Clustering algorithms: A comparative approach

Human-supervised clustering of multidimensional data using crowdsourcing

Clustering is a central task in many data analysis applications. However, there is no universally accepted metric to decide the occurrence of clusters. Ultimately, we have to resort to a consensus between experts. The problem is amplified with high-dimensional datasets where classical distances become uninformative and the ability of humans to fully apprehend the distribution of the data is challenged. In this paper, we design a mobile human-computing game as a tool to query human perception for the multidimensional data clustering problem. We propose two clustering algorithms that partially or entirely rely on aggregated human answers and report the results of two experiments conducted on synthetic and real-world datasets. We show that our methods perform on par or better than the most popular automated clustering algorithms. Our results suggest that hybrid systems leveraging annotations of partial datasets collected through crowdsourcing platforms can be an efficient strategy to capture the collective wisdom for solving abstract computational problems.

A machine learning and clustering-based approach for county-level COVID-19 analysis

COVID-19 is a global pandemic threatening the lives and livelihood of millions of people across the world. Due to its novelty and quick spread, scientists have had difficulty in creating accurate forecasts for this disease. In part, this is due to variation in human behavior and environmental factors that impact disease propagation. This is especially true for regionally specific predictive models due to either limited case histories or other unique factors characterizing the region. This paper employs both supervised and unsupervised methods to identify the critical county-level demographic, mobility, weather, medical capacity, and health related county-level factors for studying COVID-19 propagation prior to the widespread availability of a vaccine. We use this feature subspace to aggregate counties into meaningful clusters to support more refined disease analysis efforts.

Emotion Classification from Speech and Text in Videos Using a Multimodal Approach

Emotion classification is a research area in which there has been very intensive literature production concerning natural language processing, multimedia data, semantic knowledge discovery, social network mining, and text and multimedia data mining. This paper addresses the issue of emotion classification and proposes a method for classifying the emotions expressed in multimodal data extracted from videos. The proposed method models multimodal data as a sequence of features extracted from facial expressions, speech, gestures, and text, using a linguistic approach. Each sequence of multimodal data is correctly associated with the emotion by a method that models each emotion using a hidden Markov model. The trained model is evaluated on samples of multimodal sentences associated with seven basic emotions. The experimental results demonstrate a good classification rate for emotions.

An analysis framework for clustering algorithm selection with applications to spectroscopy

Cluster analysis is a valuable unsupervised machine learning technique that is applied in a multitude of domains to identify similarities or clusters in unlabelled data. However, its performance is dependent of the characteristics of the data it is being applied to. There is no universally best clustering algorithm, and hence, there are numerous clustering algorithms available with different performance characteristics. This raises the problem of how to select an appropriate clustering algorithm for the given analytical purposes. We present and validate an analysis framework to address this problem. Unlike most current literature which focuses on characterizing the clustering algorithm itself, we present a wider holistic approach, with a focus on the user’s needs, the data’s characteristics and the characteristics of the clusters it may contain. In our analysis framework, we utilize a softer qualitative approach to identify appropriate characteristics for consideration when matching clustering algorithms to the intended application. These are used to generate a small subset of suitable clustering algorithms whose performance are then evaluated utilizing quantitative cluster validity indices. To validate our analysis framework for selecting clustering algorithms, we applied it to four different types of datasets: three datasets of homemade explosives spectroscopy, eight datasets of publicly available spectroscopy data covering food and biomedical applications, a gene expression cancer dataset, and three classic machine learning datasets. Each data type has discernible differences in the composition of the data and the context within which they are used. Our analysis framework, when applied to each of these challenges, recommended differing subsets of clustering algorithms for final quantitative performance evaluation. For each application, the recommended clustering algorithms were confirmed to contain the top performing algorithms through quantitative performance indices.

Role of Nuclear-Receptor-Related 1 in the Synergistic Neuroprotective Effect of Umbilical Cord Blood and Erythropoietin Combination Therapy in Hypoxic Ischemic Encephalopathy

Neonatal hypoxic–ischemic encephalopathy (HIE) results in neurological impairments; cell-based therapy has been suggested as a therapeutic avenue. Previous research has demonstrated the synergistically potentiated therapeutic efficacy of human umbilical cord blood (UCB) by combining recombinant human erythropoietin (EPO) treatment for recovery from HIE. However, its molecular mechanism is not entirely understood. In the present study, we analyzed the mechanisms underlying the effect of combination treatment with EPO and UCB by transcriptomic analysis, followed by gene enrichment analysis. Mouse HIE model of the neonate was prepared and randomly divided into five groups: sham, HIE, and UCB, EPO, and UCB+EPO treatments after HIE. A total of 376 genes were differentially expressed when |log2FC| ≥ 1-fold change expression values were considered to be differentially expressed between UCB+EPO and HIE. Further assessment through qRT-PCR and gene enrichment analysis confirmed the expression and correlation of its potential target, Nurr1, as an essential gene involved in the synergistic effect of the UCB+EPO combination. The results indicated the remarkable activation of Wnt/β-catenin signaling by reducing the infarct size by UCB+EPO treatment, accompanied by Nurr1 activity. In conclusion, these findings suggest that the regulation of Nurr1 through the Wnt/β-catenin pathway exerts a synergistic neuroprotective effect in UCB and EPO combination treatment.

