Dimensionality Reduction Research Articles

Learning model combined with data clustering and dimensionality reduction for short-term electricity load forecasting

Electric load forecasting is crucial in the planning and operating electric power companies. It has evolved from statistical methods to artificial intelligence-based techniques that use machine learning models. In this study, we investigate short-term load forecasting (STLF) for large-scale electricity usage datasets. We propose a new prediction model for STLF that combines data clustering and dimensionality reduction schemes to handle large-scale electricity usage data effectively. Here, we adapt k-means clustering for data clustering, kernel principal component analysis (kernel PCA), universal manifold approximation and projection (UMAP), and t-stochastic nearest neighbor (t-SNE) for dimensionality reduction. To verify the effectiveness of the proposed model, we extensively apply it to neural network-based models. We compare and analyze the performance of the proposed model with the comparisons using actual electricity usage data for 4710 households. Experimental results demonstrate that data clustering with dimensionality reduction can improve the performance of baseline models. As a result, the prediction accuracy of the proposed method outperforms those of the existing methods by 1.01–1.76 times for summer data and by 1.03–1.36 times for winter data in terms of mean absolute percentage error (MAPE).

Sequential and simultaneous distance-based dimension reduction

This paper introduces a method called Sequential and Simultaneous Distance-based Dimension Reduction (S2D2R) that performs simultaneous dimension reduction for a pair of random vectors based on distance covariance (dCov). Compared with sufficient dimension reduction (SDR) and canonical correlation analysis (CCA)-based approaches, S2D2R is a model-free approach that does not impose dimensional or distributional restrictions on variables and is more sensitive to nonlinear relationships. Theoretically, we establish a non-asymptotic error bound to guarantee the performance of S2D2R. Numerically, S2D2R performs comparable to or better than other state-of-the-art algorithms and is computationally faster. All codes of our S2D2R method can be found on GitHub https://github.com/Yijin911/S2D2R.git, including an R package named S2D2R.

Lidocaine Inhibits the Proliferation of Non-Small Cell Lung Cancer and Exerts Anti-Inflammatory Effects Through the TLR-9/MyD88/NF-κB Pathway.

Lung cancer represents a significant global health burden, with non-small cell lung cancer (NSCLC) being the most common subtype. The current standard of care for NSCLC has limited efficacy, highlighting the necessity for innovative treatment options. Lidocaine, traditionally recognized as a local anesthetic, has emerged as a compound with potential antitumor and anti-inflammatory capabilities. This study was designed to explore the impact of lidocaine on NSCLC cell proliferation and inflammation, particularly focusing on the Toll-like receptor 9 (TLR)-9/MyD88/NF-κB signaling pathway. A nude mice model of NSCLC was employed, with animals receiving lidocaine at different concentrations. In vitro experiments on A549 cells involved exposure to lidocaine, followed by assessment of cell viability, cytokine expression, and TLR-9 levels using the 3-(4,5-Dimethylthiazol-2-Yl)-2,5-Diphenyltetrazolium Bromide (MTT) assay, enzyme-linked immunosorbent assay, and Quantitative Real-time polymerase chain reaction (qPCR). Protein levels were evaluated via Western blot analysis. Additionally, A549 cells were transfected with a TLR-9-overexpressing lentivirus to dissect the role of TLR-9 in lidocaine's mechanism of action. Treatment with lidocaine led to a significant reduction in tumor dimensions and a decrease in inflammatory marker expression in the NSCLC mouse model. In cellular assays, lidocaine effectively suppressed A549 cell proliferation and the expression of inflammatory cytokines. The overexpression of TLR-9 partially negated the suppressive effects of lidocaine, underscoring the significance of the TLR-9/MyD88/NF-κB pathway in mediating lidocaine's effects. Lidocaine's inhibitory effects on NSCLC cell proliferation and its anti-inflammatory mechanisms are mediated through the TLR-9/MyD88/NF-κB pathway. The study's results offer promising insights into the therapeutic potential of lidocaine in NSCLC and pave the way for future investigations into its application in cancer therapy.

Facile Access to Highly Efficient 3D Printing Using Robust Self-Healing CDs/Polymer Hybrids.

