern with a developmental phase at early ages and a stabilization phase at later ages. Shorter time intervals and a wider range of data sources provided a clearer understanding of the gut microbiota colonization and succession and their associations with pig growth and development. The rapidly changing microbiota of suckling and weaning pigs implies potential time targets for growth and health regulation through gut microbiota manipulation. Since swine gut microbiota development is predictable, swine microbiota age can be calculated and compared between animal treatment groups rather than relying only on static time-matched comparisons.

An algorithmic approach to identification of gray areas: Analysis of sleep scoring expert ensemble non agreement areas using a multinomial mixture model

Machine learning (ML) models have become a key component in modern world services. In decision-making domains where human expertise is crucial, for example, for manually scoring biological signal data, human uncertainties undermine experts’ trust in the outcomes of these models. The field of sleep staging in particular, which requires experts to score complex biological signal is notably impacted by scoring uncertainties. Data consisting of an ensemble of independent scorers are collected to estimate inter scorer agreement and the uncertainty associated with manual scoring. However, scorers’ uncertainty lacks statistical modeling, which poses difficulties in validating ML algorithms and leads to issues of reliability and explainability. From the ensemble of scorers, uncertainty zones, called gray areas, are highlighted by samples where scorers disagree. The objective of our work is to provide a framework introducing and inferring gray areas. We present a flexible and easy-to-use multi-objective method based on multinomial mixture models clustering the different levels of scorer agreement and summarize the results into two sets of high-agreement and gray area clusters, which are called supra-clusters. The threshold is selected according to the maximization of the distance between two distributions of scorers agreement measure. Effective results were obtained by the method after it was fitted on simulated data. Additionally, the method was applied to a real case of uncertainty analysis in the sleep staging domain. A series of actual sleep stages scored by an ensemble of 10 independent scorers for a dataset of 50 participants was used.

Online short text clustering using infinite extensions of discrete mixture models

AbstractShort text clustering is one of the fundamental tasks in natural language processing. Different from traditional documents, short texts are ambiguous and sparse due to their short form and the lack of recurrence in word usage from one text to another, making it very challenging to apply conventional machine learning algorithms directly. In this article, we propose two novel approaches for short texts clustering: collapsed Gibbs sampling infinite generalized Dirichlet multinomial mixture model infinite GSGDMM) and collapsed Gibbs sampling infinite Beta‐Liouville multinomial mixture model (infinite GSBLMM). We adopt two flexible and practical priors to the multinomial distribution where in the first one the generalized Dirichlet distribution is integrated, while the second one is based on the Beta‐Liouville distribution. We evaluate the proposed approaches on two famous benchmark datasets, namely, Google News and Tweet. The experimental results demonstrate the effectiveness of our models compared to basic approaches that use Dirichlet priors. We further propose to improve the performance of our methods with an online clustering procedure. We also evaluate the performance of our methods for the outlier detection task, in which we achieve accurate results.

Topic-Clustering Model with Temporal Distribution for Public Opinion Topic Analysis of Geospatial Social Media Data

Analysis of the spatiotemporal distribution of online public opinion topics can help understand the hotspots of public concern. The topic model is employed widely in public opinion topic clustering for social media data. In order to handle topic-clustering of low-quality geospatial social media data, such as microblog data, with short text and timeliness characteristics, this study proposed a Dirichlet multinomial mixture over time (DMMOT) model to cluster microblog topic for public opinion analysis. The DMMOT model assumes that a single document belongs to a single topic, in line with the characteristics of a short text, and it introduces the probability distribution of “topic-time” in the process of topic generation. The model parameter inference process was presented in detail by exploring the Gibbs sampling method. Results generated using the DMMOT model in case study show that the “topic-word” distribution is semantically aggregated within various topics, and “topic-time” distribution clustered within a time window under each topic. Furthermore, the characteristics of the trend of each topic over time are basically consistent with the corresponding trend of topic in reality in terms of content. These indicate that the DMMOT model improves topic clustering for short text to some extent. Furthermore, the DMMOT model performed well in both temporal and spatial analysis of public opinion topics based on microblog data.