Non-negative Matrix Factorization Algorithm Research Articles

Advancing Auditory Processing by Detecting Frequency-Following Responses Through a Specialized Machine Learning Model.

In this study, we explore the feasibility and performance of detecting scalp-recorded frequency-following responses (FFRs) with a specialized machine learning (ML) model. By leveraging the strengths of feature extraction of the source separation non-negative matrix factorization (SSNMF) algorithm and its adeptness in handling limited training data, we adapted the SSNMF algorithm into a specialized ML model with a hybrid architecture to enhance FFR detection amidst background noise. We recruited 40 adults with normal hearing and evoked their scalp recorded FFRs using the English vowel/i/with a rising pitch contour. The model was trained on FFR-present and FFR-absent conditions, and its performance was evaluated using sensitivity, specificity, efficiency, false-positive rate, and false-negative rate metrics. This study revealed that the specialized SSNMF model achieved heightened sensitivity, specificity, and efficiency in detecting FFRs as the number of recording sweeps increased. Sensitivity exceeded 80% at 500 sweeps and maintained over 89% from 1000 sweeps onwards. Similarly, specificity and efficiency also improved rapidly with increasing sweeps. The progressively enhanced sensitivity, specificity, and efficiency of this specialized ML model underscore its practicality and potential for broader applications. These findings have immediate implications for FFR research and clinical use, while paving the way for further advancements in the assessment of auditory processing.

Mapping emission heterogeneity in layered halide perovskites using cathodoluminescence

Recent advancements in the fabrication of layered halide perovskites and their subsequent modification for optoelectronic applications have ushered in a need for innovative characterisation techniques. In particular, heterostructures containing multiple phases and consequently featuring spatially defined optoelectronic properties are very challenging to study. Here, we adopt an approach centered on cathodoluminescence, complemented by scanning electron microscopy coupled with energy-dispersive x-ray spectroscopy analysis. Cathodoluminescence enables assessment of local emission variations by injecting charges with a nanometer-scale electron probe, which we use to investigate emission changes in three different systems: PEA2PbBr4, PEA2PbI4 and lateral heterostructures of the two, fabricated via halide substitution. We identify and map different emission bands that can be correlated with local chemical composition and geometry. One emission band is characteristic of bromine-based halide perovskite, while the other originates from iodine-based perovskite. The coexistence of these emissions bands in the halide-substituted sample confirms the formation of lateral heterostructures. To improve the signal quality of the acquired data, we employed multivariate analysis, specifically the non-negative matrix factorization algorithm, on both cathodoluminescence and compositional datasets. The resulting understanding of the halide replacement process and identification of potential synergies in the optical properties will lead to optimised architectures for optoelectronic applications.

Inter- and intra-hypergraph regularized nonnegative matrix factorization with hybrid constraints

The accurate low rank representation of high-dimensional data learned by the manifold regularized nonnegative matrix factorization framework is effective in data clustering. In previous work, this work has mainly been solved in the way of similarity matrix induction. To further increase the efficacy of low-rank representations, we propose a novel semi-supervised non-negative matrix factorization (NMF) model in this study called inter- and intra-hypergraph regularized non-negative matrix factorization with hybrid constraints (IGNMFC). IGNMFC constructs intra-hypergraph regularization and intra-hypergraph regularization by hypergraph learning, which can precisely induce high-dimensional data to map toward low-dimensional. Moreover, hybrid constraints are introduced to improve the exclusivity and sparsity of low-dimensional representations, and the result accounts for the benefit of this method in learning distinguishable subspace representations. Finally, IGNMFC is transformed into an optimal problem and an efficient iteration rule is proposed. Experiments on several datasets demonstrate that the proposed method outperforms the state-of-the-art NMF algorithms, and can achieve at least 7.9%∼15% accuracy improvement in most cases.

Single-cell dissection reveals the role of aggrephagy patterns in tumor microenvironment components aiding predicting prognosis and immunotherapy on lung adenocarcinoma.

