Multiple Latent Factors Research Articles

Dimensional subtyping of first-episode drug-naïve major depressive disorder: A multisite resting-state fMRI study

Major depressive disorder (MDD) is a heterogeneous syndrome, and understanding its neural mechanisms is crucial for the advancement of personalized medicine. However, conventional subtyping studies often categorize MDD patients into a single subgroup, neglecting the continuous interindividual variations. This implies a pressing need for a dimensional approach. 230 first-episode drug-naïve MDD patients and 395 healthy controls were obtained from 5 sites via the Rest-meta-MDD project. A Bayesian model was used to decompose the resting-state functional connectivity (RSFC) into multiple distinct RSFC patterns (refer to as “factors”), and each individual was allowed to express multiple factors to varying degrees (dimensional subtyping). The associations between demographic and clinical variables with the identified factors were calculated. We identified three latent factors with distinct but partially overlapping hypo- and hyper-RSFC patterns. Most participants co-expressed multiple latent factors. All factors shared abnormal RSFC involving the default mode network and frontoparietal network, but the directionality partially differed across factors. All factors were not significantly associated with demographic and clinical variables. These findings shed light on the interindividual variability in MDD and could form the basis for developing novel therapeutic approaches that capitalize on the heterogeneity of MDD.

Bayesian Analysis of Two-Part Latent Variable Model with Mixed Data

AbstractIn analyzing semi-continuous data, two-part model is a widely appreciated tool, in which two components are enclosed to characterize the mixing proportion of zeros and the actual level of positive values in semi-continuous data. The primary interest underlying such a model is primarily to exploit the dependence of the observed covariates on the semi-continuous variables; as such, the exploitation of unobserved heterogeneity is sometimes ignored. In this paper, we extend the conventional two-part regression model to much more general situations where multiple latent factors are considered to interpret the latent heterogeneity arising from the absence of covariates. A structural equation is constructed to describe the interrelationships between the latent factors. Moreover, a general statistical analysis procedure is developed to accommodate semi-continuous, ordered and unordered data simultaneously. A procedure for parameter estimation and model assessment is developed under a Bayesian framework. Empirical results including a simulation study and a real example are presented to illustrate the proposed methodology.

Distinct components of cardiovascular health are linked with age-related differences in cognitive abilities

Cardiovascular ageing contributes to cognitive impairment. However, the unique and synergistic contributions of multiple cardiovascular factors to cognitive function remain unclear because they are often condensed into a single composite score or examined in isolation. We hypothesized that vascular risk factors, electrocardiographic features and blood pressure indices reveal multiple latent vascular factors, with independent contributions to cognition. In a population-based deep-phenotyping study (n = 708, age 18–88), path analysis revealed three latent vascular factors dissociating the autonomic nervous system response from two components of blood pressure. These three factors made unique and additive contributions to the variability in crystallized and fluid intelligence. The discrepancy in fluid relative to crystallized intelligence, indicative of cognitive decline, was associated with a latent vascular factor predominantly expressing pulse pressure. This suggests that higher pulse pressure is associated with cognitive decline from expected performance. The effect was stronger in older adults. Controlling pulse pressure may help to preserve cognition, particularly in older adults. Our findings highlight the need to better understand the multifactorial nature of vascular aging.

GATrust: A Multi-Aspect Graph Attention Network Model for Trust Assessment in OSNs

Social trust assessment that characterizes a pairwise trustworthiness relationship can spur diversified applications. Extensive efforts have been put in exploration, but mainly focusing on applying graph convolutional network to establish a social trust evaluation model, overlooking user feature factors related to context-aware information on social trust prediction. In this article, we aim to design a new trust assessment framework GATrust which integrates multi-aspect properties of users, including user context-specific information, network topological structure information, and locally-generated social trust relationships. GATrust can assigns different attention coefficients to multi-aspect properties of users in online social networks, for improving the prediction accuracy of social trust evaluation. The framework can then learn multiple latent factors of each trustor-trustee pair to establish a social trust evaluation model, by fusing graph attention network and graph convolution network. We conduct extensive experiments on two popular real-world datasets and the results exhibit that our proposed framework can improve the precision of social trust prediction, outperforming the state-of-the-art in the literature by 4.3% and 5.5% on both two datasets, respectively.

