Integrating Multi-omics Data for Alzheimer's Disease to Explore Its Biomarkers Via the Hypergraph-Regularized Joint Deep Semi-Non-Negative Matrix Factorization Algorithm.

The number of smart phone and smartphone users has increased rapidly, boosting the growth of mobile advertising. Predicting the percentage of users who click on an advertisement when displayed on an application (CTR: Click Through Rate) plays an important role in choosing an advertising push strategy for each application. However, because the field of mobile advertising is still new, the study of solutions to these problems is only for Web advertising. This paper presents a new predictive method based on Matrix Factorization (MF) algorithm. The paper also proposes a method of integrating the reliability of data into MF algorithms. The method was tested on actual data provided by Amobi advertising company with 1.7 million records including 300 advertisements and 400 mobile device applications. The AUC (Area Under The Curve) value obtained on the test set is 0.9037 with MF and 0.9411 when integrating reliability into the MF algorithm. Thus, the results show that the predictive quality is significantly improved when integrating data reliability and evaluation into MF algorithms.

An important problem of matrix completion/approximation based on Matrix Factorization (MF) algorithms is the existence of multiple global optima; this problem is especially serious when the matrix is sparse, which is common in real-world applications such as personalized recommender systems. In this work, we clarify data sparsity by bounding the solution space of MF algorithms. We present the conditions that an MF algorithm should satisfy for reliable completion of the unobservables, and we further propose to augment current MF algorithms with extra constraints constructed by compressive sampling on the unobserved values, which is well-motivated by the theoretical analysis. Model learning and optimal solution searching is conducted in a properly reduced solution space to achieve more accurate and efficient rating prediction performances. We implemented the proposed algorithms in the Map-Reduce framework, and comprehensive experimental results on Yelp and Dianping datasets verified the effectiveness and efficiency of the augmented matrix factorization algorithms.

BackgroundThe diagnosis of Alzheimer’s disease (AD) has been greatly improved due to the fundamental role of positron emission tomography (PET) imaging. Also, predicting PET brain imaging alterations using blood‐based biomarkers is of high interest. This way, integrating PET and omics data can provide new insights into AD pathophysiology. Here, we aimed to develop a framework that combines blood transcriptomics with PET data. We hypothesized that integrating omics and PET data will help advance our understanding of AD neurobiology and may reveal relevant new peripheral biomarkers.Method[18F]Fluorodeoxyglucose ([18F]FDG)‐PET imaging and transcriptomics data were acquired from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Microarray gene expression profiling from blood samples of 99 Cognitively Unimpaired (CU) and 218 Cognitively Impaired (CI) individuals were submitted to differential expression (DE) analysis. Differentially expressed genes (DEGs) were submitted to Gene Ontology (GO) analysis. The GO terms were clustered by semantic similarity using the GOSemSim method for biological processes ontology. All computations analyses were performed in the R statistical environment. Gene clusters obtained in the previous step were selected to undergo integration with [18F]FDG‐PET images using voxel‐wise generalized linear regression (GLR) models adjusted for age, gender, years of education, and APOEε4 (RMINC package).ResultThe GO semantic similarity resulted in 16 GO clusters. The voxel‐wise correlation between [18F]FDG‐PET and GO clusters resulted in t‐statistical maps. Afterwards, only statistically significant correlated voxels (uncorrected t‐value > 2.0) were retained (Fig. 1A). A cluster related to the regulation of protein serine/threonine kinase activity showed a strong correlation with the [18F]FDG‐PET signal in the brain. At the voxel level, this cluster has a positive correlation in the precuneus’ gray matter (95.7% left, 81.1% right) as well in the medial frontal gyrus (79.6% left, 44.66% right) and cingulate region (76.9% left, 56.4% right) (Fig. 1B).ConclusionWe identified several peripheral biological processes associated with [18F]FDG‐PET metabolism in the brain of CU and CI individuals. Furthermore, the prominent enrichment of protein serine/threonine kinase activity‐related genes highlights its potential as novel AD biomarkers and as a proxy of [18F]FDG‐PET metabolism.

Crop breeders often face challenges due to limited data availability when making crucial decisions, such as selecting top-performing varieties/hybrids for further experiments, registration, and commercialization. Evaluating all varieties/hybrids across all fields is impractical due to high experimental and time costs, as well as the limited number of locations for planting. This article aims to evaluate the performance of various maize hybrids in untested locations using historical data. The problem is approached through a matrix framework, where hybrids and fields correspond to rows and columns, respectively, with entries representing the yield of a specific hybrid at a given location. As this matrix is typically sparse, the task is to fill in missing data. Agronomists are primarily interested in the performance of top hybrids at specific locations for smart seed selection. To address this, we introduce a novel application of the Data Fusion by Matrix Factorization (DFMF) algorithm for predicting crop yields using maize data from the 2019 Syngenta Crop Challenge. The DFMF results are compared with the Random Forest (RF) algorithm as a benchmark, focusing on model performance for smart seed selection. Our analysis highlights the advantages of the DFMF approach over the traditional RF method in this context.