3D printing efficiency, as a key indicator of additive manufacturing technology, directly affects its competitiveness in rapid prototyping, small batch production, and even large-scale industrial applications. Compared with traditional manufacturing methods, the high efficiency of 3D printing is often considered a bottleneck, hindering its application across various fields. Herein, a versatile and efficient strategy is proposed, namely, the dimensional reduction printing (DRP) process, to break the obstacle of high efficiency of 3D printing. Specifically, the self-healing CDs/PMMA nanocomposites were constructed via introducing carbon dots (CDs) synthesized by microfluidics into poly(methyl methacrylate) (PMMA) materials. Under the assistance of abundant hydrogen bonding and the entanglement of polymer chains, the fabricated CDs/PMMA nanocomposites exhibited outstanding self-healing properties, which can be utilized as ideal printing inks to achieve a highly efficient 3D printing process through assembling the printed simple 1D and 2D components into complex 3D models. We firmly believe that the DRP strategy opens up an idea for the design of a 3D printing process and points out the direction for the meaningful application of 3D printing technologies.

Multidimensional scaling improves distance-based clustering for microbiome data.

Clustering patients into subgroups based on their microbial compositions can greatly enhance our understanding of the role of microbes in human health and disease etiology. Distance-based clustering methods, such as partitioning around medoids (PAM), are popular due to their computational efficiency and absence of distributional assumptions. However, the performance of these methods can be suboptimal when true cluster memberships are driven by differences in the abundance of only a few microbes, a situation known as the sparse signal scenario. We demonstrate that classical multidimensional scaling (MDS), a widely used dimensionality reduction technique, effectively denoises microbiome data and enhances the clustering performance of distance-based methods. We propose a two-step procedure that first applies MDS to project high-dimensional microbiome data into a low-dimensional space, followed by distance-based clustering using the low-dimensional data. Our extensive simulations demonstrate that our procedure offers superior performance compared to directly conducting distance-based clustering under the sparse signal scenario. The advantage of our procedure is further showcased in several real data applications. The R package MDSMClust is available at https://github.com/wxy929/MDS-project. gchen25@wisc.edu. Supplementary data are available at Bioinformatics online.

Re-scaling images using a SVD-based approach

There are many instances in computer science where computational operations must be performed on matrices of different sizes. In the field of machine learning, particularly when interpreting images, it is often necessary to resize and re-scale images to achieve higher resolutions. While there are various methods for this, they typically involve fitting a simple linear or cubic model to scale the image. The singular value decomposition (SVD) is a powerful tool for dimension reduction and projection. Our proposed state-of-the-art method leverages the capabilities of SVD to create a technique for re-scaling images. This method is primarily based on modifications of the eigenvectors derived from SVD. Previous work has shown that by editing only these Eigenvectors, it is possible to minimize error propagation through the images. When applied to several well-known image processing tasks, it is possible to scale an image with reduced error compared to the current state-of-the-art methods. Additionally, we show that our method can improve the results of machine learning approaches.

Uncovering temporal patterns in visualizations of high-dimensional data

With the increasing availability of high-dimensional data, analysts often rely on exploratory data analysis to understand complex data sets. A key approach to exploring such data is dimensionality reduction, which embeds high-dimensional data in two dimensions to enable visual exploration. However, popular embedding techniques, such as t-SNE and UMAP, typically assume that data points are independent. When this assumption is violated, as in time-series data, the resulting visualizations may fail to reveal important temporal patterns and trends. To address this, we propose a formal extension to existing dimensionality reduction methods that incorporates two temporal loss terms that explicitly highlight temporal progression in the embedded visualizations. Through a series of experiments on both synthetic and real-world datasets, we demonstrate that our approach effectively uncovers temporal patterns and improves the interpretability of the visualizations. Furthermore, the method improves temporal coherence while preserving the fidelity of the embeddings, providing a robust tool for dynamic data analysis.