Accelerating Energy-Economic Simulation Models via Machine Learning-Based Emulation and Time Series Aggregation

Energy-economic simulation models with high levels of detail, high time resolutions, or large populations (e.g., distribution networks, households, electric vehicles, energy communities) are often limited due to their computational complexity. This paper introduces a novel methodology, combining cluster-based time series aggregation and sampling methods, to efficiently emulate simulation models using machine learning and significantly reduce both simulation and training time. Machine learning-based emulation models require sufficient and high-quality data to generalize the dataset. Since simulations are computationally complex, their maximum number is limited. Sampling methods come into play when selecting the best parameters for a limited number of simulations ex ante. This paper introduces and compares multiple sampling methods on three energy-economic datasets and shows their advantage over a simple random sampling for small sample-sizes. The results show that a k-means cluster sampling approach (based on unsupervised learning) and adaptive sampling (based on supervised learning) achieve the best results especially for small sample sizes. While a k-means cluster sampling is simple to implement, it is challenging to increase the sample sizes if the emulation model does not achieve sufficient accuracy. The iterative adaptive sampling is more complex during implementation, but can be re-applied until a certain accuracy threshold is met. Emulation is then applied on a case study, emulating an energy-economic simulation framework for peer-to-peer pricing models in Germany. The evaluated pricing models are the “supply and demand ratio” (SDR) and “mid-market rate pricing” (MMR). A time series aggregation can reduce time series data of municipalities by 99.4% with less than 5% error for 98.2% (load) and 95.5% (generation) of all municipalities and hence decrease the simulation time needed to create sufficient training data. This paper combines time series aggregation and emulation in a novel approach and shows significant acceleration by up to 88.9% of the model’s initial runtime for the simulation of the entire population of around 12,000 municipalities. The time for re-calculating the population (e.g., for different scenarios or sensitivity analysis) can be increased by a factor of 1100 while still retaining high accuracy. The analysis of the simulation time shows that time series aggregation and emulation, considered individually, only bring minor improvements in the runtime but can, however, be combined effectively. This can significantly speed up both the simulation itself and the training of the emulation model and allows for flexible use, depending on the capabilities of the models and the practitioners. The results of the peer-to-peer pricing for approximately 12,000 German municipalities show great potential for energy communities. The mechanisms offer good incentives for the addition of necessary flexibility.

Longitudinal clustering analysis and prediction of Parkinson's disease progression using radiomics and hybrid machine learning

BACKGROUND

We employed machine learning approaches to (I) determine distinct progression trajectories in Parkinson's disease (PD) (unsupervised clustering task), and (II) predict progression trajectories (supervised prediction task), from early (years 0 and 1) data, making use of clinical and imaging features.

METHODS

We studied PD-subjects derived from longitudinal datasets (years 0, 1, 2 & 4; Parkinson's Progressive Marker Initiative). We extracted and analyzed 981 features, including motor, non-motor, and radiomics features extracted for each region-of-interest (ROIs: left/right caudate and putamen) using our standardized standardized environment for radiomics analysis (SERA) radiomics software. Segmentation of ROIs on dopamine transposer - single photon emission computed tomography (DAT SPECT) images were performed via magnetic resonance images (MRI). After performing cross-sectional clustering on 885 subjects (original dataset) to identify disease subtypes, we identified optimal longitudinal trajectories using hybrid machine learning systems (HMLS), including principal component analysis (PCA) + K-Means algorithms (KMA) followed by Bayesian information criterion (BIC), Calinski-Harabatz criterion (CHC), and elbow criterion (EC). Subsequently, prediction of the identified trajectories from early year data was performed using multiple HMLSs including 16 Dimension Reduction Algorithms (DRA) and 10 classification algorithms.

RESULTS

We identified 3 distinct progression trajectories. Hotelling's t squared test (HTST) showed that the identified trajectories were distinct. The trajectories included those with (I, II) disease escalation (2 trajectories, 27% and 38% of patients) and (III) stable disease (1 trajectory, 35% of patients). For trajectory prediction from early year data, HMLSs including the stochastic neighbor embedding algorithm (SNEA, as a DRA) as well as locally linear embedding algorithm (LLEA, as a DRA), linked with the new probabilistic neural network classifier (NPNNC, as a classifier), resulted in accuracies of 78.4% and 79.2% respectively, while other HMLSs such as SNEA + Lib_SVM (library for support vector machines) and t_SNE (t-distributed stochastic neighbor embedding) + NPNNC resulted in 76.5% and 76.1% respectively.

CONCLUSIONS

This study moves beyond cross-sectional PD subtyping to clustering of longitudinal disease trajectories. We conclude that combining medical information with SPECT-based radiomics features, and optimal utilization of HMLSs, can identify distinct disease trajectories in PD patients, and enable effective prediction of disease trajectories from early year data.

Deciphering a Pharmacophore Network: A Case Study Using BCR-ABL Data

This paper introduces a general method that can be used to create groups of pharmacophores to support their further in-depth analysis. A BCR-ABL molecular dataset was used to calculate graph edit distances between pharmacophores and led to their organization into a novel pharmacophore network. The application of a graph layout algorithm allowed us to discriminate between the pharmacophores associated with active compounds and those associated with inactive compounds. A clustering approach was used to refine the partitioning by grouping the pharmacophores based on their structures, activities, and binding modes. Analysis of a newly spatialized pharmacophore network provided us with critical insight into structure-activity relationships, most notably those that revealed distinctions between activity classes and chemical families. As shown, this method permits us to identify families of structurally homogeneous pharmacophores.

Longitudinal Methods for Modeling Exposures in Pharmacoepidemiologic Studies in Pregnancy

In many perinatal pharmacoepidemiologic studies, exposure to a medication is classified as "ever exposed" versus "never exposed" within each trimester or even over the entire pregnancy. This approach is often far from real-world exposure patterns, may lead to exposure misclassification, and does not to incorporate important aspects such as dosage, timing of exposure, and treatment duration. Alternative exposure modeling methods can better summarize complex, individual-level medication use trajectories or time-varying exposures from information on medication dosage, gestational timing of use, and frequency of use. We provide an overview of commonly used methods for more refined definitions of real-world exposure to medication use during pregnancy, focusing on the major strengths and limitations of the techniques, including the potential for method-specific biases. Unsupervised clustering methods, including k-means clustering, group-based trajectory models, and hierarchical cluster analysis, are of interest because they enable visual examination of medication use trajectories over time in pregnancy and complex individual-level exposures, as well as providing insight into comedication and drug-switching patterns. Analytical techniques for time-varying exposure methods, such as extended Cox models and Robins' generalized methods, are useful tools when medication exposure is not static during pregnancy. We propose that where appropriate, combining unsupervised clustering techniques with causal modeling approaches may be a powerful approach to understanding medication safety in pregnancy, and this framework can also be applied in other areas of epidemiology.