Lung adenocarcinoma (LUAD) is one of the leading malignant cancers. Aggrephagy plays a critical role in key genetic events for various cancers; yet, how aggrephagy functions within the tumor microenvironment (TME) in LUAD remains to be elucidated. In this study, by sequential non-negative matrix factorization (NMF) algorithm, pseudotime analysis, cell-cell interaction analysis, and SCENIC analysis, we have shown that aggrephagy genes demonstrated various patterns among different cell types in LUAD TME. LUAD and Immunotherapy cohorts from public repository were used to determine the prognosis and immune response of aggrephagy TME subtypes. The aggrephagy-deprived prognostic score (ADPS) was quantified based on machine learning algorithms. The cancer-associated fibroblasts (CAFs), tumor-associated macrophages (TAMs), and CD8+ T cells have various aggrephagy patterns, which enhance the intensity of intercellular communication and transcription factor activation. Furthermore, based on the signatures of the newly defined aggrephagy cell subtypes and expression profiles of large cohorts in LUAD patients, we determine that DYNC1I2+CAF-C1, DYNLL1+CAF-C2, PARK7+CAF-C3, VIM+Mac-C1, PARK7+Mac-C2, VIM+CD8+T_cells-C1, UBA52+CD8+T_cells-C2, TUBA4A+CD8+T_ cells-C3, and TUBA1A+CD8+T_cells-C4 are crucial prognostic factors for LUAD patients. The developed ADPS could predict survival outcomes and immunotherapeutic response across ten cohorts (n = 1838), and patients with low ADPS owned a better prognosis, lower genomic alterations, and are more sensitive to immunotherapy. Meanwhile, based on PRISM, CTRP, and CMAP databases, PLK inhibitor BI-2536, may be a potential agent for patients with high ADPS. Taken together, our novel and systematic single-cell analysis has revealed the unique role of aggrephagy in remodeling the TME of LUAD. As a newly demonstrated biomarker, the ADPS facilitates the clinical management and individualized treatment of LUAD.

Incremental semi-supervised graph learning NMF with block-diagonal

Non-negative matrix factorization (NMF), as a good data dimensionality reduction method, is widely used in the field of image recognition. Incremental non-negative matrix factorization (INMF) as an improvement solves the problem of inefficiency caused by repeated running of data samples during online learning. However, in the traditional incremental non-negative matrix factorization algorithm, the newly added training samples do not contain label information. Some samples may be marked in both the initial sample and the new training sample in the real application scenario. In order to make full use of the label information carried by the dataset, in this paper, we propose a semi-supervised non-negative matrix factorization model for batch incremental data, incremental semi-supervised graph learning NMF with block diagonal (ISGDNMF). The model is divided into three cases according to the degree of label-carrying for the new batch data: all label-carrying, no label-carrying, and partial label-carrying. The label information is also used to add a diagonal structure to the coefficient matrix, which makes it possible to have stronger discriminatory ability and to distinguish different classes of images more easily. And graph regularization information is added in order to maintain the spatial-geometric structure of the data. Experiments on six image datasets show that this algorithm has superior performance relative to the other seven NMF-based algorithms.

A breast cancer classification and immune landscape analysis based on cancer stem-cell-related risk panel

This study sought to identify molecular subtypes of breast cancer (BC) and develop a breast cancer stem cells (BCSCs)-related gene risk score for predicting prognosis and assessing the potential for immunotherapy. Unsupervised clustering based on prognostic BCSC genes was used to determine BC molecular subtypes. Core genes of BC subtypes identified by non-negative matrix factorization algorithm (NMF) were screened using weighted gene co-expression network analysis (WGCNA). A risk model based on prognostic BCSC genes was constructed using machine learning as well as LASSO regression and multivariate Cox regression. The tumor microenvironment and immune infiltration were analyzed using ESTIMATE and CIBERSORT, respectively. A CD79A+CD24-PANCK+-BCSC subpopulation was identified and its spatial relationship with microenvironmental immune response state was evaluated by multiplexed quantitative immunofluorescence (QIF) and TissueFAXS Cytometry. We identified two distinct molecular subtypes, with Cluster 1 displaying better prognosis and enhanced immune response. The constructed risk model involving ten BCSC genes could effectively stratify patients into subgroups with different survival, immune cell abundance, and response to immunotherapy. In subsequent QIF validation involving 267 patients, we demonstrated the existence of CD79A+CD24-PANCK+-BCSC in BC tissues and revealed that this BCSC subtype located close to exhausted CD8+FOXP3+ T cells. Furthermore, both the densities of CD79A+CD24-PANCK+-BCSCs and CD8+FOXP3+T cells were positively correlated with poor survival. These findings highlight the importance of BCSCs in prognosis and reshaping the immune microenvironment, which may provide an option to improve outcomes for patients.