A Gene Rank Based Approach for Single Cell Similarity Assessment and Clustering.

Single-cell RNA sequencing (scRNA-seq) technology provides quantitative gene expression profiles at single-cell resolution. As a result, researchers have established new ways to explore cell population heterogeneity and genetic variability of cells. One of the current research directions for scRNA-seq data is to identify different cell types accurately through unsupervised clustering methods. However, scRNA-seq data analysis is challenging because of their high noise level, high dimensionality and sparsity. Moreover, the impact of multiple latent factors on gene expression heterogeneity and on the ability to accurately identify cell types remains unclear. How to overcome these challenges to reveal the biological difference between cell types has become the key to analyze scRNA-seq data. For these reasons, the unsupervised learning for cell population discovery based on scRNA-seq data analysis has become an important research area. A cell similarity assessment method plays a significant role in cell clustering. Here, we present BioRank, a new cell similarity assessment method based on annotated gene sets and gene ranks. To evaluate the performances, we cluster cells by two classical clustering algorithms based on the similarity between cells obtained by BioRank. In addition, BioRank can be used by any clustering algorithm that requires a similarity matrix. Applying BioRank to 12 public scRNA-seq datasets, we show that it is better than or at least as well as several popular similarity assessment methods for single cell clustering.

Feature Extraction and Identification of Alzheimer's Disease based on Latent Factor of Multi-Channel EEG.

Alzheimer's disease is a neurodegenerative disease in old age, early diagnosis will help to delay the progression of the disease. Presently, the features of brain functional diseases can be obtained with EEG analysis, but the relationship between characteristics of EEG and Alzheimer's disease has not been clearly clarified. In this work, we hypothesize that there exist default brain variables (latent factors) across subjects in disease processes, decoding latent factor from brain activity contributes to the study of cognitive impairment. To that end, this work proposes to extract characteristics of Alzheimer's disease by combing latent factors of EEG with variational auto-encoder to realize disease identification. Primarily, power spectrum characteristics is investigated and it is found that the dominant frequency of two groups is different. Further analysis reveals that latent factor distribution of Alzheimer's disease exists obvious differences with normal group in the theta frequency band. Moreover, the latent factors are projected onto the three-dimensional state space and the transient rotation of neural state is found, which shows the dynamic characteristics of latent factors. In addition, Takagi-Sugeno-Kang classifier is adopted and multiple latent factors are fed into Takagi-Sugeno-Kang classifier for decoding. Compared with linear classifier, Takagi-Sugeno-Kang fuzzy classifier has better performance in classification of energy feature from sub-frequency bands of latent factors. The accuracy of identification could up to 98.10% when the combination of energy features of four frequency bands is used as model input. Collectively, this work provides a feasible tool for identification of neurological dysfunction from the view of latent factors, especially contributing to the diagnosis of Alzheimer's disease.

Application of exploratory factor analysis to address the challenge of measuring social capital in a rural communal setting in South Africa

ABSTRACTSocial capital (SC) is an umbrella concept combining attributes of multiple latent factors that are not directly observable, making it difficult to measure and express as a single variable. Despite its multidimensional nature, the bulk of empirical studies continue to construct and use unidimensional indicators of SC, generating notable disparities in results derived from the use of these alternative measures. This study employed exploratory factor analysis to search for and construct composite measures capturing the multidimensional facets of structural and cognitive social capital (SC) in a rural communal setting in Africa. Our factor analysis revealed eleven factors describing a diversity of SC components, with clear evidence of multiple features of cognitive SC at the household level. On the other hand, evidence of presence of structural SC was limited, which is a general finding of household level analysis of determinants of SC. Not adequately accounting for elements of informal social networks, is an important weakness of our study, and we accordingly strongly recommend that SC research in developing countries should include, as indicators of structural SC, measures of informal social networks.