MarkovHC: Markov hierarchical clustering for the topological structure of high-dimensional single-cell omics data with transition pathway and critical point detection

Clustering cells and depicting the lineage relationship among cell subpopulations are fundamental tasks in single-cell omics studies. However, existing analytical methods face challenges in stratifying cells, tracking cellular trajectories, and identifying critical points of cell transitions. To overcome these, we proposed a novel Markov hierarchical clustering algorithm (MarkovHC), a topological clustering method that leverages the metastability of exponentially perturbed Markov chains for systematically reconstructing the cellular landscape. Briefly, MarkovHC starts with local connectivity and density derived from the input and outputs a hierarchical structure for the data. We firstly benchmarked MarkovHC on five simulated datasets and ten public single-cell datasets with known labels. Then, we used MarkovHC to investigate the multi-level architectures and transition processes during human embryo preimplantation development and gastric cancer procession. MarkovHC found heterogeneous cell states and sub-cell types in lineage-specific progenitor cells and revealed the most possible transition paths and critical points in the cellular processes. These results demonstrated MarkovHC's effectiveness in facilitating the stratification of cells, identification of cell populations, and characterization of cellular trajectories and critical points.

SRIQ clustering: A fusion of Random Forest, QT clustering, and KNN concepts

Gene expression profiling together with unsupervised analysis methods, typically clustering methods, has been used extensively in cancer research to unravel, e.g., new molecular subtypes that hold promise of disease refinement that may ultimately benefit patients. However, many of the commonly used methods require a prespecified number of clusters to extract and frequently require some type of feature pre-selection, e.g. variance filtering. This introduces subjectivity to the process of cluster discovery and the definition of putative novel tumor subtypes. Here, we introduce SRIQ, a novel unsupervised clustering method that could circumvent some of the issues in commonly used unsupervised analysis methods. SRIQ incorporates concepts from random forest machine learning as well as quality threshold- and k-nearest neighbor clustering. It is implemented as a Java and Python pipeline including data pre-processing, differential expression analysis, and pathway analysis. Using 434 lung adenocarcinomas profiled by RNA sequencing, we demonstrate the technical reproducibility of SRIQ and benchmark its performance compared to the commonly used consensus clustering method. Based on differential gene expression analysis and auxiliary molecular data we show that SRIQ can define new tumor subsets that appear biologically relevant and consistent compared and that these new subgroups seem to refine existing transcriptional subtypes that were defined using consensus clustering. Together, this provides support that SRIQ may be a useful new tool for unsupervised analysis of gene expression data from human malignancies.

Unsupervised Algorithms for Microarray Sample Stratification

The amount of data made available by microarrays gives researchers the opportunity to delve into the complexity of biological systems. However, the noisy and extremely high-dimensional nature of this kind of data poses significant challenges. Microarrays allow for the parallel measurement of thousands of molecular objects spanning different layers of interactions. In order to be able to discover hidden patterns, the most disparate analytical techniques have been proposed. Here, we describe the basic methodologies to approach the analysis of microarray datasets that focus on the task of (sub)group discovery.

Relationship between Visibility, Air Pollution Index and Annual Mortality Rate in Association with the Occurrence of Rainfall—A Probabilistic Approach

An innovative method was proposed to facilitate the analyses of meteorological conditions and selected air pollution indices’ influence on visibility, air quality index and mortality. The constructed calculation algorithm is dedicated to simulating the visibility in a single episode, first of all. It was derived after applying logistic regression methodology. It should be stressed that eight visibility thresholds (Vis) were adopted in order to build proper classification models with a number of relevant advantages. At first, there exists the possibility to analyze the impact of independent variables on visibility with the consideration of its’ real variability. Secondly, through the application of the Monte Carlo method and the assumed classification algorithms, it was made possible to model the number of days during a precipitation and no-precipitation periods in a yearly cycle, on which the visibility ranged practically: Vis < 8; Vis = 8–12 km, Vis = 12–16 km, Vis = 16–20 km, Vis = 20–24 km, Vis = 24–28 km, Vis = 28–32 km, Vis > 32 km. The derived algorithm proved a particular role of precipitation and no-precipitation periods in shaping the air visibility phenomena. Higher visibility values and a lower number of days with increased visibility were found for the precipitation period contrary to no-precipitation one. The air quality index was lower for precipitation days, and moreover, strong, non-linear relationships were found between mortality and visibility, considering precipitation and seasonality effects.

Characterisation of Temporal Patterns in Step Count Behaviour from Smartphone App Data: An Unsupervised Machine Learning Approach

The increasing ubiquity of smartphone data, with greater spatial and temporal coverage than achieved by traditional study designs, have the potential to provide insight into habitual physical activity patterns. This study implements and evaluates the utility of both K-means clustering and agglomerative hierarchical clustering methods in identifying weekly and yearlong physical activity behaviour trends. Characterising the demographics and choice of activity type within the identified clusters of behaviour. Across all seven clusters of seasonal activity behaviour identified, daylight saving was shown to play a key role in influencing behaviour, with increased activity in summer months. Investigation into weekly behaviours identified six clusters with varied roles, of weekday versus weekend, on the likelihood of meeting physical activity guidelines. Preferred type of physical activity likewise varied between clusters, with gender and age strongly associated with cluster membership. Key relationships are identified between weekly clusters and seasonal activity behaviour clusters, demonstrating how short-term behaviours contribute to longer-term activity patterns. Utilising unsupervised machine learning, this study demonstrates how the volume and richness of secondary app data can allow us to move away from aggregate measures of physical activity to better understand temporal variations in habitual physical activity behaviour.

New confinement index and new perspective for comparing countries - COVID-19

BACKGROUND AND OBJECTIVE

In the difficult problem of comparing countries regarding their lockdown measures or deaths caused by the COVID-19, there is still no agreement on what is the best strategy to follow. Thus, we propose a new way of comparison countries that avoids the main difficulties in the comparison by using three-dimensional trajectories for this type of data.

METHODS

We introduce a new index to analyze the level of confinement that each country was subject to overtime, based on the Community Mobility Reports published by Google resorting to Principal Component Analysis. Subsequently, by using longitudinal clustering, we divide the European countries into similar groups according to the COVID-19 obits and also to the confinement index. However, to make the most out of the clustering methods we resort to artificial longitudinal data to evaluate both the methods and the indices.