Single-cell histone chaperones patterns guide intercellular communication of tumor microenvironment that contribute to breast cancer metastases

BackgroundHistone chaperones (HCs) are crucial for governing genome stability and gene expression in multiple cancers. However, the functioning of HCs in the tumor microenvironment (TME) is still not clearly understood.MethodsSelf-tested single-cell RNA-seq data derived from 6 breast cancer (BC) patients with brain and liver metastases were reanalyzed by nonnegative matrix factorization (NMF) algorithm for 36 HCs. TME subclusters were observed with BC and immunotherapy public cohorts to assess their prognosis and immune response. The biological effect of HSPA8, one of the HCs, was verified by transwell assay and wound-healing assays.ResultsCells including fibroblasts, macrophages, B cells, and T cells, were classified into various subclusters based on marker genes. Additionally, it showed that HCs might be strongly associated with biological and clinical features of BC metastases, along with the pseudotime trajectory of each TME cell type. Besides, the results of bulk-seq analysis revealed that TME cell subclusters mediated by HCs distinguished significant prognostic value for BC patients and were relevant to patients’ immunotherapy responses, especially for B cells and macrophages. In particular, CellChat analysis exhibited that HCs-related TME cell subclusters revealed extensive and diverse interactions with malignant cells. Finally, transwell and wound-healing assays exhibited that HSPA8 deficiency inhibited BC cell migration and invasion.ConclusionsCollectively, our study first dissected HCs-guided intercellular communication of TME that contribute to BC metastases.

Prior-free mixed speckle simplex separation strategy for multi-object imaging through thin scattering media beyond the optical memory effect

The optical technique of imaging through scattering media based on the optical memory effect (OME) sustains a limited field-of-view (FOV). Therefore, a prior-free imaging method is proposed to reconstruct multiple objects through the scattering media beyond the OME range. Based on the mixed speckle simplex separation strategy designed in this method, separating speckles of sub-objects is simplified as seeking the vertices of the mixed speckle simplex. An effective initial analysis of the mixed speckle simplex constructed by random intensity modulation is provided by vertex component analysis. The exact speckles of sub-objects are then separated by the specially designed non-negative matrix factorization algorithm. The multiple hidden objects can be recovered from the separated speckles respectively. The feasibility and imaging effects of the proposed method have been demonstrated via experiments. Multi-object imaging through the scattering media beyond at least four times the OME range has been realized. This work effectively advances speckle separation strategies to enlarge the limited FOV for imaging through scattering media.

Introgression as an Important Driver of Geographic Genetic Differentiation within European White Oaks

The genetic composition of 5797 white oaks assigned in forest inventories as Quercus robur (3342), Quercus petraea (2090), Quercus pubescens (170), or as unspecified Quercus. spp. (195) sampled all over Europe were genotyped at 355 nuclear SNPs and 28 maternally inherited SNPs of the chloroplast and mitochondria. The sampling had a focus on Central and Eastern Europe, as well as the Black Sea and Caucasus region. Using a sparse nonnegative matrix factorization (snmf) algorithm, the nuclear genetic information was best represented by K = 4 different genetic clusters, whereas a principal component analysis visualized three different groups. The snmf run with K = 3 corresponded, for most individuals with the assignment in the forest inventories, to the three different species. The majority of the samples (88%) had an admixture coefficient q > 0.8 for one of the three species clusters, underlining the species integrity with a minor level of admixture. In contrast to Q. petraea, Q. robur and Q. pubescens showed a clear geographic genetic substructure. These large-scale within-species genetic structures were correlated to regionally variable levels of introgression between the species. For Q. petraea, introgression from Q. robur and Q. pubescens was less focused to particular regions, and this widespread inter-specific gene flow reduced the geographic genetic differentiation. The genetic variation at the maternally inherited SNPs led to 12 different haplotypes with a clear cross-species geographic pattern, further supporting the observation of significant hybridization and introgression among the species.