Multidimensional clinical data denoising via Bayesian CP factorization

CANDECOMP/PARAFAC (CP) tensor factorization is an efficient technique for incomplete tensor-data processing through capturing the multilinear latent factors. Based on the incorporate a sparsity-inducing prior over multiple latent factors and appropriate hyper-priors over all hyper-parameters, a Bayesian-based hierarchical probabilistic CP factorization model could be formed. By this way, the rank of the incomplete tensor can be determined automatically. In this paper, we explored the tensor completion method in processing incomplete multidimensional electroencephalogram (EEG) and magnetic resonance imaging (MRI) clinical data. The empirical results indicated that the Bayesian CP tensor factorization of incomplete data method can effectively recover EEG signal with missing data and denoised the noisy MRI data.

An Adversarial Neuro-Tensorial Approach for Learning Disentangled Representations

Several factors contribute to the appearance of an object in a visual scene, including pose, illumination, and deformation, among others. Each factor accounts for a source of variability in the data, while the multiplicative interactions of these factors emulate the entangled variability, giving rise to the rich structure of visual object appearance. Disentangling such unobserved factors from visual data is a challenging task, especially when the data have been captured in uncontrolled recording conditions (also referred to as “in-the-wild”) and label information is not available. In this paper, we propose a pseudo-supervised deep learning method for disentangling multiple latent factors of variation in face images captured in-the-wild. To this end, we propose a deep latent variable model, where the multiplicative interactions of multiple latent factors of variation are explicitly modelled by means of multilinear (tensor) structure. We demonstrate that the proposed approach indeed learns disentangled representations of facial expressions and pose, which can be used in various applications, including face editing, as well as 3D face reconstruction and classification of facial expression, identity and pose.

One health behavior or many health-related behaviors?

ObjectiveOf the many existing health models, models of health behavior are considered optimal for research and application as they focus on concrete forms of behavior that support, maintain, or undermine one’s health, and they accentuate the individual as the initiator of this behavior. Research in this area follows a broad range of concrete partial manifestations of health behavior. Is it necessary to differentiate between various types of health behavior or could these partial manifestations be combined under one common scale?MethodsData acquisition tool: Health-Related Behavior Scale (HRBS, 42 items). Data processing methods: principal component analysis (the internal structure of HRBS), confirmatory factor analysis (the latent factor structure of four tested models). Sample: N=1,664 adult respondents.ResultsThe HRBS described ten areas of health-related behavior (ten extracted factors). All tested models of latent structure showed almost identical mathematical and statistical values of the model.ConclusionHealth-related behavior includes a set of partial behaviors (behavior related to nutrition, addictive substances, movement, and physical exercises). An unambiguous latent factor structure has not been revealed. An open question remains whether there is one latent factor behind all health-related behaviors or whether there are multiple latent factors. The use of one or the other model should be deduced from the underlying theory and research objectives. To find a reliable model of health behavior, it is necessary to include moderators and mediators such as personality, attitude, or economic status.

Forecasting Financial Stress Indices in Korea: A Factor Model Approach

We propose factor-based out-of-sample forecast models for the financial stress index and its 4 sub-indices developed by the Bank of Korea. We employ the method of the principal components for 198 monthly frequency macroeconomic data to extract multiple latent factors that summarize the common components of the entire data set. We evaluate the out-of-sample predictability of our models via the ratio of the root mean squared prediction errors and the Diebold-Mariano-West statistics. Our factor models overall outperform the random walk model in forecasting the financial stress indices for up to 1-year forecast horizon. Our models also perform fairly well relative to a stationary autoregressive model especially when the forecast horizon is short, which is practically useful because financial crises often occur abruptly with no systemic warning signals. Parsimonious models with small number of factors perform as well as bigger models. Overall, our findings imply that not only financial data but also real activity variables can help out-of-sample forecast the vulnerability in the financial markets.

Fine-Tuning of the RIG-I-Like Receptor/Interferon Regulatory Factor 3-Dependent Antiviral Innate Immune Response by the Glycogen Synthase Kinase 3/β-Catenin Pathway.