RESULTS

By using artificial data, we discover that Calinski-Harabasz outperformed other internal indices in indicating the real number of clusters. The tests also suggested that K-means with Euclidean distance was the best method among the ones studied. With the application to both the mobility and fatalities datasets, we found two groups in each one.

CONCLUSIONS

Our analysis enables us to discover that European northern countries had more mobility during the first confinement and that the deaths caused by COVID-19 started to drop around the 40th day since the first death.

Gene Expression Analysis through Parallel Non-Negative Matrix Factorization

Genetic expression analysis is a principal tool to explain the behavior of genes in an organism when exposed to different experimental conditions. In the state of art, many clustering algorithms have been proposed. It is overwhelming the amount of biological data whose high-dimensional structure exceeds mostly current computational architectures. The computational time and memory consumption optimization actually become decisive factors in choosing clustering algorithms. We propose a clustering algorithm based on Non-negative Matrix Factorization and K-means to reduce data dimensionality but whilst preserving the biological context and prioritizing gene selection, and it is implemented within parallel GPU-based environments through the CUDA library. A well-known dataset is used in our tests and the quality of the results is measured through the Rand and Accuracy Index. The results show an increase in the acceleration of 6.22× compared to the sequential version. The algorithm is competitive in the biological datasets analysis and it is invariant with respect to the classes number and the size of the gene expression matrix.

Feature Selection and Ensemble-Based Intrusion Detection System: An Efficient and Comprehensive Approach

The emergence of ground-breaking technologies such as artificial intelligence, cloud computing, big data powered by the Internet, and its highly valued real-world applications consisting of symmetric and asymmetric data distributions, has significantly changed our lives in many positive aspects. However, it equally comes with the current catastrophic daily escalating cyberattacks. Thus, raising the need for researchers to harness the innovative strengths of machine learning to design and implement intrusion detection systems (IDSs) to help mitigate these unfortunate cyber threats. Nevertheless, trustworthy and effective IDSs is a challenge due to low accuracy engendered by vast, irrelevant, and redundant features; inept detection of all types of novel attacks by individual machine learning classifiers; costly and faulty use of labeled training datasets cum significant false alarm rates (FAR) and the excessive model building and testing time. Therefore, this paper proposed a promising hybrid feature selection (HFS) with an ensemble classifier, which efficiently selects relevant features and provides consistent attack classification. Initially, we harness the various strengths of CfsSubsetEval, genetic search, and a rule-based engine to effectively select subsets of features with high correlation, which considerably reduced the model complexity and enhanced the generalization of learning algorithms, both of which are symmetry learning attributes. Moreover, using a voting method and average of probabilities, we present an ensemble classifier that used K-means, One-Class SVM, DBSCAN, and Expectation-Maximization, abbreviated (KODE) as an enhanced classifier that consistently classifies the asymmetric probability distributions between malicious and normal instances. HFS-KODE achieves remarkable results using 10-fold cross-validation, CIC-IDS2017, NSL-KDD, and UNSW-NB15 datasets and various metrics. For example, it outclassed all the selected individual classification methods, cutting-edge feature selection, and some current IDSs techniques with an excellent performance accuracy of 99.99%, 99.73%, and 99.997%, and a detection rate of 99.75%, 96.64%, and 99.93% for CIC-IDS2017, NSL-KDD, and UNSW-NB15, respectively based on only 11, 8, 13 selected relevant features from the above datasets. Finally, considering the drastically reduced FAR and time, coupled with no need for labeled datasets, it is self-evident that HFS-KODE proves to have a remarkable performance compared to many current approaches.

Altered metabolomic states elicited by Flg22 and FlgII-28 in Solanum lycopersicum: intracellular perturbations and metabolite defenses

BACKGROUND

Surveillance of potential pathogens is a key feature of plant innate immunity. For non-self-recognition plants rely on the perception of pathogen-derived molecules. Early post-perception events activate signaling cascades, leading to the synthesis of defense-related proteins and specialized metabolites, thereby providing a broad-spectrum antimicrobial coverage. This study was concerned with tracking changes in the tomato plant metabolome following perception of the flagellum-derived elicitors (Flg22 and FlgII-28).

RESULTS

Following an untargeted metabolomics workflow, the metabolic profiles of a Solanum lycopersicum cultivar were monitored over a time range of 16-32 h post-treatment. Liquid chromatography was used to resolve the complex mixture of metabolites and mass spectrometry for the detection of differences associated with the elicitor treatments. Stringent data processing and multivariate statistical tools were applied to the complex dataset to extract relevant metabolite features associated with the elicitor treatments. Following perception of Flg22 and FlgII-28, both elicitors triggered an oxidative burst, albeit with different kinetic responses. Signatory biomarkers were annotated from diverse metabolite classes which included amino acid derivatives, lipid species, steroidal glycoalkaloids, hydroxybenzoic acids, hydroxycinnamic acids and derivatives, as well as flavonoids.

CONCLUSIONS

An untargeted metabolomics approach adequately captured the subtle and nuanced perturbations associated with elicitor-linked plant defense responses. The shared and unique features characterizing the metabolite profiles suggest a divergence of signal transduction events following perception of Flg22 vs. FlgII-28, leading to a differential reorganization of downstream metabolic pathways.

Network Theory and Switching Behaviors: A User Guide for Analyzing Electronic Records Databases

As part of studies that employ health electronic records databases, this paper advocates the employment of graph theory for investigating drug-switching behaviors. Unlike the shared approach in this field (comparing groups that have switched with control groups), network theory can provide information about actual switching behavior patterns. After a brief and simple introduction to fundamental concepts of network theory, here we present (i) a Python script to obtain an adjacency matrix from a records database and (ii) an illustrative example of the application of network theory basic concepts to investigate drug-switching behaviors. Further potentialities of network theory (weighted matrices and the use of clustering algorithms), along with the generalization of these methods to other kinds of switching behaviors beyond drug switching, are discussed.

