Uniform Manifold Approximation Research Articles

174-OR: Identification of Gestational Diabetes (GDM) Subtypes via Machine Learning and Associations with Perinatal Complications

There has been increasing recognition of GDM heterogeneity. We aimed to identify GDM subtypes using 45 variables from the electronic health records via Uniform Manifold Approximation and Projection and Density-Based Spatial Clustering of Applications with Noise, and assess associations with risk of perinatal complications via logistic regression. Since risk factors for GDM and complications were used to identify subtypes, association between subtypes and complications were not adjusted. In a cohort study of 37,561 racially and ethnically diverse individuals with GDM in 2011-2021, we split the sample into training/validation (70%/30%) sets. In the training set, 4 clusters (C) were identified. C1 (14.4%) and C2 (7.8%) were more likely to have early GDM diagnosed by Carpenter-Coustan criteria, pre-existing conditions (hypertension, prediabetes, polycystic ovary syndrome, and obesity), and to be older and multiparous, compared to C3 (65.7%) and C4 (12.1%; Table). Compared to C3 as the mildest and most common subtype, C2 and C4 had higher risk of preeclampsia/eclampsia, C2 had higher risk of placental abruption, and C1, C2, and C4 had higher risk of preterm birth, large for gestational age, and neonatal intensive care unit admission. Findings were confirmed in the validation set. Subtyping GDM may inform personalized risk assessment and management to improve perinatal outcomes. Disclosure Y.Zhu: None. R.Harvill: None. M.Greenberg: None. A.Ngo: None. L.D.Liao: None. R.F.Chehab: None. A.Ferrara: None. Funding National Institute of Diabetes and Digestive and Kidney Diseases (K01DK120807); National Heart, Lung, and Blood Institute (R01HL157666)

Pathological changes or technical artefacts? The problem of the heterogenous databases in COVID-19 CXR image analysis

BackgroundWhen the COVID-19 pandemic commenced in 2020, scientists assisted medical specialists with diagnostic algorithm development. One scientific research area related to COVID-19 diagnosis was medical imaging and its potential to support molecular tests. Unfortunately, several systems reported high accuracy in development but did not fare well in clinical application. The reason was poor generalization, a long-standing issue in AI development. Researchers found many causes of this issue and decided to refer to them as confounders, meaning a set of artefacts and methodological errors associated with the method. We aim to contribute to this steed by highlighting an undiscussed confounder related to image resolution. Methods20 216 chest X-ray images (CXR) from worldwide centres were analyzed. The CXRs were bijectively projected into the 2D domain by performing Uniform Manifold Approximation and Projection (UMAP) embedding on the radiomic features (rUMAP) or CNN-based neural features (nUMAP) from the pre-last layer of the pre-trained classification neural network. Additional 44 339 thorax CXRs were used for validation. The comprehensive analysis of the multimodality of the density distribution in rUMAP/nUMAP domains and its relation to the original image properties was used to identify the main confounders. ResultsnUMAP revealed a hidden bias of neural networks towards the image resolution, which the regular up-sampling procedure cannot compensate for. The issue appears regardless of the network architecture and is not observed in a high-resolution dataset. The impact of the resolution heterogeneity can be partially diminished by applying advanced deep-learning-based super-resolution networks. ConclusionsrUMAP and nUMAP are great tools for image homogeneity analysis and bias discovery, as demonstrated by applying them to COVID-19 image data. Nonetheless, nUMAP could be applied to any type of data for which a deep neural network could be constructed. Advanced image super-resolution solutions are needed to reduce the impact of the resolution diversity on the classification network decision.

Microscopy-Directed Imaging Mass Spectrometry for Rapid High Spatial Resolution Molecular Imaging of Glomeruli.