Induction of an antiviral innate immune response relies on pattern recognition receptors, including retinoic acid-inducible gene 1-like receptors (RLR), to detect invading pathogens, resulting in the activation of multiple latent transcription factors, including interferon regulatory factor 3 (IRF3). Upon sensing of viral RNA and DNA, IRF3 is phosphorylated and recruits coactivators to induce type I interferons (IFNs) and selected sets of IRF3-regulated IFN-stimulated genes (ISGs) such as those for ISG54 (Ifit2), ISG56 (Ifit1), and viperin (Rsad2). Here, we used wild-type, glycogen synthase kinase 3α knockout (GSK-3α(-/-)), GSK-3β(-/-), and GSK-3α/β double-knockout (DKO) embryonic stem (ES) cells, as well as GSK-3β(-/-) mouse embryonic fibroblast cells in which GSK-3α was knocked down to demonstrate that both isoforms of GSK-3, GSK-3α and GSK-3β, are required for this antiviral immune response. Moreover, the use of two selective small-molecule GSK-3 inhibitors (CHIR99021 and BIO-acetoxime) or ES cells reconstituted with the catalytically inactive versions of GSK-3 isoforms showed that GSK-3 activity is required for optimal induction of antiviral innate immunity. Mechanistically, GSK-3 isoform activation following Sendai virus infection results in phosphorylation of β-catenin at S33/S37/T41, promoting IRF3 DNA binding and activation of IRF3-regulated ISGs. This study identifies the role of a GSK-3/β-catenin axis in antiviral innate immunity.

Bayesian Robust Tensor Factorization for Incomplete Multiway Data

We propose a generative model for robust tensor factorization in the presence of both missing data and outliers. The objective is to explicitly infer the underlying low-CANDECOMP/PARAFAC (CP)-rank tensor capturing the global information and a sparse tensor capturing the local information (also considered as outliers), thus providing the robust predictive distribution over missing entries. The low-CP-rank tensor is modeled by multilinear interactions between multiple latent factors on which the column sparsity is enforced by a hierarchical prior, while the sparse tensor is modeled by a hierarchical view of Student-t distribution that associates an individual hyperparameter with each element independently. For model learning, we develop an efficient variational inference under a fully Bayesian treatment, which can effectively prevent the overfitting problem and scales linearly with data size. In contrast to existing related works, our method can perform model selection automatically and implicitly without the need of tuning parameters. More specifically, it can discover the groundtruth of CP rank and automatically adapt the sparsity inducing priors to various types of outliers. In addition, the tradeoff between the low-rank approximation and the sparse representation can be optimized in the sense of maximum model evidence. The extensive experiments and comparisons with many state-of-the-art algorithms on both synthetic and real-world data sets demonstrate the superiorities of our method from several perspectives.

Bayesian CP Factorization of Incomplete Tensors with Automatic Rank Determination.

CANDECOMP/PARAFAC (CP) tensor factorization of incomplete data is a powerful technique for tensor completion through explicitly capturing the multilinear latent factors. The existing CP algorithms require the tensor rank to be manually specified, however, the determination of tensor rank remains a challenging problem especially for CP rank . In addition, existing approaches do not take into account uncertainty information of latent factors, as well as missing entries. To address these issues, we formulate CP factorization using a hierarchical probabilistic model and employ a fully Bayesian treatment by incorporating a sparsity-inducing prior over multiple latent factors and the appropriate hyperpriors over all hyperparameters, resulting in automatic rank determination. To learn the model, we develop an efficient deterministic Bayesian inference algorithm, which scales linearly with data size. Our method is characterized as a tuning parameter-free approach, which can effectively infer underlying multilinear factors with a low-rank constraint, while also providing predictive distributions over missing entries. Extensive simulations on synthetic data illustrate the intrinsic capability of our method to recover the ground-truth of CP rank and prevent the overfitting problem, even when a large amount of entries are missing. Moreover, the results from real-world applications, including image inpainting and facial image synthesis, demonstrate that our method outperforms state-of-the-art approaches for both tensor factorization and tensor completion in terms of predictive performance.

SAGA: sparse and geometry-aware non-negative matrix factorization through non-linear local embedding

This paper presents a new non-negative matrix factorization technique which (1) allows the decomposition of the original data on multiple latent factors accounting for the geometrical structure of the manifold embedding the data; (2) provides an optimal representation with a controllable level of sparsity; (3) has an overall linear complexity allowing handling in tractable time large and high dimensional datasets. It operates by coding the data with respect to local neighbors with non-linear weights. This locality is obtained as a consequence of the simultaneous sparsity and convexity constraints. Our method is demonstrated over several experiments, including a feature extraction and classification task, where it achieves better performances than the state-of-the-art factorization methods, with a shorter computational time.