Risk profiles for negative and positive COVID-19 hospitalized patients

COVID-19 is a viral infection that affects people differently, where the majority of cases develop mild symptoms, some people require hospitalization, and unfortunately, a small number of patients perish. Hence, identifying risk factors is critical for physicians to make treatment decisions. The purpose of this article is to determine whether unsupervised analysis of risk factors in positive and negative COVID-19 subjects can aid in the identification of a set of reliable and clinically relevant risk profiles. Positive and negative patients hospitalized were randomly selected from the Mexican Open Registry between March and May 2020. Thirteen risk factors, three distinct outcomes, and COVID-19 test results were used to categorize registry patients. As a result, the dataset was reported via 6144 different risk profiles for each age group. The unsupervised learning method is proposed in this study to discover the most prevalent risk profiles. The data was partitioned into discovery (70%) and validation (30%) sets. The discovery set was analyzed using the partition around medoids (PAM) method, and the stable set of risk profiles was estimated using robust consensus clustering. The PAM models' reliability was validated by predicting the risk profile of subjects from the validation set and patients admitted in November 2020. In the validation set, the clinical relevance of the risk profiles was evaluated by determining the prevalence of three patient outcomes: pneumonia diagnosis, ICU admission, or death. Six positive and five negative COVID-19 risk profiles were identified, with significant statistical differences between them. As a result, PAM clustering with consensus mapping is a viable method for discovering unsupervised risk profiles in subjects with severe respiratory health problems.

The utility of clusters and a Hungarian clustering algorithm

Implicit in the k-means algorithm is a way to assign a value, or utility, to a cluster of points. It works by taking the centroid of the points and the value of the cluster is the sum of distances from the centroid to each point in the cluster. The aim in this paper is to introduce an alternative way to assign a value to a cluster. Motivation is provided. Moreover, whereas the k-means algorithm does not have a natural way to determine k if it is unknown, we can use our method of evaluating a cluster to find good clusters in a sequential manner. The idea uses optimizations over permutations and clusters are set by the cyclic groups; generated by the Hungarian algorithm.

The topology of data: Opportunities for cancer research

Motivation

Topological methods have recently emerged as a reliable and interpretable framework for extracting information from high-dimensional data, leading to the creation of a branch of applied mathematics called Topological Data Analysis (TDA). Since then, TDA has been progressively adopted in biomedical research. Biological data collection can result in enormous datasets, comprising thousands of features and spanning diverse datatypes. This presents a barrier to initial data analysis as the fundamental structure of the dataset becomes hidden, obstructing the discovery of important features and patterns. TDA provides a solution to obtain the underlying shape of datasets over continuous resolutions, corresponding to key topological features independent of noise. TDA has the potential to support future developments in healthcare as biomedical datasets rise in complexity and dimensionality. Previous applications extend across the fields of neuroscience, oncology, immunology and medical image analysis. TDA has been used to reveal hidden subgroups of cancer patients, construct organizational maps of brain activity and classify abnormal patterns in medical images. The utility of TDA is broad and to understand where current achievements lie, we have evaluated the present state of TDA in cancer data analysis.

Results

This article aims to provide an overview of TDA in Cancer Research. A brief introduction to the main concepts of TDA is provided to ensure that the article is accessible to readers who are not familiar with this field. Following this, a focussed literature review on the field is presented, discussing how TDA has been applied across heterogeneous datatypes for cancer research.

An SVM approach towards breast cancer classification from H&E-stained histopathology images based on integrated features

Breast cancer is one among the most frequent reasons of women's death worldwide. Nowadays, healthcare informatics is mainly focussing on the classification of breast cancer images, due to the lethal nature of this cancer. There are chances of inter- and intra-observer variability that may lead to misdiagnosis in the detection of cancer. This study proposed an automatic breast cancer classification system that uses support vector machine (SVM) classifier based on integrated features (texture, geometrical, and color). The University of California Santa Barbara (UCSB) dataset and BreakHis dataset, which are available in public domain, were used. A classification comparison module which involves SVM, k-nearest neighbor (k-NN), random forest (RF), and artificial neural network (ANN) was also proposed to determine the classifier that best suits for the application of breast cancer detection from histopathology images. The performance of these classifiers was analyzed against metrics like accuracy, specificity, sensitivity, balanced accuracy, and F-score. Results showed that among the classifiers, the SVM classifier performed better with a test accuracy of approximately 90% on both the datasets. Additionally, the significance of the proposed integrated SVM model was statistically analyzed against other classifier models.

Novel Metrics to Characterize In Vitro Pollen Tube Growth Performance of Apple Cultivars

In vitro germination assays are frequently used in screening trials to evaluate the pollen viability of pollinizers. To be effective, screening trials must have defined threshold criteria, from which individuals can then be assessed. However, despite decades of research on pollen viability, no established threshold is available to categorize apple cultivars based on their in vitro pollen tube lengths. This study aimed to identify and characterize the subgroups of cultivars based on their pollen tube growth performance. In vitro pollen tube lengths of 41 individuals were determined by incubating samples on artificial germination media at 15 and 25 °C. A six-number summary statistic was calculated, and hierarchical clustering on principal component (HCPC) analysis was used to determine and characterize subgroups. Furthermore, a decision tree model was used to predict class membership for future datasets. HCPC analysis partitioned the 41 individuals into three subgroups with different performances. The decision tree quickly predicted the cluster membership based on the second quartile at 15 °C and the third quartile at 25 °C. The thresholds from the decision tree can be used to characterize new observations. The use of the methods will be demonstrated using a case study with 29 apple accessions.