The glomerulus is a multicellular functional tissue unit (FTU) of the nephron that is responsible for blood filtration. Each glomerulus contains multiple substructures and cell types that are crucial for their function. To understand normal aging and disease in kidneys, methods for high spatial resolution molecular imaging within these FTUs across whole slide images is required. Here we demonstrate a workflow using microscopy-driven selected sampling to enable 5 μm pixel size matrix-assisted laser desorption/ionization imaging mass spectrometry (MALDI IMS) of all glomeruli within whole slide human kidney tissues. Such high spatial resolution imaging entails large numbers of pixels, increasing the data acquisition times. Automating FTU-specific tissue sampling enables high-resolution analysis of critical tissue structures, while concurrently maintaining throughput. Glomeruli were automatically segmented using coregistered autofluorescence microscopy data, and these segmentations were translated into MALDI IMS measurement regions. This allowed high-throughput acquisition of 268 glomeruli from a single whole slide human kidney tissue section. Unsupervised machine learning methods were used to discover molecular profiles of glomerular subregions and differentiate between healthy and diseased glomeruli. Average spectra for each glomerulus were analyzed using Uniform Manifold Approximation and Projection (UMAP) and k-means clustering, yielding 7 distinct groups of differentiated healthy and diseased glomeruli. Pixel-wise k-means clustering was applied to all glomeruli, showing unique molecular profiles localized to subregions within each glomerulus. Automated microscopy-driven, FTU-targeted acquisition for high spatial resolution molecular imaging maintains high-throughput and enables rapid assessment of whole slide images at cellular resolution and identification of tissue features associated with normal aging and disease.

Joint Design of a Simultaneous Reflection and Transmission RIS in Mode-Switching Mode to Assist NOMA Systems.

Simultaneous transmitting and reflecting reconfigurable intelligent surfaces (STAR-RISs) can reflect signals and transmissive signals simultaneously and can extend the coverage of signals. A conventional RIS mainly focuses on the case where the signal source and the target are on the same side. In this paper, a STAR-RIS-assisted non-orthogonal multiple access (NOMA) downlink communication system is considered to maximize the achievable rate for users by jointly optimizing the power-allocation coefficients, active beamforming, and STAR-RIS beamforming under the mode-switching (MS) protocol. The critical information of the channel is first extracted using the Uniform Manifold Approximation and Projection (UMAP) method. Based on the key extracted channel features, STAR-RIS elements and users are clustered individually using the fuzzy C-mean clustering (FCM) method. The alternating optimization method decomposes the original optimization problem into three sub-optimization problems. Finally, the sub-problems are converted to unconstrained optimization methods using penalty functions for the solution. Simulation results show that when the number of elements of RIS is 60, the achievable rate of the STAR-RIS-NOMA system is about 18% higher than that of the RIS-NOMA system.

Explainable AI for wearable seizure logging: Impact of data quality, patient age, and antiseizure medication on performance

ObjectiveObjective seizure count estimates are crucial for ambulatory epilepsy management. Wearables have shown promise for the detection of tonic-clonic seizures but may suffer from false alarms and undetected seizures. Seizure signatures recorded by wearables often occur over prolonged periods, including increased levels of electrodermal activity and heart rate long after seizure EEG onset, however, previous detection methods only partially exploited these signatures. Understanding the utility of these prolonged signatures for seizure count estimation and what factors generally determine seizure logging performance, including the role of data quality vs. algorithm performance, is thus crucial for improving wearables-based epilepsy monitoring and determining which patients benefit most from this technology. MethodsIn this retrospective study we examined 76 pediatric epilepsy patients during multiday video-EEG monitoring equipped with a wearable (Empatica E4; records of electrodermal activity, EDA, accelerometry, ACC, heart rate, HR; 1983 h total recording time; 45 tonic-clonic seizures). To log seizures on prolonged data trends, we applied deep learning on continuous overlapping 1-hour segments of multimodal data in a leave-one-subject-out approach. We systematically examined factors influencing logging performance, including patient age, antiseizure medication (ASM) load, seizure type and duration, and data artifacts. To gain insights into algorithm function and feature importance we applied Uniform Manifold Approximation and Projection (UMAP, to represent the separability of learned features) and SHapley Additive exPlanations (SHAP, to represent the most informative data signatures). ResultsPerformance for tonic-clonic seizure logging increased systematically with patient age (AUC 0.61 for patients 〈 11 years, AUC 0.77 for patients between 11–15 years, AUC 0.85 for patients 〉 15 years). Across all ages, AUC was 0.75 corresponding to a sensitivity of 0.52 and a false alarm rate of 0.28/24 h. Seizures under high ASM load or with shorter duration were detected worse (P=.025, P=.033, respectively). UMAP visualized discriminatory power at the individual patient level, SHAP analyses identified clonic motor activity and peri/postictal increases in HR and EDA as most informative. In contrast, in missed seizures, these features were absent indicating that recording quality but not the algorithm caused the low sensitivity in these patients. SignificanceOur results demonstrate the utility of prolonged, postictal data segments for seizure logging, contribute to algorithm explainability and point to influencing factors, including high ASM dose and short seizure duration. Collectively, these results may help to identify patients who particularly benefit from such technology.