Integration of bioinformatics and machine learning strategies identifies APM-related gene signatures to predict clinical outcomes and therapeutic responses for breast cancer patients

BackgroundTumor antigenicity and efficiency of antigen presentation jointly influence tumor immunogenicity, which largely determines the effectiveness of immune checkpoint blockade (ICB). However, the role of altered antigen processing and presentation machinery (APM) in breast cancer (BRCA) has not been fully elucidated. MethodsA series of bioinformatic analyses and machine learning strategies were performed to construct APM-related gene signatures to guide personalized treatment for BRCA patients. A single-sample gene set enrichment analysis (ssGSEA) algorithm and weighted gene co-expression network analysis (WGCNA) were combined to screen for BRCA-specific APM-related genes. The non-negative matrix factorization (NMF) algorithm was used to divide the cohort into different clusters and the fgsea algorithm was applied to investigate the altered signaling pathways. Random survival forest (RSF) and the least absolute shrinkage and selection operator (Lasso) Cox regression analysis were combined to construct an APM-related risk score (APMrs) signature to predict overall survival. Furthermore, a nomogram and decision tree were generated to improve predictive accuracy and risk stratification for individual patients. Based on Tumor Immune Dysfunction and Exclusion (TIDE) method, random forest (RF) and Lasso logistic regression model were combined to establish an APM-related immunotherapeutic response score (APMis). Finally, immune infiltration, immunomodulators, mutational patterns, and potentially applicable drugs were comprehensively analyzed in different APM-related risk groups. IHC staining was used to assess the expression of APM-related genes in clinical samples. ResultsIn this study, APMrs and APMis showed favorable performances in risk stratification and therapeutic prediction for BRCA patients. APMrs exhibited more powerful prognostic capacity and accurate survival prediction compared to conventional clinicopathological features. APMrs was closely associated with distinct mutational patterns, immune cell infiltration and immunomodulators expression. Furthermore, the two APM-related gene signatures were independently validated in external cohorts with prognosis or immunotherapeutic responses. Potential applicable drugs and targets were mined in the APMrs-high group. APM-related genes were further validated in our in-house samples. ConclusionThe APM-related gene signatures established in our study could improve the personalized assessment of survival risk and guide ICB decision-making for BRCA patients.

Partial NMF-based hyperspectral unmixing methods for linear mixing models addressing intra-class variability

Linear Mixing Models (LMMs) are the most popular ones used in the linear hyperspectral unmixing field. However, several of them do not take into account available additional information and intra-class variability. In this investigation, two informed LMMs, with their related partial unmixing algorithms, which simultaneously address intra-class variability and exploit some available information, are introduced here for hyperspectral unmixing. More precisely, partial, by means of some available known endmember spectra, iterative and multiplicative Nonnegative Matrix Factorization (NMF) algorithms are designed to unmix hyperspectral data by considering the two informed LMMs that deal with intra-class variability. These two unmixing algorithms based on partial NMF, specifically designed to address intra-class variability, can be applied for automated detection, precise mapping, and accurate estimation of areas containing materials with known spectra in hyperspectral data. Experiments, based on synthetic and real data, are carried out, and obtained results are compared to those achieved by some literature methods. These experiments clearly demonstrate the effectiveness of the two proposed informed LMMs and their associated partial NMF-based unmixing algorithms.

Topological analysis of overlapping dynamic brain networks under naturalistic stimuli based on sparse non-negative matrix factorization algorithm

Based on the overlapping hypothesis of human brain networks, the sparse non-negative matrix factorization algorithm was used to decompose the dynamic brain functional networks under naturalistic stimuli into a set of spatiotemporal overlapping subgraphs and weight coefficients. This is followed by an analysis of how these subgraphs characterize the topographies in dynamic brain networks. It is also found that specific local topology is modulated by emotional activities. Default Mode Network, Ventral Attention Network, Auditory Network, Visual Network, and some brain regions in other cognitive networks are mainly involved. Our results are expected to provide solutions for future brain function detection and diagnosis of brain diseases based on naturalistic stimuli.

A novel force-constrained non-negative matrix factorization algorithm demonstrates that muscle synergies are useful in force control

A novel force-constrained non-negative matrix factorization algorithm demonstrates that muscle synergies are useful in force control

Hub gene identification and molecular subtype construction for Helicobacter pylori in gastric cancer via machine learning methods and NMF algorithm.

