A novel interpretable deep transfer learning combining diverse learnable parameters for improved T2D prediction based on single-cell gene regulatory networks.

Accurate deep learning (DL) models to predict type 2 diabetes (T2D) are concerned not only with targeting the discrimination task but also with learning useful feature representation. However, existing DL tools are far from perfect and do not provide appropriate interpretation as a guideline to explain and promote superior performance in the target task. Therefore, we provide an interpretable approach for our presented deep transfer learning (DTL) models to overcome such drawbacks, working as follows. We utilize several pre-trained models including SEResNet152, and SEResNeXT101. Then, we transfer knowledge from pre-trained models via keeping the weights in the convolutional base (i.e., feature extraction part) while modifying the classification part with the use of Adam optimizer to deal with classifying healthy controls and T2D based on single-cell gene regulatory network (SCGRN) images. Another DTL models work in a similar manner but just with keeping weights of the bottom layers in the feature extraction unaltered while updating weights of consecutive layers through training from scratch. Experimental results on the whole 224 SCGRN images using five-fold cross-validation show that our model (TFeSEResNeXT101) achieving the highest average balanced accuracy (BAC) of 0.97 and thereby significantly outperforming the baseline that resulted in an average BAC of 0.86. Moreover, the simulation study demonstrated that the superiority is attributed to the distributional conformance of model weight parameters obtained with Adam optimizer when coupled with weights from a pre-trained model.

Application note: TDbasedUFE and TDbasedUFEadv: bioconductor packages to perform tensor decomposition based unsupervised feature extraction.

Motivation: Tensor decomposition (TD)-based unsupervised feature extraction (FE) has proven effective for a wide range of bioinformatics applications ranging from biomarker identification to the identification of disease-causing genes and drug repositioning. However, TD-based unsupervised FE failed to gain widespread acceptance due to the lack of user-friendly tools for non-experts. Results: We developed two bioconductor packages-TDbasedUFE and TDbasedUFEadv-that enable researchers unfamiliar with TD to utilize TD-based unsupervised FE. The packages facilitate the identification of differentially expressed genes and multiomics analysis. TDbasedUFE was found to outperform two state-of-the-art methods, such as DESeq2 and DIABLO. Availability and implementation: TDbasedUFE and TDbasedUFEadv are freely available as R/Bioconductor packages, which can be accessed at https://bioconductor.org/packages/TDbasedUFE and https://bioconductor.org/packages/TDbasedUFEadv, respectively.

Applying Deep Transfer Learning to Assess the Impact of Imaging Modalities on Colon Cancer Detection.

The use of medical images for colon cancer detection is considered an important problem. As the performance of data-driven methods relies heavily on the images generated by a medical method, there is a need to inform research organizations about the effective imaging modalities, when coupled with deep learning (DL), for detecting colon cancer. Unlike previous studies, this study aims to comprehensively report the performance behavior for detecting colon cancer using various imaging modalities coupled with different DL models in the transfer learning (TL) setting to report the best overall imaging modality and DL model for detecting colon cancer. Therefore, we utilized three imaging modalities, namely computed tomography, colonoscopy, and histology, using five DL architectures, including VGG16, VGG19, ResNet152V2, MobileNetV2, and DenseNet201. Next, we assessed the DL models on the NVIDIA GeForce RTX 3080 Laptop GPU (16GB GDDR6 VRAM) using 5400 processed images divided equally between normal colons and colons with cancer for each of the imaging modalities used. Comparing the imaging modalities when applied to the five DL models presented in this study and twenty-six ensemble DL models, the experimental results show that the colonoscopy imaging modality, when coupled with the DenseNet201 model under the TL setting, outperforms all the other models by generating the highest average performance result of 99.1% (99.1%, 99.8%, and 99.1%) based on the accuracy results (AUC, precision, and F1, respectively).

Tensor decomposition discriminates tissues using scATAC-seq.

ATAC-seq is a powerful tool for measuring the landscape structure of a chromosome. scATAC-seq is a recently updated version of ATAC-seq performed in a single cell. The problem with scATAC-seq is data sparsity and most of the genomic sites are inaccessible. Here, tensor decomposition (TD) was used to fill in missing values. In this study, TD was applied to massive scATAC-seq datasets generated by approximately 200 bp intervals, and this number can reach 13,627,618. Currently, no other methods can deal with large sparse matrices. The proposed method could not only provide UMAP embedding that coincides with tissue specificity, but also select genes associated with various biological enrichment terms and transcription factor targeting. This suggests that TD is a useful tool to process a large sparse matrix generated from scATAC-seq.