Exploratory Data Analysis of Earthquake Moment Tensor Catalog in Japan Using Non-linear Graph-Based Dimensionality Reduction

The catalog of moment tensor solutions, which contains the information on spatial location, origin time and fault mechanism type of earthquakes, has been interpreted by researchers based on a wealth of past knowledge. However, the long-term routine analyses have led to the accumulation of a huge amount of data in the moment tensor catalog, and it is worth considering moving away from the artisanal approach. In this study, using dimensionality reduction of unsupervised machine learning, we performed exploratory data analysis of the moment tensor catalog in Japan to objectively obtain comprehensive images of seismic activity and to acquire knowledge on the spatial and temporal characteristics of the earthquake mechanism. Source parameters of the moment tensor catalog in Japan, spatial location (latitude, longitude, and focal depth) and source-mechanism diagram information were embedded in two-dimensional space via a non-linear graph-based dimensionality-reduction method, Uniform Manifold Approximation and Projection. On the embedding map, earthquakes in eastern and western Japan are distributed separately and are further embedded to reflect their characteristic fault mechanism and focal depth in each region. The similarity degree of the earthquakes can be obtained as the distance on the embedding map. This study demonstrates that the data visualization using dimensionality reduction is useful for intuitively and objectively understanding the regional characteristics of earthquake mechanisms. The embedding map can also be employed to visualize temporal changes in regional seismic activity and to perform a similarity search with a past event.

Misinformation and Public Health Messaging in the Early Stages of the Mpox Outbreak: Mapping the Twitter Narrative With Deep Learning.

Shortly after the worst of the COVID-19 pandemic, an outbreak of mpox introduced another critical public health emergency. Like the COVID-19 pandemic, the mpox outbreak was characterized by a rising prevalence of public health misinformation on social media, through which many US adults receive and engage with news. Digital misinformation continues to challenge the efforts of public health officials in providing accurate and timely information to the public. We examine the evolving topic distributions of social media narratives during the mpox outbreak to map the tension between rapidly diffusing misinformation and public health communication. This study aims to observe topical themes occurring in a large-scale collection of tweets about mpox using deep learning. We leveraged a data set comprised of all mpox-related tweets that were posted between May 7, 2022, and July 23, 2022. We then applied Sentence Bidirectional Encoder Representations From Transformers (S-BERT) to the content of each tweet to generate a representation of its content in high-dimensional vector space, where semantically similar tweets will be located closely together. We projected the set of tweet embeddings to a 2D map by applying principal component analysis and Uniform Manifold Approximation Projection (UMAP). Finally, we group these data points into 7 topical clusters using k-means clustering and analyze each cluster to determine its dominant topics. We analyze the prevalence of each cluster over time to evaluate longitudinal thematic changes. Our deep-learning pipeline revealed 7 distinct clusters of content: (1) cynicism, (2) exasperation, (3) COVID-19, (4) men who have sex with men, (5) case reports, (6) vaccination, and (7) World Health Organization (WHO). Clusters that largely communicated erroneous or irrelevant information began earlier and grew faster, reaching a wider audience than later communications by official instances and health officials. Within a few weeks of the first reported mpox cases, an avalanche of mostly false, misleading, irrelevant, or damaging information started to circulate on social media. Official institutions, including the WHO, acted promptly, providing case reports and accurate information within weeks, but were overshadowed by rapidly spreading social media chatter. Our results point to the need for real-time monitoring of social media content to optimize responses to public health emergencies.

Time-series transcriptome analysis of peripheral blood mononuclear cells obtained from individuals who received the SARS-CoV-2 mRNA vaccine.