Radiogenomics and machine learning predict oncogenic signaling pathways in glioblastoma

BackgroundGlioblastoma (GBM) is a highly aggressive brain tumor associated with a poor patient prognosis. The survival rate remains low despite standard therapies, highlighting the urgent need for novel treatment strategies. Advanced imaging techniques, particularly magnetic resonance imaging (MRI), are crucial in assessing GBM. Disruptions in various oncogenic signaling pathways, such as Receptor Tyrosine Kinase (RTK)-Ras-Extracellular signal-regulated kinase (ERK) signaling, Phosphoinositide 3- Kinases (PI3Ks), tumor protein p53 (TP53), and Neurogenic locus notch homolog protein (NOTCH), contribute to the development of different tumor types, each exhibiting distinct morphological and phenotypic features that can be observed at a microscopic level. However, identifying genetic abnormalities for targeted therapy often requires invasive procedures, prompting exploration into non-invasive approaches like radiogenomics. This study explores the utility of radiogenomics and machine learning (ML) in predicting these oncogenic signaling pathways in GBM patients.MethodsWe collected post-operative MRI scans (T1w, T1c, FLAIR, T2w) from the BRATS-19 dataset, including scans from patients with both GBM and LGG, linked to genetic and clinical data via TCGA and CPTAC. Signaling pathway data was manually extracted from cBioPortal. Radiomic features were extracted from four MRI modalities using PyRadiomics. Dimensionality reduction and feature selection were applied and Data imbalance was addressed with SMOTE. Five ML models were trained to predict signaling pathways, with Grid Search optimizing hyperparameters and 5-fold cross-validation ensuring unbiased performance. Each model’s performance was evaluated using various metrics on test data.ResultsOur results showed a positive association between most signaling pathways and the radiomic features derived from MRI scans. The best models achieved high AUC scores, namely 0.7 for RTK-RAS, 0.8 for PI3K, 0.75 for TP53, and 0.4 for NOTCH, and therefore, demonstrated the potential of ML models in accurately predicting oncogenic signaling pathways from radiomic features, thereby informing personalized therapeutic approaches and improving patient outcomes.ConclusionWe present a novel approach for the non-invasive prediction of deregulation in oncogenic signaling pathways in glioblastoma (GBM) by integrating radiogenomic data with machine learning models. This research contributes to advancing precision medicine in GBM management, highlighting the importance of integrating radiomics with genomic data to understand tumor behavior and treatment response better.

Using space-filling curves and fractals to reveal spatial and temporal patterns in neuroimaging data

Objective. Magnetic resonance imaging (MRI), functional MRI (fMRI) and other neuroimaging techniques are routinely used in medical diagnosis, cognitive neuroscience or recently in brain decoding. They produce three- or four-dimensional scans reflecting the geometry of brain tissue or activity, which is highly correlated temporally and spatially. While there exist numerous theoretically guided methods for analyzing correlations in one-dimensional data, they often cannot be readily generalized to the multidimensional geometrically embedded setting.Approach. We present a novel method, Fractal Space-Curve Analysis (FSCA), which combines Space-Filling Curve (SFC) mapping for dimensionality reduction with fractal Detrended Fluctuation Analysis. We conduct extensive feasibility studies on diverse, artificially generated data with known fractal characteristics: the fractional Brownian motion, Cantor sets, and Gaussian processes. We compare the suitability of dimensionality reduction via Hilbert SFC and a data-driven alternative. FSCA is then successfully applied to real-world MRI and fMRI scans.Main results. The method utilizing Hilbert curves is optimized for computational efficiency, proven robust against boundary effects typical in experimental data analysis, and resistant to data sub-sampling. It is able to correctly quantify and discern correlations in both stationary and dynamic two-dimensional images. In MRI Alzheimer's dataset, patients reveal a progression of the disease associated with a systematic decrease of the Hurst exponent. In fMRI recording of breath-holding task, the change in the exponent allows distinguishing different experimental phases.Significance. This study introduces a robust method for fractal characterization of spatial and temporal correlations in many types of multidimensional neuroimaging data. Very few assumptions allow it to be generalized to more dimensions than typical for neuroimaging and utilized in other scientific fields. The method can be particularly useful in analyzing fMRI experiments to compute markers of pathological conditions resulting from neurodegeneration. We also showcase its potential for providing insights into brain dynamics in task-related experiments.