Helicobacter pylori (HP) is a gram-negative and spiral-shaped bacterium colonizing the human stomach and has been recognized as the risk factor of gastritis, peptic ulcer disease, and gastric cancer (GC). Moreover, it was recently identified as a class I carcinogen, which affects the occurrence and progression of GC via inducing various oncogenic pathways. Therefore, identifying the HP-related key genes is crucial for understanding the oncogenic mechanisms and improving the outcomes of GC patients. We retrieved the list of HP-related gene sets from the Molecular Signatures Database. Based on the HP-related genes, unsupervised non-negative matrix factorization (NMF) clustering method was conducted to stratify TCGA-STAD, GSE15459, GSE84433 samples into two clusters with distinct clinical outcomes and immune infiltration characterization. Subsequently, two machine learning (ML) strategies, including support vector machine-recursive feature elimination (SVM-RFE) and random forest (RF), were employed to determine twelve hub HP-related genes. Beyond that, receiver operating characteristic and Kaplan-Meier curves further confirmed the diagnostic value and prognostic significance of hub genes. Finally, expression of HP-related hub genes was tested by qRT-PCR array and immunohistochemical images. Additionally, functional pathway enrichment analysis indicated that these hub genes were implicated in the genesis and progression of GC by activating or inhibiting the classical cancer-associated pathways, such as epithelial-mesenchymal transition, cell cycle, apoptosis, RAS/MAPK, etc. In the present study, we constructed a novel HP-related tumor classification in different datasets, and screened out twelve hub genes via performing the ML algorithms, which may contribute to the molecular diagnosis and personalized therapy of GC.

Music emotion representation based on non-negative matrix factorization algorithm and user label information.

Music emotion representation learning forms the foundation of user emotion recognition, addressing the challenges posed by the vast volume of digital music data and the scarcity of emotion annotation data. This article introduces a novel music emotion representation model, leveraging the nonnegative matrix factorization algorithm (NMF) to derive emotional embeddings of music by utilizing user-generated listening lists and emotional labels. This approach facilitates emotion recognition by positioning music within the emotional space. Furthermore, a dedicated music emotion recognition algorithm is formulated, alongside the proposal of a user emotion recognition model, which employs similarity-weighted calculations to obtain user emotion representations. Experimental findings demonstrate the method's convergence after a mere 400 iterations, yielding a remarkable 47.62% increase in F1 value across all emotion classes. In practical testing scenarios, the comprehensive accuracy rate of user emotion recognition attains an impressive 52.7%, effectively discerning emotions within seven emotion categories and accurately identifying users' emotional states.

Consistent control information driven musculoskeletal model for multiday myoelectric control

Objective. Musculoskeletal model (MM)-based myoelectric interface has aroused great interest in human-machine interaction. However, the performance of electromyography (EMG)-driven MM in long-term use would be degraded owing to the inherent non-stationary characteristics of EMG signals. Here, to improve the estimation performance without retraining, we proposed a consistent muscle excitation extraction approach based on an improved non-negative matrix factorization (NMF) algorithm for MM when applied to simultaneous hand and wrist movement prediction. Approach. We added constraints and L 2-norm regularization terms to the objective function of classic NMF regarding muscle weighting matrix and time-varying profiles, through which stable muscle synergies across days were identified. The resultant profiles of these synergies were then used to drive the MM. Both offline and online experiments were conducted to evaluate the performance of the proposed method in inter-day scenarios. Main results. The results demonstrated significantly better and more robust performance over several competitive methods in inter-day experiments, including machine learning methods, EMG envelope-driven MM, and classic NMF-based MM. Furthermore, the analysis of control information on different days revealed the effectiveness of the proposed method in obtaining consistent muscle excitations. Significance. The outcomes potentially provide a novel and promising pathway for the robust and zero-retraining control of myoelectric interfaces.

Auto-adjustable hypergraph regularized non-negative matrix factorization for image clustering

Non-negative matrix factorization (NMF) is an effective method for image clustering. However, relatively fixed graph regularization terms and loss functions have been adopted by recently proposed variants of NMF, and their clustering performance can be improved by incorporating configurable parameters. In this paper, an auto-adjustable hypergraph regularized non-negative matrix factorization (AHRNMF) algorithm was proposed. In the AHRNMF framework, we proposed a piecewise loss function and an innovative auto-adjustable hypergraph. The loss function incorporates two adaptive parameters, harmonizing reconstruction error and anti-outlier efficacy. Hypergraph construction relies on the calculation of two k-nearest neighbors (KNN) with different scales. Furthermore, an KNN-based algorithm was developed to assist AHRNMF in achieving auto-adjustment, which can automatically detect outliers without determining the number of clusters in advance. It was demonstrated by extensive experiments that the proposed AHRNMF outperforms other state-of-the-art methods.