VICTOR: A visual analytics web application for comparing cluster sets

Clustering is the process of grouping different data objects based on similar properties. Clustering has applications in various case studies from several fields such as graph theory, image analysis, pattern recognition, statistics and others. Nowadays, there are numerous algorithms and tools able to generate clustering results. However, different algorithms or parameterizations may produce quite dissimilar cluster sets. In this way, the user is often forced to manually filter and compare these results in order to decide which of them generate the ideal clusters. To automate this process, in this study, we present VICTOR, the first fully interactive and dependency-free visual analytics web application which allows the visual comparison of the results of various clustering algorithms. VICTOR can handle multiple cluster set results simultaneously and compare them using ten different metrics. Clustering results can be filtered and compared to each other with the use of data tables or interactive heatmaps, bar plots, correlation networks, sankey and circos plots. We demonstrate VICTOR's functionality using three examples. In the first case, we compare five different network clustering algorithms on a Yeast protein-protein interaction dataset whereas in the second example, we test four different parameters of the MCL clustering algorithm on the same dataset. Finally, as a third example, we compare four different meta-analyses with hierarchically clustered differentially expressed genes found to be involved in myocardial infarction. VICTOR is available at http://victor.pavlopouloslab.info or http://bib.fleming.gr:3838/VICTOR.

Improving tuberculosis surveillance by detecting international transmission using publicly available whole genome sequencing data

IntroductionImproving the surveillance of tuberculosis (TB) is especially important for multidrug-resistant (MDR) and extensively drug-resistant (XDR) TB. The large amount of publicly available whole genome sequencing (WGS) data for TB gives us the chance to re-use data and to perform additional analyses at a large scale.AimWe assessed the usefulness of raw WGS data of global MDR/XDR Mycobacterium tuberculosis isolates available from public repositories to improve TB surveillance.MethodsWe extracted raw WGS data and the related metadata of M. tuberculosis isolates available from the Sequence Read Archive. We compared this public dataset with WGS data and metadata of 131 MDR- and XDR M. tuberculosis isolates from Germany in 2012 and 2013.ResultsWe aggregated a dataset that included 1,081 MDR and 250 XDR isolates among which we identified 133 molecular clusters. In 16 clusters, the isolates were from at least two different countries. For example, Cluster 2 included 56 MDR/XDR isolates from Moldova, Georgia and Germany. When comparing the WGS data from Germany with the public dataset, we found that 11 clusters contained at least one isolate from Germany and at least one isolate from another country. We could, therefore, connect TB cases despite missing epidemiological information.ConclusionWe demonstrated the added value of using WGS raw data from public repositories to contribute to TB surveillance. Comparing the German with the public dataset, we identified potential international transmission events. Thus, using this approach might support the interpretation of national surveillance results in an international context.

A novel hybrid predictive maintenance model based on clustering, smote and multi-layer perceptron neural network optimised with grey wolf algorithm

Abstract

Considering the complexities and challenges in the classification of multiclass and imbalanced fault conditions, this study explores the systematic combination of unsupervised and supervised learning by hybridising clustering (CLUST) and optimised multi-layer perceptron neural network with grey wolf algorithm (GWO-MLP). The hybrid technique was meticulously examined on a historical hydraulic system dataset by first, extracting and selecting the most significant statistical time-domain features. The selected features were then grouped into distinct clusters allowing for reduced computational complexity through a comparative study of four different and frequently used categories of unsupervised clustering algorithms in fault classification. The Synthetic Minority Over Sampling Technique (SMOTE) was then employed to balance the classes of the training samples from the various clusters which then served as inputs for training the supervised GWO-MLP. To validate the proposed hybrid technique (CLUST-SMOTE-GWO-MLP), it was compared with its distinct modifications (variants). The superiority of CLUST-SMOTE-GWO-MLP is demonstrated by outperforming all the distinct modifications in terms of test accuracy and seven other statistical performance evaluation metrics (error rate, sensitivity, specificity, precision, F score, Mathews Correlation Coefficient and geometric mean). The overall analysis indicates that the proposed CLUST-SMOTE-GWO-MLP is efficient and can be used to classify multiclass and imbalanced fault conditions.

Article Highlights

Governance modes in supply chains and financial performance at buyer, supplier and dyadic levels: the positive impact of power balance

Purpose

This study analyzes whether power in the supply chain, based on governance modes and network centrality, explain financial performance at different levels of analysis: buyers, suppliers and dyads.

Design/methodology/approach

The study employs a dual macro-micro lens based on global value chain (i.e. market, modular, relational and captive governance modes) and social network analysis (network centrality) to assess the impact of power (im)balance onto financial performance. Different from previous research, this study adopts information reliability techniques – such as information entropy – to differentiate the weights of distinct financial performance metrics in terms of the maximal entropy principle. This principle states that the probability distribution that best represents the current state of knowledge given prior data is the one with largest entropy. These weights are used in TOPSIS analysis.

Findings

Results offer insightful reflections to SCM research. We show that buyers outperform suppliers due to power asymmetry. We ground our findings both analyzing across governance modes and comparing network centrality. We show that market and modular governances (where power balance prevails) outperform relational and captive modes at the dyadic level – thus inferring that in the long run these governance modes may lead to financially healthier supply chains.

Originality/value

This study advances SCM research by exploring the impact of governance modes and network centrality on performance at both firm and dyadic levels while employing an innovative combination of secondary data and robust set of techniques including TOPSIS, WASPAS and information entropy.

Genome Reconstruction Attacks Against Genomic Data-Sharing Beacons

Sharing genome data in a privacy-preserving way stands as a major bottleneck in front of the scientific progress promised by the big data era in genomics. A community-driven protocol named genomic data-sharing beacon protocol has been widely adopted for sharing genomic data. The system aims to provide a secure, easy to implement, and standardized interface for data sharing by only allowing yes/no queries on the presence of specific alleles in the dataset. However, beacon protocol was recently shown to be vulnerable against membership inference attacks. In this paper, we show that privacy threats against genomic data sharing beacons are not limited to membership inference. We identify and analyze a novel vulnerability of genomic data-sharing beacons: genome reconstruction. We show that it is possible to successfully reconstruct a substantial part of the genome of a victim when the attacker knows the victim has been added to the beacon in a recent update. In particular, we show how an attacker can use the inherent correlations in the genome and clustering techniques to run such an attack in an efficient and accurate way. We also show that even if multiple individuals are added to the beacon during the same update, it is possible to identify the victim's genome with high confidence using traits that are easily accessible by the attacker (e.g., eye color or hair type). Moreover, we show how a reconstructed genome using a beacon that is not associated with a sensitive phenotype can be used for membership inference attacks to beacons with sensitive phenotypes (e.g., HIV+). The outcome of this work will guide beacon operators on when and how to update the content of the beacon and help them (along with the beacon participants) make informed decisions.