Machine Learning Based Finger Print Analysis for Gender Detection: A Review

Gender detection using fingerprint biometrics has emerged as a promising area of research due to its non-intrusive nature and potential applications in biometric identification systems. The procedure can involve multiple steps are the size of finger print and their ridge pattern, minutiae point, machine learning and image processing and accuracy and limitations. This review explores the effectiveness of machine learning techniques for gender classification based on fingerprint patterns, emphasizing the role of advanced classification algorithms and feature extraction methods. Machine learning is crucial for gender detection since it classifies fingerprint patterns and biometric information using models like Convolutional Neural Networks (CNN) and Support Vector Machines (SVM). To identify traits unique to a gender, such as ridge density and minutiae points, these algorithms are trained using labelled datasets. Compared to manual procedures, these models are more effective at handling high-dimensional data and identifying subtle gender-related patterns. Although hybrid models like CNN-DNN and AlexNet further increase classification precision, Convolutional Neural Networks (CNNs) are especially effective due to their automatic feature extraction capabilities. Despite their effectiveness, factors like as picture resolution, demographic balance, and dataset heterogeneity might affect performance, highlighting the need for carefully selected datasets and improved model designs. A structured comparative analysis of multiple studies reveals the impact of various datasets, feature types, and model architectures on classification accuracy and reliability. The findings suggest that deep learning models often outperform traditional classifiers, while dimensionality reduction and hybrid approaches can further enhance performance. However, challenges such as dataset imbalances, limited diversity, and susceptibility to low-quality fingerprint data remain prominent barriers to achieving consistent results. This review also outlines key limitations observed across the studies and provides recommendations for future research, including the need for more diverse datasets and optimized classification frameworks. This study aims to improve fingerprint feature extraction for gender detection, reduce processing costs, fix dataset imbalances, and increase classification accuracy. By stating the objective, the scope and objectives of each investigation are made clear. The generalizability of machine learning models is significantly impacted by the amount, variety, and quality of the dataset. The analysis aims to support the development of more accurate, inclusive, and scalable fingerprint-based gender detection systems.

The Predictive Value of a Nomogram Based on Ultrasound Radiomics, Clinical Factors, and Enhanced Ultrasound Features for Central Lymph Node Metastasis in Papillary Thyroid Microcarcinoma.

This study aims to establish and validate an ultrasound radiomics nomogram for preoperative prediction of central lymph node metastasis in papillary thyroid microcarcinoma (PTMC) before operation. A retrospective analysis conducted on ultrasonic images and clinical features derived from 288 PTMC patients, who were divided into training cohorts (n = 201) and validating cohorts (n = 87) in a ratio of 7:3 base on the principle of random allocation. Radiomics features were extracted from the PTMC patients after ultrasonic examination, followed by dimension reduction and characteristic selection to construct the radiomics score (Radscore) using LASSO regression analysis. Subsequently, the models, ultrasound features plus clinical features (US-Clin), radiomics score model, and combined model of clinical features plus ultrasound features and Radscore (Combined-model) were built through multi-factor logistic regression analysis. After that, the nomograms were developed for visualization and presentation of these models. The discriminative power, calibration and clinical utility of the nomogram models were evaluated in the training and validating cohorts. The Radscore model comprised 12 carefully selected features. The independent risk factors for conventional ultrasound features and clinical features of PTMC in predicting CLNM included age <45 years, tumor envelope invasion, male gender and presence of microcalcifications, while the enhanced ultrasound features risk factor was extrathyroidal expansion. The combined model showed good performance in predicting PTMC CLNM, with AUCs of 0.921 and 0.889 in the training and validating cohorts, respectively. And DCA based on the prediction model showed good clinical utility. The nomogram developed based on preoperative clinical data, ultrasound features, and Radscore of PTMC patients can more accurately predict central lymph node metastasis (CLNM) in PTMC patients. However, it needs to be validated for clinical applicability in multicenter studies with larger sample sizes and combined with genomic mutation analyses of the tumors.