Principal component analysis- and tensor decomposition-based unsupervised feature extraction to select more suitable differentially methylated cytosines: Optimization of standard deviation versus state-of-the-art methods.

In contrast to RNA-seq analysis, which has various standard methods, no standard methods for identifying differentially methylated cytosines (DMCs) exist. To identify DMCs, we tested principal component analysis and tensor decomposition-based unsupervised feature extraction with optimized standard deviation, which has been shown to be effective for differentially expressed gene (DEG) identification. The proposed method outperformed certain conventional methods, including those that assume beta-binomial distribution for methylation as the proposed method does not require this, especially when applied to methylation profiles measured using high throughput sequencing. DMCs identified by the proposed method also significantly overlapped with various functional sites, including known differentially methylated regions, enhancers, and DNase I hypersensitive sites. The proposed method was applied to data sets retrieved from The Cancer Genome Atlas to identify DMCs using American Joint Committee on Cancer staging system edition labels. This suggests that the proposed method is a promising standard method for identifying DMCs.

A new machine learning based computational framework identifies therapeutic targets and unveils influential genes in pancreatic islet cells.

Pancreatic islets comprise a group of cells that produce hormones regulating blood glucose levels. Particularly, the alpha and beta islet cells produce glucagon and insulin to stabilize blood glucose. When beta islet cells are dysfunctional, insulin is not secreted, inducing a glucose metabolic disorder. Identifying effective therapeutic targets against the disease is a complicated task and is not yet conclusive. To close the wide gap between understanding the molecular mechanism of pancreatic islet cells and providing effective therapeutic targets, we present a computational framework to identify potential therapeutic targets against pancreatic disorders. First, we downloaded three transcriptome expression profiling datasets pertaining to pancreatic islet cells (GSE87375, GSE79457, GSE110154) from the Gene Expression Omnibus database. For each dataset, we extracted expression profiles for two cell types. We then provided these expression profiles along with the cell types to our proposed constrained optimization problem of a support vector machine and to other existing methods, selecting important genes from the expression profiles. Finally, we performed (1) an evaluation from a classification perspective which showed the superiority of our methods against the baseline; and (2) an enrichment analysis which indicated that our methods achieved better outcomes. Results for the three datasets included 44 unique genes and 10 unique transcription factors (SP1, HDAC1, EGR1, E2F1, AR, STAT6, RELA, SP3, NFKB1, and ESR1) which are reportedly related to pancreatic islet functions, diseases, and therapeutic targets.

A tensor decomposition-based integrated analysis applicable to multiple gene expression profiles without sample matching.

The integrated analysis of multiple gene expression profiles previously measured in distinct studies is problematic since missing both sample matches and common labels prevent their integration in fully data-driven, unsupervised training. In this study, we propose a strategy to enable the integration of multiple gene expression profiles among multiple independent studies with neither labeling nor sample matching using tensor decomposition unsupervised feature extraction. We apply this strategy to Alzheimer's disease (AD)-related gene expression profiles that lack precise correspondence among samples, including AD single-cell RNA sequence (scRNA-seq) data. We were able to select biologically reasonable genes using the integrated analysis. Overall, integrated gene expression profiles can function analogously to prior- and/or transfer-learning strategies in other machine-learning applications. For scRNA-seq, the proposed approach significantly reduces the required computational memory.

Adapted tensor decomposition and PCA based unsupervised feature extraction select more biologically reasonable differentially expressed genes than conventional methods.

Tensor decomposition- and principal component analysis-based unsupervised feature extraction were proposed almost 5 and 10 years ago, respectively; although these methods have been successfully applied to a wide range of genome analyses, including drug repositioning, biomarker identification, and disease-causing genes' identification, some fundamental problems have been identified: the number of genes identified was too small to assume that there were no false negatives, and the histogram of P values derived was not fully coincident with the null hypothesis that principal component and singular value vectors follow the Gaussian distribution. Optimizing the standard deviation such that the histogram of P values is as much as possible coincident with the null hypothesis results in an increase in the number and biological reliability of the selected genes. Our contribution was that we improved these methods so as to be able to select biologically more reasonable differentially expressed genes than the state of art methods that must empirically assume negative binomial distributions and dispersion relation, which is required for the selecting more expressed genes than less expressed ones, which can be achieved by the proposed methods that do not have to assume these.

Projection in genomic analysis: A theoretical basis to rationalize tensor decomposition and principal component analysis as feature selection tools.