Messenger ribonucleic acid (mRNA) vaccination against coronavirus disease 2019 (COVID-19) is an effective prevention strategy, despite a limited understanding of the molecular mechanisms underlying the host immune system and individual heterogeneity of the variable effects of mRNA vaccination. We assessed the time-series changes in the comprehensive gene expression profiles of 200 vaccinated healthcare workers by performing bulk transcriptome and bioinformatics analyses, including dimensionality reduction utilizing the uniform manifold approximation and projection (UMAP) technique. For these analyses, blood samples, including peripheral blood mononuclear cells (PBMCs), were collected from 214 vaccine recipients before vaccination (T1) and on Days 22 (T2, after second dose), 90, 180 (T3, before a booster dose), and 360 (T4, after a booster dose) after receiving the first dose of BNT162b2 vaccine (UMIN000043851). UMAP successfully visualized the main cluster of gene expression at each time point in PBMC samples (T1-T4). Through differentially expressed gene (DEG) analysis, we identified genes that showed fluctuating expression levels and gradual increases in expression levels from T1 to T4, as well as genes with increased expression levels at T4 alone. We also succeeded in dividing these cases into five types based on the changes in gene expression levels. High-throughput and temporal bulk RNA-based transcriptome analysis is a useful approach for inclusive, diverse, and cost-effective large-scale clinical studies.

0424 Visualizing insomnia phenotypes using dimensionality reduction techniques

Abstract Introduction A large number of features can be extracted from a single hypnogram, such as stages durations, onsets, or transitions probabilities. Those numerous indicators can turn a collection of sleep records into a high dimension space. Dimensionality reduction techniques are then useful to reveal patterns in data. We used 3 dimensionality reduction techniques to visualize insomnia phenotypes from a dataset of insomnia and control sleep records: principal component analysis (PCA), t-distributed stochastic neighbor embedding (t-SNE) and uniform manifold approximation and projection (UMAP). Methods 519 sleep records have been included with the following diagnoses: 46 sleep onset insomnia, 83 sleep state misperception, 223 sleep maintenance insomnia, 117 sleep onset and maintenance insomnia and 50 controls (good sleep). We limited the feature extraction to the hypnogram as macrostructure constitute the primary source of information used for diagnosis in polysomnography. We used common macrostructure indicators such as wake after sleep onset (WASO), as well as more intricate features such as stages transitions probabilities. A first set of 54 features per hypnogram was computed. Those features were then projected in a 2 dimensional space using PCA, t-SNE and UMAP. Results Co-ranking matrix of the projections was computed for the 3 techniques (PCA: Kmax=103, Qlocal=0.38; tSNE: Kmax=20, Qlocal=0.44; UMAP: Kmax=160, Qlocal=0.44). UMAP was the technique that projected the hypnograms feature sets in the most meaningful way, by reflecting the individual diagnoses. Interestingly in the UMAP representation, group outliers, such as controls having nevertheless experienced nocturnal awakenings were projected next to similar insomnia subjects, which indicate a fine-grained capture of sleep quality at the individual level. Conclusion Dimensionality reduction and unsupervised techniques are promising methods when approaching high dimensionality spaces. A fine analysis of projected clusters could show subgroups in already known phenotypes. As literature showed robustness and applicability of those methods to larger datasets, dimensionality reduction of insomnia records could be improved by the addition of new features computed from hypnodensities generated by machine learning algorithms, or by spectral features extracted from polysomnographic signals. Support (if any) None

Visual Extraction of Refined Operation Mode of New Power System Based on IPSO-Kmeans

Due to the influence of the high proportion of renewable energy penetration, the time-varying and complex operation mode of the new power system is gradually increasing, leading to a lack of fineness and practicality of traditional operation modes. To this end, a new visual extraction method for fine operation mode of power system is proposed. Specifically, aiming at the dimensional problem between high-dimensional electrical characteristic variables, a power grid operation data preprocessing method based on maximum absolute standardization (MaxAbs) is designed. Then, in order to reduce the impact of redundant features on the accuracy of the operation mode extraction results, the Pearson correlation coefficient is introduced to optimize the feature space relationship matrix, constructing a screening model of operating mode characteristic variables based on pearson kernel principal component analysis (P_KPCA). Then, with the clustering elbow index as the constraint condition, a K-means algorithm based on improved particle swarm optimization (IPSO-Kmeans) was proposed to realize fine operation mode extraction. Finally, the experimental analysis is carried out with the actual operation data of the power grid for one year and based on uniform manifold approximation and projection (UMAP) to visualize the extraction results of the operation mode. The validity and accuracy of the proposed method are verified.

Unsupervised machine learning framework for discriminating major variants of concern during COVID-19.