Hyperspectral Band Selection Method Based on Global Partition Clustering

Band selection is an important step in the dimensionality reduction processing of hyperspectral images and is highly important for eliminating redundant spectral information and reducing computational costs. In recent years, band selection methods based on ordered partition have been widely used in the dimensionality reduction processing of hyperspectral images. However, most of the methods use coarse and fine partition for band subspace partition, so that the partition results are affected by equal interval partition. Furthermore, existing methods usually select a representative band in each band subspace but do not consider the relationship between the output bands in the selection process, resulting in a certain degree of redundancy between the output bands. To solve the above problems, we propose a band selection method based on global partition clustering that contains band subspace partition and band selection. The band subspace partition is based on coarse and fine partition and the similarity-based ranking-structural similarity method, which divides the band space into band subspaces according to the relationship between the number of bands in the hyperspectral image and the number of selected bands. Band selection is based on the sequential forward selection method, which iteratively selects one band as the output band in each band subspace. The proposed method makes two main contributions. Firstly, this method avoids the negative effect of equal interval partition on the results. Secondly, this method fully considers the relationship between the selected bands and the bands to be selected. The effectiveness of the proposed method is demonstrated via ablation and comparison experiments on three publicly available datasets. Comparison experiments show that the classification accuracy of the proposed method can exceed the accuracy of the comparison methods.

Robust Dimensionality Reduction: A Bootstrap-Based Evaluation of PCA with Applications in Nutritional and Environmental Sciences

The complex structures of vast amounts of data provide a considerable challenge to researchers. Dimensional reduction methods are reliable and transform high-dimensional data into lower-dimensional representations while preserving most original information. Principal Component Analysis (PCA) is a commonly used approach for dimensionality reduction that transforms data into a lower-dimensional space while preserving important information. The variability within the sample may affect the stability and reliability of PCA results. The constraint of existing approaches compromises the accuracy of PCA stability assessments in practical data scenarios. These methodologies frequently depend on linear assumptions and encounter difficulties when addressing high-dimensional data. This study used the bootstrap method to assess the stability of PCA by assessing the variability of eigenvalues and principal components over several bootstrap iterations. We evaluate how stability metrics, particularly confidence intervals for eigenvalues and the proportion of variance clarified, can assist in determining the optimal number of principal components. The results indicate that the bootstrap provides a helpful framework for evaluating the robustness of PCA and guiding informed decisions on dimensionality reduction in many applications, including data compression, visualization, and classification. Moreover, the results illustrate the efficacy of this method in enhancing the reliability and interpretability of PCA findings among distinct data-driven research endeavors. This study enhances understanding of how principal component analysis (PCA) tackles data unpredictability while delivering valuable insights for professionals in several disciplines.

Reverse-engineering the FLT3-PI3K/AKT axis to enhance TILs function and improve prognosis in ovarian and cervical cancers

BackgroundOvarian cancers (OC) and cervical cancers (CC) have poor survival rates. Tumor-infiltrating lymphocytes (TILs) play a pivotal role in prognosis, but shared immune mechanisms remain elusive.MethodsWe integrated single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics (ST) to explore immune regulation in OC and CC, focusing on the PI3K/AKT pathway and FLT3 as key modulators. Seurat and Harmony were employed for batch correction and dimensionality reduction. FLT3 expression was mapped with spatial data from 10 × Genomics.ResultsFLT3, identified as a regulator through the PI3K/AKT pathway, showed positive correlations with T cells, NK cells, and B cells. FLT3-high regions exhibited increased immune infiltration, particularly in CC, enhancing survival outcomes.ConclusionThis study provides the first spatially resolved evidence of FLT3's immune-modulatory role in OC and CC, positioning it as a promising immunotherapeutic target. FLT3-targeted strategies may offer new options for patients resistant to conventional therapies.

Diffusive topology preserving manifold distances for single-cell data analysis

Manifold learning techniques have emerged as crucial tools for uncovering latent patterns in high-dimensional single-cell data. However, most existing dimensionality reduction methods primarily rely on 2D visualization, which can distort true data relationships and fail to extract reliable biological information. Here, we present DTNE (diffusive topology neighbor embedding), a dimensionality reduction framework that faithfully approximates manifold distance to enhance cellular relationships and dynamics. DTNE constructs a manifold distance matrix using a modified personalized PageRank algorithm, thereby preserving topological structure while enabling diverse single-cell analyses. This approach facilitates distribution-based cellular relationship analysis, pseudotime inference, and clustering within a unified framework. Extensive benchmarking against mainstream algorithms on diverse datasets demonstrates DTNE's superior performance in maintaining geodesic distances and revealing significant biological patterns. Our results establish DTNE as a powerful tool for high-dimensional data analysis in uncovering meaningful biological insights.