Clustering with partial supervision background information or semi-supervised clustering, learning from a combination of both labeled and unlabeled data, has received a lot of attention over the last decade. The supervisory information is usually used as the constraints to bias clustering towards a good region of search space. This paper proposes a semi-supervised algorithm, called constrained non-negative matrix factorization (Constrained-NMF), with a few labeled examples as constraints to improve performance. The proposed algorithm is a matrix factorization algorithm, in which initialization of matrices is required at the beginning. Although the benefits of good initialization are well-known, randomized seeding of basis matrix and coefficient matrix is still the standard approach for many non-negative matrix factorization (NMF) algorithms. This work devises an algorithm called entropy-based weighted semi-supervised fuzzy c-means (EWSS-FCM) algorithm to initialize the seeds. The experimental results indicate that the proposed Constrained-NMF can benefit from the initialization obtained from EWSS-FCM, which emphasizes the role of labeled examples and automatically weights them during the course of clustering. This work considers labeled examples in the objective functions to devise the two algorithms, in which the labeled information is propagated to unlabeled examples iteratively. We further analyze the proposed Constrained-NMF and give convergence justifications. The experiments are conducted on five real data sets, and experimental results indicate that the proposed algorithm generally outperforms the other alternatives.

Crowdsourcing has arisen as a new problem-solving paradigm for tasks that are difficult for computers but easy for humans. However, since the answers collected from the recruited participants (workers) may contain sensitive information, crowdsourcing raises serious privacy concerns. In this paper, we investigate the problem of protecting answer privacy under local differential privacy (LDP), by which individual workers randomize their answers independently and send the perturbed answers to the task requester. The utility goal is to enable to infer the true answer (i.e., truth) from the perturbed data with high accuracy. One of the challenges of LDP perturbation is the sparsity of worker answers (i.e., each worker only answers a small number of tasks). Simple extension of the existing approaches (e.g., Laplace perturbation and randomized response) may incur large error of truth inference on sparse data. Thus we design an efficient new matrix factorization (MF) algorithm under LDP. We prove that our MF algorithm can provide both LDP guarantee and small error of truth inference, regardless of the sparsity of worker answers. We perform extensive experiments on real-world and synthetic datasets, and demonstrate that the MF algorithm performs better than the existing LDP algorithms on sparse crowdsourcing data.

Face recognition is a challenging issue in field of multi-science, main contents of research is how to make computer have ability of face recognition face recognition technology involved in a lot, which is a key feature extraction and classification method, this paper focuses on study of related theory. Non-negative matrix factorization trapped MF) algorithm and local non-negative matrix factorization (LNMF) algorithm is a feature extraction method based on local features, has been successfully used in face recognition NMF algorithm in face recognition rate is low, although LNMF algorithm to a certain extent, improve recognition rate, but its price is to increase number of iterations. In addition, two algorithms have failed to solve good nonlinear separable problems the kernel method combined with LNMF algorithm, kernel local non-negative matrix factorization (KLNMF) algorithm, first by a nonlinear transformation of original space to high-dimensional space, making samples linearly separable, and then use LNMF algorithm to extract face features. In classification part, paper presents decision rules of classification of their own, and design based on NMF subspace classifier. DOI : http://dx.doi.org/10.11591/telkomnika.v12i3.4356 Full Text: PDF