Due to the high mutation rate of the virus, the COVID-19 pandemic evolved rapidly. Certain variants of the virus, such as Delta and Omicron emerged with altered viral properties leading to severe transmission and death rates. These variants burdened the medical systems worldwide with a major impact to travel, productivity, and the world economy. Unsupervised machine learning methods have the ability to compress, characterize, and visualize unlabelled data. This paper presents a framework that utilizes unsupervised machine learning methods to discriminate and visualize the associations between major COVID-19 variants based on their genome sequences. These methods comprise a combination of selected dimensionality reduction and clustering techniques. The framework processes the RNA sequences by performing a k-mer analysis on the data and further visualises and compares the results using selected dimensionality reduction methods that include principal component analysis (PCA), t-distributed stochastic neighbour embedding (t-SNE), and uniform manifold approximation projection (UMAP). Our framework also employs agglomerative hierarchical clustering to visualize the mutational differences among major variants of concern and country-wise mutational differences for selected variants (Delta and Omicron) using dendrograms. We also provide country-wise mutational differences for selected variants via dendrograms. We find that the proposed framework can effectively distinguish between the major variants and has the potential to identify emerging variants in the future.

Exploratory data analysis (EDA) machine learning approaches for ocean world analog mass spectrometry

Many upcoming and proposed missions to ocean worlds such as Europa, Enceladus, and Titan aim to evaluate their habitability and the existence of potential life on these moons. These missions will suffer from communication challenges and technology limitations. We review and investigate the applicability of data science and unsupervised machine learning (ML) techniques on isotope ratio mass spectrometry data (IRMS) from volatile laboratory analogs of Europa and Enceladus seawaters as a case study for development of new strategies for icy ocean world missions. Our driving science goal is to determine whether the mass spectra of volatile gases could contain information about the composition of the seawater and potential biosignatures. We implement data science and ML techniques to investigate what inherent information the spectra contain and determine whether a data science pipeline could be designed to quickly analyze data from future ocean worlds missions. In this study, we focus on the exploratory data analysis (EDA) step in the analytics pipeline. This is a crucial unsupervised learning step that allows us to understand the data in depth before subsequent steps such as predictive/supervised learning. EDA identifies and characterizes recurring patterns, significant correlation structure, and helps determine which variables are redundant and which contribute to significant variation in the lower dimensional space. In addition, EDA helps to identify irregularities such as outliers that might be due to poor data quality. We compared dimensionality reduction methods Uniform Manifold Approximation and Projection (UMAP) and Principal Component Analysis (PCA) for transforming our data from a high-dimensional space to a lower dimension, and we compared clustering algorithms for identifying data-driven groups (“clusters”) in the ocean worlds analog IRMS data and mapping these clusters to experimental conditions such as seawater composition and CO2 concentration. Such data analysis and characterization efforts are the first steps toward the longer-term science autonomy goal where similar automated ML tools could be used onboard a spacecraft to prioritize data transmissions for bandwidth-limited outer Solar System missions.

A Model-Based Approach Machine Learning to Scalable Portfolio Selection

This study proposes a scalable asset selection and allocation approach using machine learning that integrates clustering methods into portfolio optimization models. The methodology applies the Uniform Manifold Approximation and Projection method and ensemble clustering techniques to preselect assets from the Ibovespa and S&P 500 indices. The research compares three allocation models and finds that the Hierarchical Risk Parity model outperformed the others, with a Sharpe ratio of 1.11. Despite the pandemic's impact on the portfolios, with drawdowns close to 30%, they recovered in 111 to 149 trading days. The portfolios outperformed the indices in cumulative returns, with similar annual volatilities of 20%. Preprocessing with UMAP allowed for finding clusters with higher discriminatory power, evaluated through internal cluster validation metrics, helping to reduce the problem's size during optimal portfolio allocation. Overall, this study highlights the potential of machine learning in portfolio optimization, providing a useful framework for investment practitioners.

Combining the Classification and Pharmacophore Approaches to Understand Homogeneous Olfactory Perceptions at Peripheral Level: Focus on Two Aroma Mixtures.