A Spectral Algorithm for Latent Dirichlet Allocation

Topic modeling is a generalization of clustering that posits that observations (words in a document) are generated by multiple latent factors (topics), as opposed to just one. The increased representational power comes at the cost of a more challenging unsupervised learning problem for estimating the topic-word distributions when only words are observed, and the topics are hidden. This work provides a simple and efficient learning procedure that is guaranteed to recover the parameters for a wide class of multi-view models and topic models, including latent Dirichlet allocation (LDA). For LDA, the procedure correctly recovers both the topic-word distributions and the parameters of the Dirichlet prior over the topic mixtures, using only trigram statistics (i.e., third order moments, which may be estimated with documents containing just three words). The method is based on an efficiently computable orthogonal tensor decomposition of low-order moments.

Multilinear Graph Embedding: Representation and Regularization for Images

Given a set of images, finding a compact and discriminative representation is still a big challenge especially when multiple latent factors are hidden in the way of data generation. To represent multifactor images, although multilinear models are widely used to parameterize the data, most methods are based on high-order singular value decomposition (HOSVD), which preserves global statistics but interprets local variations inadequately. To this end, we propose a novel method, called multilinear graph embedding (MGE), as well as its kernelization MKGE to leverage the manifold learning techniques into multilinear models. Our method theoretically links the linear, nonlinear, and multilinear dimensionality reduction. We also show that the supervised MGE encodes informative image priors for image regularization, provided that an image is represented as a high-order tensor. From our experiments on face and gait recognition, the superior performance demonstrates that MGE better represents multifactor images than classic methods, including HOSVD and its variants. In addition, the significant improvement in image (or tensor) completion validates the potential of MGE for image regularization.

Modeling individual subtests of the WAIS IV with multiple latent factors.

Performance on a cognitive test can be viewed either as measuring a unitary function or as reflecting the operation of multiple factors. Individual subtests in batteries designed to measure human abilities are commonly modeled as a single latent factor. Several latent factors are then used to model groups of subtests. However these latent factors are not independent as they are related through hierarchical or oblique structures. As a result, the simple structure of subtest performance results in complex latent factors. The present study used structural equation modeling to evaluate several multidimensional models of the Wechsler Adult Intelligence Scales- fourth edition (WAIS-IV) subtests. Multidimensional models of subtest performance provided better model fit as compared to several previously proposed one dimensional models. These multidimensional models also generalized well to new samples of populations differing in age from that used to estimate the model parameters. Overall these results show that models that describe subtests as multidimensional functions of uncorrelated factors provided a better fit to the WAIS-IV correlations than models that describe subtests as one dimensional functions of correlated factors. There appears to be a trade-off in modeling subtests as one dimensional and modeling with homogeneous latent traits. More consideration should be given to models that include multiple uncorrelated latent factors as determinants of the performance on a given subtest. These results support the view that performance on any given cognitive test is potentially the result of multiple factors. Simple structure may be too simple.

Return, Trading Volume, and Market Depth in Currency Futures Markets

We use a class of stochastic volatility models with multiple latent factors to investigate the joint dynamics of return, trading volume, and open interest (a proxy for market depth) in currency futures markets. In accordance with theory, the empirical evidence indicates that there is more than one latent factor affecting these three variables. However, the evidence is ambivalent on the choice between two- and three-latent-factor models. These three variables also display different patterns of information spillovers across currency futures.

Leverage and Volatility Feedback Effects in High-Frequency Data

We examine the relationship between volatility and past and future returns using high-frequency aggregate equity index data. Consistent with a prolonged ‘‘leverage’’ effect, we find the correlations between absolute high-frequency returns and current and past high-frequency returns to be significantly negative for several days, whereas the reverse cross-correlations are generally negligible. We also find that high-frequency data may be used in more accurately assessing volatility asymmetries over longer daily return horizons. Furthermore, our analysis of several popular continuous-time stochastic volatility models clearly points to the importance of allowing for multiple latent volatility factors for satisfactorily describing the observed volatility asymmetries.