Elucidating Cuproptosis-Associated Genes in the Progression from Nash to HCC Using Bulk and Single-Cell RNA Sequencing Analyses and Experimental Validation.

Background and Objectives: Non-alcoholic steatohepatitis (NASH) is a significant risk factor for hepatocellular carcinoma (HCC) development. Timely treatment during the NASH stage is essential to minimize the possibility of disease progression to HCC. Cuproptosis is a newly identified form of cellular death that could impact the progression of various diseases and cancers. Materials and Methods: Transcriptome and single-cell sequencing datasets were utilized to investigate the role of cuproptosis-related genes (CRGs) in NASH progression to HCC. FDX1, LIPT1, and PDHP were identified as CRGs in NASH patients, and FDX1, DBT, GCSH, SLC31A1, and DLAT were identified as CRGs in patients with NASH progressing to HCC. FDX1 was found to play a significant role in both NASH patients and patients with NASH progressing to HCC. This study constructed cuproptosis-related clusters (CRCs) using the Nonnegative Matrix Factorization algorithm, and they were linked to fatty acid metabolism and the PPAR signaling pathway in both NASH CRCs and HCC CRCs. The Weighted Correlation Network Analysis algorithm identified CRP, CRC, TAT, CXCL10, and ACTA1 as highly relevant genes in NASH CRCs and HCC CRCs. The expression of FDX1 was validated in both mouse models and human NASH samples. Results: The investigation highlights FDX1 as a pivotal CRG in both NASH and NASH progression to HCC. The comprehensive characterization of CRGs sheds light on their potential biofunctional importance in the context of NASH and HCC. Our experimental results show that FDX1 expression was significantly increased in NASH patients. Conclusions: The present study identified key CRGs, revealing their potential impact on NASH and HCC. Meanwhile, targeting FDX1 may prevent the progression of NASH to HCC.

Distributed out-of-memory NMF on CPU/GPU architectures

AbstractWe propose an efficient distributed out-of-memory implementation of the non-negative matrix factorization (NMF) algorithm for heterogeneous high-performance-computing systems. The proposed implementation is based on prior work on NMFk, which can perform automatic model selection and extract latent variables and patterns from data. In this work, we extend NMFk by adding support for dense and sparse matrix operation on multi-node, multi-GPU systems. The resulting algorithm is optimized for out-of-memory problems where the memory required to factorize a given matrix is greater than the available GPU memory. Memory complexity is reduced by batching/tiling strategies, and sparse and dense matrix operations are significantly accelerated with GPU cores (or tensor cores when available). Input/output latency associated with batch copies between host and device is hidden using CUDA streams to overlap data transfers and compute asynchronously, and latency associated with collective communications (both intra-node and inter-node) is reduced using optimized NVIDIA Collective Communication Library (NCCL) based communicators. Benchmark results show significant improvement, from 32X to 76x speedup, with the new implementation using GPUs over the CPU-based NMFk. Good weak scaling was demonstrated on up to 4096 multi-GPU cluster nodes with approximately 25,000 GPUs when decomposing a dense 340 Terabyte-size matrix and an 11 Exabyte-size sparse matrix of density $$10^{-6}$$ 10 - 6 .

Comodule Estimation of Cognitive Sensor Networks Based on Partial Clustering for Partial Observed Data

The proposed study is on the partial clustering algorithms for cognitive sensor networks that deal with partially observed data. The proposed algorithms aim to estimate clusters in the presence of missing values and leverage data imputation techniques to fill in the gaps in the target and station device matrices. A modified loss function is introduced to shape the cluster centers, and robust Non-negative Matrix Factorization (NMF) algorithms are utilized to enhance the robustness of the clustering process. This research contributes to the field of cognitive sensor networks by providing insights into the challenges of partial clustering and presenting effective algorithms to address them. The proposed methods have the potential to enhance the performance of clustering tasks in various domains, including sensor networks, by accounting for missing data and producing accurate cluster reconstructions.