The mechanisms involved in the homogeneous perception of odorant mixtures remain largely unknown. With the aim of enhancing knowledge about blending and masking mixture perceptions, we focused on structure-odor relationships by combining the classification and pharmacophore approaches. We built a dataset of about 5000 molecules and their related odors and reduced the multidimensional space defined by 1014 fingerprints representing the structures to a tridimensional 3D space using uniform manifold approximation and projection (UMAP). The self-organizing map (SOM) classification was then performed using the 3D coordinates in the UMAP space that defined specific clusters. We explored the allocating in these clusters of the components of two aroma mixtures: a blended mixture (red cordial (RC) mixture, 6 molecules) and a masking binary mixture (isoamyl acetate/whiskey-lactone [IA/WL]). Focusing on clusters containing the components of the mixtures, we looked at the odor notes carried by the molecules belonging to these clusters and also at their structural features by pharmacophore modeling (PHASE). The obtained pharmacophore models suggest that WL and IA could have a common binding site(s) at the peripheral level, but that would be excluded for the components of RC. In vitro experiments will soon be carried out to assess these hypotheses.

Evaluation of nutritional status and clinical depression classification using an explainable machine learning method.

Depression is a prevalent disorder worldwide, with potentially severe implications. It contributes significantly to an increased risk of diseases associated with multiple risk factors. Early accurate diagnosis of depressive symptoms is a critical first step toward management, intervention, and prevention. Various nutritional and dietary compounds have been suggested to be involved in the onset, maintenance, and severity of depressive disorders. Despite the challenges to better understanding the association between nutritional risk factors and the occurrence of depression, assessing the interplay of these markers through supervised machine learning remains to be fully explored. This study aimed to determine the ability of machine learning-based decision support methods to identify the presence of depression using publicly available health data from the Korean National Health and Nutrition Examination Survey. Two exploration techniques, namely, uniform manifold approximation and projection and Pearson correlation, were performed for explanatory analysis among datasets. A grid search optimization with cross-validation was performed to fine-tune the models for classifying depression with the highest accuracy. Several performance measures, including accuracy, precision, recall, F1 score, confusion matrix, areas under the precision-recall and receiver operating characteristic curves, and calibration plot, were used to compare classifier performances. We further investigated the importance of the features provided: visualized interpretation using ELI5, partial dependence plots, and local interpretable using model-agnostic explanations and Shapley additive explanation for the prediction at both the population and individual levels. The best model achieved an accuracy of 86.18% for XGBoost and an area under the curve of 84.96% for the random forest model in original dataset and the XGBoost algorithm with an accuracy of 86.02% and an area under the curve of 85.34% in the quantile-based dataset. The explainable results revealed a complementary observation of the relative changes in feature values, and, thus, the importance of emergent depression risks could be identified. The strength of our approach is the large sample size used for training with a fine-tuned model. The machine learning-based analysis showed that the hyper-tuned model has empirically higher accuracy in classifying patients with depressive disorder, as evidenced by the set of interpretable experiments, and can be an effective solution for disease control.

Application of Aligned-UMAP to longitudinal biomedical studies

High-dimensional data analysis starts with projecting the data to low dimensions to visualize and understand the underlying data structure. Several methods have been developed for dimensionality reduction, but they are limited to cross-sectional datasets. The recently proposed Aligned-UMAP, an extension of the uniform manifold approximation and projection (UMAP) algorithm, can visualize high-dimensional longitudinal datasets. We demonstrated its utility for researchers to identify exciting patterns and trajectories within enormous datasets in biological sciences. We found that the algorithm parameters also play a crucial role and must be tuned carefully to utilize the algorithm's potential fully. We also discussed key points to remember and directions for future extensions of Aligned-UMAP. Further, we made our code open source to enhance the reproducibility and applicability of our work. We believe our benchmarking study becomes more important as more and more high-dimensional longitudinal data in biomedical research become available.

An Unsupervised Classification Algorithm for Heterogeneous Cryo-EM Projection Images Based on Autoencoders.

Heterogeneous three-dimensional (3D) reconstruction in single-particle cryo-electron microscopy (cryo-EM) is an important but very challenging technique for recovering the conformational heterogeneity of flexible biological macromolecules such as proteins in different functional states. Heterogeneous projection image classification is a feasible solution to solve the structural heterogeneity problem in single-particle cryo-EM. The majority of heterogeneous projection image classification methods are developed using supervised learning technology or require a large amount of a priori knowledge, such as the orientations or common lines of the projection images, which leads to certain limitations in their practical applications. In this paper, an unsupervised heterogeneous cryo-EM projection image classification algorithm based on autoencoders is proposed, which only needs to know the number of heterogeneous 3D structures in the dataset and does not require any labeling information of the projection images or other a priori knowledge. A simple autoencoder with multi-layer perceptrons trained in iterative mode and a complex autoencoder with residual networks trained in one-pass learning mode are implemented to convert heterogeneous projection images into latent variables. The extracted high-dimensional features are reduced to two dimensions using the uniform manifold approximation and projection dimensionality reduction algorithm, and then clustered using the spectral clustering algorithm. The proposed algorithm is applied to two heterogeneous cryo-EM datasets for heterogeneous 3D reconstruction. Experimental results show that the proposed algorithm can effectively extract category features of heterogeneous projection images and achieve high classification and reconstruction accuracy, indicating that the proposed algorithm is effective for heterogeneous 3D reconstruction in single-particle cryo-EM.