The correct interpretation of clinical positron emission tomography (PET) data depends largely on the physical limits of the PET scanner. The partial volume effect (PVE) is related to the size of the studied object compared to the spatial resolution. It represents one of the most important limiting factors in quantitative data analysis. This effect is increased in the case of atrophy, as in patients with Alzheimer disease (AD), and it influences measurement of the metabolic reduction generally seen in cerebral degeneration. In this case, interpretation can be biased, because cortical activity will be underestimated due to the atrophy. In general, anatomical images of AD patients have shown diffuse atrophy, while PET studies have found widespread hypometabolism affecting the parietal and temporal lobes. Although hypometabolic areas usually correspond to atrophic regions, they also occur without such changes. Thus, the aim is to differentiate authentic hypometabolism (decrease of glucose consumption per unit volume of gray matter) from that due to PVE from atrophy (cell loss). Consequently, we are using a method for three-dimensional (3D) correction of human PET data with 3D magnetic resonance imaging (MRI). We measured atrophy and metabolism by using both T1-weighted MR images and high and medium resolution PET scans. We injected 12 patients and controls with [18F]fluorodeoxyglucose for glucose consumption measurements. Atrophy was estimated in the following way. We isolated the cerebral structures, using a segmentation technique on the MRI scans, into gray matter (GM), white matter, and cerebrospinal fluid. We superimposed the PET images onto the MR images to obtain anatomo-functional correlations. We degraded the segmented MR images to the resolution of the PET images by a convolution process to create a PET image correction map. We corrected the metabolic PET data for the PVE. We studied the cerebral metabolic rate of glucose in the GM where metabolic variation is the most relevant to AD. By dealing with problems relating to the sensitivity to the segmentation and to the PET-MRI coregistration, computation of MRI convolution processes provided the degree of PVE on a pixel-by-pixel basis, allowing correction of hypometabolisms contained in GM PET values. Global cortical metabolism increased after correction for PVE by, on average, 29 and 24% for tomographs acquired with medium (TTV03 LETI) and high (ECAT 953B CTI/Siemens) resolution, respectively, whereas the cortical metabolism increased by 75 and 65% for the respective tomographs in AD patients. The difference of metabolism between scans after correction for PVE was less than before correction, decreasing from 31 to 17%. This difference was most marked in the frontal and temporal lobes. Fusion imaging allowed correction for PVE in metabolic data using 3D MRI and determination of whether a change in the apparent radiotracer concentration in PET data reflected an alteration in GM volume, a change in radiotracer concentration per unit volume of GM, or both.

The non-negative matrix factorization (NMF) algorithm is a classical matrix factorization and dimension reduction method in machine learning and data mining. However, in real problems, we always have to run the algorithm for several times and use the best matrix factorization result as the final output because of the random initialization of the matrix factorization. In this paper, we proposed an improved non-negative matrix factorization algorithm based on genetic algorithm (GA), which uses the internal parallelism and the random search of genetic algorithm to get the optimal solution of matrix factorization. It could have larger searching area and higher accuracy in matrix factorization. In the document clustering problem, we use the TDT2 dataset and design several contrast experiments on the classical NMF and the improved NMF based on genetic algorithm, the experiment results show that our improved non-negative matrix factorization algorithm has higher clustering accuracy and better robustness.

In this paper, we describe scalable parallel algorithms for symmetric sparse matrix factorization, analyze their performance and scalability, and present experimental results for up to 1,024 processors on a Gray T3D parallel computer. Through our analysis and experimental results, we demonstrate that our algorithms substantially improve the state of the art in parallel direct solution of sparse linear systems-both in terms of scalability and overall performance. It is a well known fact that dense matrix factorization scales well and can be implemented efficiently on parallel computers. In this paper, we present the first algorithms to factor a wide class of sparse matrices (including those arising from two- and three-dimensional finite element problems) that are asymptotically as scalable as dense matrix factorization algorithms on a variety of parallel architectures. Our algorithms incur less communication overhead and are more scalable than any previously known parallel formulation of sparse matrix factorization. Although, in this paper, we discuss Cholesky factorization of symmetric positive definite matrices, the algorithms can be adapted for solving sparse linear least squares problems and for Gaussian elimination of diagonally dominant matrices that are almost symmetric in structure. An implementation of one of our sparse Cholesky factorization algorithms delivers up to 20 GFlops on a Gray T3D for medium-size structural engineering and linear programming problems. To the best of our knowledge, this is the highest performance ever obtained for sparse Cholesky factorization on any supercomputer.

Matrix factorization has been widely applied to various applications. With the fast development of storage and internet technologies, we have been witnessing a rapid increase of data. In this paper, we propose new algorithms for matrix factorization with the emphasis on efficiency. In addition, most existing methods of matrix factorization only consider a general smooth least square loss. Differently, many real-world applications have distinctive characteristics. As a result, different losses should be used accordingly. Therefore, it is beneficial to design new matrix factorization algorithms that are able to deal with both smooth and non-smooth losses. To this end, one needs to analyze the characteristics of target data and use the most appropriate loss based on the analysis. We particularly study two representative cases of low-rank matrix recovery, i.e., collaborative filtering for recommendation and high dynamic range imaging. To solve these two problems, we respectively propose a stage-wise matrix factorization algorithm by exploiting manifold optimization techniques. From our theoretical analysis, they are both are provably guaranteed to converge to a stationary point. Extensive experiments on recommender systems and high dynamic range imaging demonstrate the satisfactory performance and efficiency of our proposed method on large-scale real data.