Large-Scale 802.11 Wireless Networks Data Analysis Based on Graph Clustering

This paper analyzes a large-scale dataset of real-world Wi-Fi operating networks, collected from more than 9,000 access points (APs) for 1 year. The APs are distributed among more than 1,200 educational centers in the context of a nation-wide one-to-one computing program, being most of them primary and secondary schools. The data corresponds to RSSI measurements between APs used to build the conflict graphs for each school Wi-Fi network. We propose a simple embedding for the Wi-Fi network conflict graphs based on classical graph features, which proves to be useful to analyze the behavior of the wireless networks, showing a high discrimination power among the different school networks. Moreover, we discuss some practical applications of the embedding. In particular, it enables to study the Wi-Fi network dynamics at each school, analyzing the conflict graphs temporal variations through clustering techniques. The presented methodology allows us to successfully separate the most stable scenarios from those with more significant variability, which therefore require more technical resources to optimize the network. Besides, we also compared the behaviour of the Wi-Fi networks of the different schools, which enable us to reuse the optimal configuration found for one school in all those sites that have similar conflict graph patterns.

Premature mortality due to air pollution in European cities: a health impact assessment

BACKGROUND

Ambient air pollution is a major environmental cause of morbidity and mortality worldwide. Cities are generally hotspots for air pollution and disease. However, the exact extent of the health effects of air pollution at the city level is still largely unknown. We aimed to estimate the proportion of annual preventable deaths due to air pollution in almost 1000 cities in Europe.

METHODS

We did a quantitative health impact assessment for the year 2015 to estimate the effect of air pollution exposure (PM_2·5 and NO₂) on natural-cause mortality for adult residents (aged ≥20 years) in 969 cities and 47 greater cities in Europe. We retrieved the cities and greater cities from the Urban Audit 2018 dataset and did the analysis at a 250 m grid cell level for 2015 data based on the global human settlement layer residential population. We estimated the annual premature mortality burden preventable if the WHO recommended values (ie, 10 μg/m³ for PM_2·5 and 40 μg/m³ for NO₂) were achieved and if air pollution concentrations were reduced to the lowest values measured in 2015 in European cities (ie, 3·7 μg/m³ for PM_2·5 and 3·5 μg/m³ for NO₂). We clustered and ranked the cities on the basis of population and age-standardised mortality burden associated with air pollution exposure. In addition, we did several uncertainty and sensitivity analyses to test the robustness of our estimates.

FINDINGS

Compliance with WHO air pollution guidelines could prevent 51 213 (95% CI 34 036-68 682) deaths per year for PM_2·5 exposure and 900 (0-2476) deaths per year for NO₂ exposure. The reduction of air pollution to the lowest measured concentrations could prevent 124 729 (83 332-166 535) deaths per year for PM_2·5 exposure and 79 435 (0-215 165) deaths per year for NO₂ exposure. A great variability in the preventable mortality burden was observed by city, ranging from 0 to 202 deaths per 100 000 population for PM_2·5 and from 0 to 73 deaths for NO₂ per 100 000 population when the lowest measured concentrations were considered. The highest PM_2·5 mortality burden was estimated for cities in the Po Valley (northern Italy), Poland, and Czech Republic. The highest NO₂ mortality burden was estimated for large cities and capital cities in western and southern Europe. Sensitivity analyses showed that the results were particularly sensitive to the choice of the exposure response function, but less so to the choice of baseline mortality values and exposure assessment method.

INTERPRETATION

A considerable proportion of premature deaths in European cities could be avoided annually by lowering air pollution concentrations, particularly below WHO guidelines. The mortality burden varied considerably between European cities, indicating where policy actions are more urgently needed to reduce air pollution and achieve sustainable, liveable, and healthy communities. Current guidelines should be revised and air pollution concentrations should be reduced further to achieve greater protection of health in cities.

FUNDING

Spanish Ministry of Science and Innovation, Internal ISGlobal fund.

Application of Plant Growth Regulators Modulates the Profile of Chlorogenic Acids in Cultured Bidens pilosa Cells

Plant cell culture offers an alternative to whole plants for the production of biologically important specialised metabolites. In cultured plant cells, manipulation by auxin and cytokinin plant growth regulators (PGRs) may lead to in vitro organogenesis and metabolome changes. In this study, six different combination ratios of 2,4-dichlorophenoxyacetic acid (2,4-D) and benzylaminopurine (BAP) were investigated with the aim to induce indirect organogenesis from Bidens pilosa callus and to investigate the associated induced changes in the metabolomes of these calli. Phenotypic appearance of the calli and total phenolic contents of hydromethanolic extracts indicated underlying biochemical differences that were investigated using untargeted metabolomics, based on ultra-high-performance liquid chromatography quadrupole time-of-flight mass spectrometry (UHPLC–qTOF–MS), combined with multivariate data analysis. The concentration and combination ratios of PGRs were shown to induce differential metabolic responses and, thus, distinct metabolomic profiles, dominated by chlorogenic acids consisting of caffeoyl- and feruloyl-derivatives of quinic acid. Although organogenesis was not achieved, the results demonstrate that exogenous application PGRs can be used to manipulate the metabolome of B. pilosa for in vitro production of specialised metabolites with purported pharmacological properties.

Characterization and Monitoring of Antigen-Responsive T Cell Clones Using T Cell Receptor Gene Expression Analysis

High-throughput T-cell receptor repertoire sequencing constitutes a powerful tool to study T cell responses at the clonal level. However, it does not give information on the functional phenotype of the responding clones and lacks a statistical framework for quantitative evaluation. To overcome this, we combined datasets from different experiments, all starting from the same blood samples. We used a novel, sensitive, UMI-based protocol to perform repertoire analysis on experimental replicates. Applying established bioinformatic routines for transcriptomic expression analysis we explored the dynamics of antigen-induced clonal expansion after in vitro stimulation, identified antigen-responsive clones, and confirmed their activation status using the expression of activation markers upon antigen re-challenge. We demonstrate that the addition of IL-4 after antigen stimulation drives the expansion of T cell clones encoding unique receptor sequences. We show that our approach represents a scalable, high-throughput immunological tool, which can be used to identify and characterize antigen-responsive T cells at clonal level.