NKX3.1 immunohistochemistry and methylome profiling in mesenchymal chondrosarcoma: additional diagnostic value for a well-defined disease?

Mesenchymal chondrosarcoma (MCS) is a rare and highly aggressive tumour of soft tissue and bone that is defined by an underlying and highly specific fusion transcript involving HEY1 and NCOA2. Histologically, the tumours show a biphasic appearance consisting of an undifferentiated blue and round cell component as well as islands of highly differentiated cartilage. Particularly in core needle biopsies, the chondromatous component can be missed and the non-specific morphology and immunophenotype of the round cell component can cause diagnostic challenges. We applied NKX3.1 immunohistochemistry which was recently reported as a highly specific marker as well as methylome and copy number profiling to a set of 45 well characterised MCS cases to evaluate their potential diagnostic value. Methylome profiling revealed a highly distinct cluster for MCS. Notably, the findings were reproducible also when analysing the round cell and cartilaginous component separately. Furthermore, four outliers were identified by methylome profiling for which the diagnosis had to be revised. NKX3.1 immunohistochemistry showed positivity in 36% of tumours, the majority of which was rather focal and weak. Taken together, NKX3.1 expression showed a low sensitivity but a high specificity in our analysis. Methylome profiling on the other hand represents a sensitive, specific and reliable tool to support the diagnosis of MCS, particularly if only the round cell component is obtained in a biopsy and the diagnosis is not suspected. Furthermore, it can aid in confirming the diagnosis in case RNA sequencing for the HEY1::NCOA2 fusion transcript is not available.

Classifying FRB spectrograms using nonlinear dimensionality reduction techniques

ABSTRACT Fast radio bursts (FRBs) are mysterious astronomical phenomena, and it is still uncertain whether they consist of multiple types. In this study, we use two nonlinear dimensionality reduction algorithms – Uniform Manifold Approximation and Projection (UMAP) and t-distributed stochastic neighbour embedding (t-SNE) – to differentiate repeaters from apparently non-repeaters in FRBs. Based on the first Canadian Hydrogen Intensity Mapping Experiment (CHIME) FRB catalogue, these two methods are applied to standardized parameter data and image data from a sample of 594 sub-bursts and 535 FRBs, respectively. Both methods are able to differentiate repeaters from apparently non-repeaters. The UMAP algorithm using image data produces more accurate results and is a more model-independent method. Our result shows that in general repeater clusters tend to be narrowband, which implies a difference in burst morphology between repeaters and apparently non-repeaters. We also compared our UMAP predictions with the CHIME/FRB discovery of six new repeaters, the performance was generally good except for one outlier. Finally, we highlight the need for a larger and more complete sample of FRBs.

Population Characteristics of Spirlin Alburnoides bipunctatus (Bloch, 1782) in Serbia (Central Balkans): Implications for Conservation

The aim of this study was to evaluate the population characteristics of spirlin, Alburnoides bipunctatus, in Serbia, since this small fish species is facing a severe decline in its abundance and its natural habitats in Europe. We investigated the spirlin population dynamics, including size, age structure, growth pattern, mortality, and exploitation rate. Additionally, we used the Uniform Manifold Approximation and Projection approach with the Decision Tree algorithm to investigate the influence of different environmental parameters on the population parameters to unveil which factors shape the abundance and distribution of spirlin. The results showed that the highest values of production, abundance, and biomass were estimated in sites with low temperature, optimal pH, and well-oxygenated water, even though we found them in heavily polluted waters with extremely high values of conductivity. Moreover, we observed a pattern of migratory behavior, in which spirlin migrate upstream to sites at a higher altitude in early summer and autumn. Despite the putative vulnerability and high sensitivity of spirlin populations, our results showed that the species was abundant, occurring in altered habitats (due to pollution, climate change, anthropogenic pressure, etc.).