Matrix factorization is a key component for solving several computer vision problems. It is particularly challenging in the presence of missing or erroneous data, which often arise in structure-from-motion. We propose batch algorithms for matrix factorization. They are based on closure and basis constraints, that are used either on the cameras or the structure, leading to four possible algorithms. The constraints are robustly computed from complete measurement sub-matrices with e.g. random data sampling. The cameras and 3D structure are then recovered through linear least squares. Prior information about the scene such as identical camera positions or orientations, smooth camera trajectory, known 3D points and coplanarity of some 3D points can be directly incorporated. We demonstrate our algorithms on challenging image sequences with tracking error and more than 95% missing data.

Traditional Collaborative filtering (CF) is one of the techniques of recommender systems that has been successfully exploited in various applications, but sometimes they fail to provide accurate recommendations because they depend majorly on the rating matrix, which is always scanty and of very high dimension. Matrix factorization (MF) algorithms are variants of latent factor models, which are easy, fast, and efficient. This article reviews the related research and advances in the application of matrix factorization techniques in recommender systems. Popular matrix factorization algorithms utilized in recommender systems were reviewed. The peculiar challenges of using matrix factorization in recommender systems were also enumerated and discussed with the goal of identifying the different problems solved with the use of matrix factorization techniques as applied in recommender systems.

PCANet is a deep learning network that uses an orthogonal factorization form (i.e., principal component analysis, PCA) to learn the filter bank. It shows superior performance for image classification task. However, PCA only considers the statistics of the image patches while it does not consider the part-based matrix factorization (i.e. Non-negative Matrix Factorization). In this paper, we proposed to replace the PCA algorithm with the Semi-Non-Negative Matrix Factorization (SNMF) algorithm and to construct the SNMFNet. Specifically, first, we propose a seminon- negative matrix factorization algorithm, which uses the l2 norm regularizer to restrict base vectors to substitute the nonnegative constraint on basis matrix. Second, we introduce the SNMFNet, which uses the nonorthogonal matrix factorization algorithm to replace the orthogonal matrix factorization such as PCANet. Experimental results show that our proposed SNMFNet achieves superior performance to PCANet on several benchmark datasets.

Amyloid positron emission tomography (PET) imaging with florbetapir 18F (18F-AV-45) allows in vivo assessment of cerebral amyloid load and can be used in the evaluation of progression of Alzheimer's disease (AD) and other dementias associated with b-amyloid. However, cortical amyloid deposition can occur in healthy cases, as well as in patients with AD and quantification of cortical amyloid burden can improve the 18F-AV-45 PET imaging evaluations. The quantification is mostly performed by cortical-to-cerebellum standardized uptake value ratio (SUVr). The aim of our study was to compare two methods for SUVr calculations in amyloid florbetapir 18F PET brain imaging. In amyloid florbetapir 18F PET brain imaging study, we imaged 42 cases with the mean age of 72.6 ± 9.9 (mean ± standard deviation). They were imaged on different PET/computed tomography systems with 369.0 ± 34.2 kBq of 18F florbetapir. Data were reconstructed using the vendor's reconstruction software. Corresponding magnetic resonance imaging (MRI) data were retrieved, and matched PET and MRI data were transferred to a common platform. Two methods were used for the calculation of the ratio of cortical-to-cerebellar signal (SUVr). One method was based on the MIM Software Inc., Version 6.4 software and only uses PET data. The second approach used the PMOD Neuro tool (version 3.5). This approach utilizes PET and corresponding MRI data (preferably T1-weighted) for better brain segmentation. For all the 42 cases, the average SUVr values for MIM and PMOD applications were 1.24 ± 0.26 and 1.22 ± 0.25, respectively, with a mean difference of 0.02 ± 0.15. The repeatability coefficient was 0.15 (12.3% of the mean). The Spearman's rank correlation coefficient was very high, r = 0.96. For amyloid-negative cases, the average SUVr values were lower than all group SUVr average values, 0.96 ± 0.07 and 1.00 ± 0.09, for MIM and PMOD applications, respectively. A mean difference was 0.04 ± 0.12, the repeatability coefficient was 0.12 (12.9% of the mean) and the Spearman's rank correlation coefficient was modest, r = 0.55. For amyloid-positive patients, the average SUVr values were higher than the same all group values, 1.34 ± 0.16 and 1.35 ± 0.20, respectively, with a mean difference of 0.01 ± 0.16. The repeatability coefficient was 0.16 (11.9% of the mean). The Spearman's rank correlation coefficient was high, r = 0.93. Our results indicated that the SUVr values derived using MIM and PMOD Neuro are effectively interchangeable and well correlated. However, PET template-based quantification (MIM approach) is clinically friendlier and easier to use. MRI template-based quantification (PMOD Neuro) better delineates different regions of the brain, can be used with any tracer, and therefore is more suitable for research.