Single-cell sequencing uncovers the mechanistic role of DAPK1 in glioma and its diagnostic and prognostic implications

BackgroundWe conducted an investigation into the characteristics of single-cell differentiation data in gliomas, with a focus on developing DAPK1-based prognostic markers to predict patient outcomes. Dysregulated expression of DAPK1 has been associated with the invasive behavior of various malignancies, including gliomas. However, the precise role and underlying mechanisms of DAPK1 in gliomas remain inadequately understood.MethodsWe performed analyses on RNA-seq and microarray datasets from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO), in addition to single-cell RNA sequencing (scRNA-seq) data from glioma patients available in GEO. Utilizing the Seurat R package, we identified gene clusters associated with survival from the scRNA-seq data. Prognostic models were developed using LASSO and stepwise regression algorithms. Furthermore, we assessed the predictive potential of these genes within the immune microenvironment and their relevance in immunotherapy contexts.ResultsOur scRNA-seq data analysis revealed 32 distinct cell clusters corresponding to 10 cell types. Through dimensionality reduction and clustering, we identified three glial cell subpopulations based on their differentiation trajectories. DAPK1, serving as a marker gene for the terminal subpopulation, exhibited an association with poor prognosis.ConclusionsDAPK1-based prognostic models show promise for accurately predicting outcomes in glioblastoma and glioma. An in-depth examination of DAPK1’s specific mechanisms in glioblastoma could elucidate its role in immunotherapy response. Targeting the DAPK1 gene may offer therapeutic benefits for glioma patients.

STAT4 gene polymorphism may be associated with microscopic polyangiitis susceptibility in a Chinese Guangxi population: A case-control analysis based on propensity score matching.

Microscopic polyangiitis (MPA) is a severe multisystem autoimmune disease featured by small-vessel vasculitis with few or no immune complex, also has a significant genetic predisposition. Growing evidence has confirmed that STAT4 gene is tightly associated with multiple autoimmune diseases, but its contribution to MPA onset is still elusive. The aim was to investigated the association between STAT4 gene polymorphisms (rs7572482, rs7574865 and rs12991409) and MPA susceptibility in a Guangxi population of China. 260 MPA patients and 295 healthy adult volunteers were selected, 1:1 propensity score matching (PSM) was performed to control potential confounding variables, then 199 MPA patients and 199 healthy adult volunteers matched in gender, ethnicity and age were included in this study. High-throughput sequencing and multiplex PCR were applied to detect the target STAT4 SNPs. SHEsis and SNPstats were used to evaluated the allele frequency, genotype frequency, linkage disequilibrium (LD), haplotype, and the association between SNPs and the MPA susceptibility in multiple genetic models. SNP-SNP interactions were explored based on generalized multifactor dimensionality reduction (GMDR) algorithm. Some clinical indicators, such as renal pathology and therapeutic effects, were collected and compared. The allele and genotype frequencies of rs7574865 displayed significant diversities between case group and control group (p < 0.05). Strong LD was found between rs7572482 and rs12991409 (D'=0.9). The haplotype GGT was related to a reduced risk of MPA (OR = 0.661, 95 %CI: 0.469-0.931, p = 0.017), and haplotype GTT might perform an increased risk of MPA (OR = 1.922, 95 %CI: 1.225-3.015, p = 0.004). Rs7574865 polymorphism was associated with an increased risk of MPA in codominant model (OR:2.03; p = 0.0093), dominant model (OR: 1.88p = 0.0023), and overdominant model (OR:1.57; p = 0.027). In Han and male subgroups, rs7574865 polymorphism dramatically increased the MPA risk. GMDR suggested that STAT4 rs7574865 and PTPN22 rs3811021 composed the most risk combinations (p = 0.0010). Moreover, renal pathology, Birmingham vasculitis activity score (BVAS), and alanine aminotransferase (ALT) might be linked with STAT4 gene polymorphisms (p < 0.05). The genetic polymorphism of STAT4 may be associated with MPA susceptibility and renal pathological classification in Chinese Guangxi population; the T allele of rs7574865 may be an important risk factor for MPA.