Unsupervised learning of Swiss population spatial distribution

The paper deals with the analysis of spatial distribution of Swiss population using fractal concepts and unsupervised learning algorithms. The research methodology is based on the development of a high dimensional feature space by calculating local growth curves, widely used in fractal dimension estimation and on the application of clustering algorithms in order to reveal the patterns of spatial population distribution. The notion “unsupervised” also means, that only some general criteria—density, dimensionality, homogeneity, are used to construct an input feature space, without adding any supervised/expert knowledge. The approach is very powerful and provides a comprehensive local information about density and homogeneity/fractality of spatially distributed point patterns.

A method for predicting crash configurations using counterfactual simulations and real-world data

Traffic safety technologies revolve around two principle ideas; crash avoidance and injury mitigation for inevitable crashes. The development of relevant vehicle injury mitigating technologies should consider the interaction of those two technologies, ensuring that the inevitable crashes can be adequately managed by the occupant and vulnerable road user (VRU) protection systems. A step towards that is the accurate description of the expected crashes remaining when crash-avoiding technologies are available in vehicles. With the overall objective of facilitating the assessment of future traffic safety, this study develops a method for predicting crash configurations when introducing crash-avoiding countermeasures. The predicted crash configurations are one important factor for prioritizing the evaluation and development of future occupant and VRU protection systems. By using real-world traffic accident data to form the baseline and performing counterfactual model-in-the-loop (MIL) pre-crash simulations, the change in traffic situations (vehicle crashes) provided by vehicles with crash-avoiding technologies can be predicted. The method is built on a novel crash configuration definition, which supports further analysis of the in-crash phase. By clustering and grouping the remaining crashes, a limited number of crash configurations can be identified, still representing and covering the real-world variation. The developed method was applied using Swedish national- and in-depth accident data related to urban intersections and highway driving, and a conceptual Autonomous Emergency Braking system (AEB) computational model. Based on national crash data analysis, the conflict situations Same-Direction rear-end frontal (SD-ref) representing 53 % of highway vehicle-to-vehicle (v2v) crashes, and Straight Crossing Path (SCP) with 21 % of urban v2v intersection crashes were selected for this study. Pre-crash baselines, for SD-ref (n = 1010) and SCP (n = 4814), were prepared based on in-depth accident data and variations of these. Pre-crash simulations identified the crashes not avoided by the conceptual AEB, and the clustering of these revealed 5 and 52 representative crash configurations for the highway SD-ref and urban intersection SCP conflict situations, respectively, to be used in future crashworthiness studies. The results demonstrated a feasible way of identifying, in a predictive way, relevant crash configurations for in-crash testing of injury prevention capabilities.

Cluster-Based Relocation of Stations for Efficient Forest Fire Management in the Province of Valencia (Spain)

Forest fires are undesirable situations with tremendous impacts on wildlife and people's lives. Reaching them quickly is essential to slowing down their expansion and putting them out in an effective manner. This work proposes an optimized distribution of fire stations in the province of Valencia (Spain) to minimize the impacts of forest fires. Using historical data about fires in the Valencia province, together with the location information about existing fire stations and municipalities, two different clustering techniques have been applied. Floyd-Warshall dynamic programming algorithm has been used to estimate the average times to reach fires among municipalities and fire stations in order to quantify the impacts of station relocation. The minimization was done approximately through k-means clustering. The outcomes with different numbers of clusters determined a predicted tradeoff between reducing the time and the cost of more stations. The results show that the proposed relocation of fire stations generally ensures faster arrival to the municipalities compared to the current disposition of fire stations. In addition, deployment costs associated with station relocation are also of paramount importance, so this factor was also taken into account in the proposed approach.

Behavioral structure of users in cryptocurrency market

Human behavior as they engaged in financial activities is intimately connected to the observed market dynamics. Despite many existing theories and studies on the fundamental motivations of the behavior of humans in financial systems, there is still limited empirical deduction of the behavioral compositions of the financial agents from a detailed market analysis. Blockchain technology has provided an avenue for the latter investigation with its voluminous data and its transparency of financial transactions. It has enabled us to perform empirical inference on the behavioral patterns of users in the market, which we explore in the bitcoin and ethereum cryptocurrency markets. In our study, we first determine various properties of the bitcoin and ethereum users by a temporal complex network analysis. After which, we develop methodology by combining k-means clustering and Support Vector Machines to derive behavioral types of users in the two cryptocurrency markets. Interestingly, we found four distinct strategies that are common in both markets: optimists, pessimists, positive traders and negative traders. The composition of user behavior is remarkably different between the bitcoin and ethereum market during periods of local price fluctuations and large systemic events. We observe that bitcoin (ethereum) users tend to take a short-term (long-term) view of the market during the local events. For the large systemic events, ethereum (bitcoin) users are found to consistently display a greater sense of pessimism (optimism) towards the future of the market.

mbkmeans: Fast clustering for single cell data using mini-batch k-means

Single-cell RNA-Sequencing (scRNA-seq) is the most widely used high-throughput technology to measure genome-wide gene expression at the single-cell level. One of the most common analyses of scRNA-seq data detects distinct subpopulations of cells through the use of unsupervised clustering algorithms. However, recent advances in scRNA-seq technologies result in current datasets ranging from thousands to millions of cells. Popular clustering algorithms, such as k-means, typically require the data to be loaded entirely into memory and therefore can be slow or impossible to run with large datasets. To address this problem, we developed the mbkmeans R/Bioconductor package, an open-source implementation of the mini-batch k-means algorithm. Our package allows for on-disk data representations, such as the common HDF5 file format widely used for single-cell data, that do not require all the data to be loaded into memory at one time. We demonstrate the performance of the mbkmeans package using large datasets, including one with 1.3 million cells. We also highlight and compare the computing performance of mbkmeans against the standard implementation of k-means and other popular single-cell clustering methods. Our software package is available in Bioconductor at https://bioconductor.org/packages/mbkmeans.