Interpretability of fingerprint presentation attack detection systems: a look at the “representativeness” of samples against never-seen-before attacks

Nowadays, fingerprint Presentation Attack Detection systems (PADs) are primarily based on deep learning architectures subjected to massive training. However, their performance decreases to never-seen-before attacks. With the goal of contributing to explaining this issue, we hypothesized that this limited ability to generalize is due to the lack of "representativeness" of the samples available for the PAD training. "Representativeness" is treated here from a geometrical perspective: the spread of samples into the feature space, especially near the decision boundaries. In particular, we explored the possibility of adopting three-dimensionality reduction methods to make the problem affordable through visual inspection. These methods enable visual inspection and interpretation by projecting data into two-dimensional spaces, facilitating the identification of weak areas in the decision regions estimated after the training phase. Our analysis delineates the benefits and drawbacks of each dimensionality reduction method and leads us to make substantial recommendations in the crucial phase of the training design.

How complex are galaxies? A non-parametric estimation of the intrinsic dimensionality of wide-band photometric data

Abstract Galaxies are complex objects, yet the number of independent parameters to describe them remains unknown. We present here a non-parametric method to estimate the intrinsic dimensionality of large datasets. We apply it to wide-band photometric data drawn from the COSMOS2020 catalogue and a comparable mock catalogue from the Horizon-AGN simulation. Our galaxy catalogues are limited in signal-to-noise ratio in all optical and NIR bands. Our results reveal that most of the variance in the wide-band photometry of this galaxy sample can be described with at most 4.3 ± 0.5 independent parameters for star-forming galaxies and 2.9 ± 0.2 for passive ones, both in the observed and simulated catalogues. We identify one of these parameters to be noise-driven, and recover that stellar mass and redshift are two key independent parameters driving the magnitudes. Our findings support the idea that wide-band photometry does not provide more than one additional independent parameter for star-forming galaxies. Although our sample is not mass-limited and may miss some passive galaxies due to our cut in SNR, our work suggests that dimensionality reduction techniques may be effectively used to explore and analyse wide-band photometric data, provided the used latent space is at least four-dimensional.

Classification of Fibro-osseous Tumors in the Craniofacial Bones using DNA Methylation and Copy Number Alterations.

Fibro-osseous tumors of the craniofacial bones are a heterogeneous group of lesions comprising cemento-osseous dysplasia (COD), cemento-ossifying fibroma (COF), juvenile trabecular ossifying fibroma (JTOF), psammomatoid ossifying fibroma (PsOF), fibrous dysplasia (FD), and low-grade osteosarcoma (LGOS) with overlapping clinicopathological features. However, their clinical behavior and treatment differ significantly, underlining the need for accurate diagnosis. Molecular diagnostic markers exist for subsets of these tumors, including GNAS mutations in FD, SATB2 fusions in PsOF, mutations involving the RAS-MAPK signaling pathway in COD, and MDM2 amplification in LGOS. Since DNA methylation and copy number profiling are well established for the classification of central nervous system tumors, our aim was to investigate whether this tool might be used as well for classifying fibro-osseous tumors in the craniofacial bones. We collected a well-characterized, multicenter cohort with available molecular data, including COD (n = 20), COF (n = 13), JTOF (n = 10), PsOF (n = 25), FD (n = 23), LGOS (n = 4), and high-grade osteosarcoma (HGOS; n = 11). Genome-wide DNA methylation and copy number variation data were generated using the Illumina Infinium Methylation EPIC array interrogating >850 000 CpG sites. DNA methylation profiling yielded evaluable results in 73/106 tumors, including 6 CODs, 12 COFs, 6 JTOFs, 19 PsOFs, 18 FDs, 2 LGOSs, and 10 HGOSs. Unsupervised clustering and dimensionality reduction (Uniform Manifold Approximation and Projection) revealed that FD, extragnatic PsOF, and HGOS formed distinct clusters. Surprisingly, COD, COF, JTOF, and mandibular PsOF clustered together, apart from other craniofacial bone tumors. LGOS did not form a distinct cluster, likely due to the low number of cases. Copy number analysis revealed that FD, COD, COF, JTOF, and PsOF were typically characterized by flat copy number profiles compared to LGOS with gains of chromosome 12 and HGOS with multiple heterogeneous copy number alterations. In conclusion, using DNA methylation and copy number profiles, benign fibro-osseous tumors can be separated from low-grade and high-grade osteosarcomas in the craniofacial bones, which is of diagnostic value in challenging cases with overlapping clinicopathological features.