Identifying differentially expressed genes is difficult because of the small number of available samples compared with the large number of genes. Conventional gene selection methods employing statistical tests have the critical problem of heavy dependence of P-values on sample size. Although the recently proposed principal component analysis (PCA) and tensor decomposition (TD)-based unsupervised feature extraction (FE) has often outperformed these statistical test-based methods, the reason why they worked so well is unclear. In this study, we aim to understand this reason in the context of projection pursuit (PP) that was proposed a long time ago to solve the problem of dimensions; we can relate the space spanned by singular value vectors with that spanned by the optimal cluster centroids obtained from K-means. Thus, the success of PCA- and TD-based unsupervised FE can be understood by this equivalence. In addition to this, empirical threshold adjusted P-values of 0.01 assuming the null hypothesis that singular value vectors attributed to genes obey the Gaussian distribution empirically corresponds to threshold-adjusted P-values of 0.1 when the null distribution is generated by gene order shuffling. For this purpose, we newly applied PP to the three data sets to which PCA and TD based unsupervised FE were previously applied; these data sets treated two topics, biomarker identification for kidney cancers (the first two) and the drug discovery for COVID-19 (the thrid one). Then we found the coincidence between PP and PCA or TD based unsupervised FE is pretty well. Shuffling procedures described above are also successfully applied to these three data sets. These findings thus rationalize the success of PCA- and TD-based unsupervised FE for the first time.

Integrated Analysis of Tissue-Specific Gene Expression in Diabetes by Tensor Decomposition Can Identify Possible Associated Diseases.

In the field of gene expression analysis, methods of integrating multiple gene expression profiles are still being developed and the existing methods have scope for improvement. The previously proposed tensor decomposition-based unsupervised feature extraction method was improved by introducing standard deviation optimization. The improved method was applied to perform an integrated analysis of three tissue-specific gene expression profiles (namely, adipose, muscle, and liver) for diabetes mellitus, and the results showed that it can detect diseases that are associated with diabetes (e.g., neurodegenerative diseases) but that cannot be predicted by individual tissue expression analyses using state-of-the-art methods. Although the selected genes differed from those identified by the individual tissue analyses, the selected genes are known to be expressed in all three tissues. Thus, compared with individual tissue analyses, an integrated analysis can provide more in-depth data and identify additional factors, namely, the association with other diseases.

Novel feature selection method via kernel tensor decomposition for improved multi-omics data analysis.

BACKGROUND: Feature selection of multi-omics data analysis remains challenging owing to the size of omics datasets, comprising approximately [Formula: see text]-[Formula: see text] features. In particular, appropriate methods to weight individual omics datasets are unclear, and the approach adopted has substantial consequences for feature selection. In this study, we extended a recently proposed kernel tensor decomposition (KTD)-based unsupervised feature extraction (FE) method to integrate multi-omics datasets obtained from common samples in a weight-free manner. METHOD: KTD-based unsupervised FE was reformatted as the collection of kernelized tensors sharing common samples, which was applied to synthetic and real datasets. RESULTS: The proposed advanced KTD-based unsupervised FE method showed comparative performance to that of the previously proposed KTD method, as well as tensor decomposition-based unsupervised FE, but required reduced memory and central processing unit time. Moreover, this advanced KTD method, specifically designed for multi-omics analysis, attributes P values to features, which is rare for existing multi-omics-oriented methods. CONCLUSIONS: The sample R code is available at https://github.com/tagtag/MultiR/ .

Effects of Collagen-Glycosaminoglycan Mesh on Gene Expression as Determined by Using Principal Component Analysis-Based Unsupervised Feature Extraction.

The development of the medical applications for substances or materials that contact cells is important. Hence, it is necessary to elucidate how substances that surround cells affect gene expression during incubation. In the current study, we compared the gene expression profiles of cell lines that were in contact with collagen-glycosaminoglycan mesh and control cells. Principal component analysis-based unsupervised feature extraction was applied to identify genes with altered expression during incubation in the treated cell lines but not in the controls. The identified genes were enriched in various biological terms. Our method also outperformed a conventional methodology, namely, gene selection based on linear regression with time course.

Application of Tensor Decomposition to Gene Expression of Infection of Mouse Hepatitis Virus Can Identify Critical Human Genes and Efffective Drugs for SARS-CoV-2 Infection.

To better understand the genes with altered expression caused by infection with the novel coronavirus strain SARS-CoV-2 causing COVID-19 infectious disease, a tensor decomposition (TD)-based unsupervised feature extraction (FE) approach was applied to a gene expression profile dataset of the mouse liver and spleen with experimental infection of mouse hepatitis virus, which is regarded as a suitable model of human coronavirus infection. TD-based unsupervised FE selected 134 altered genes, which were enriched in protein-protein interactions with orf1ab, polyprotein, and 3C-like protease that are well known to play critical roles in coronavirus infection, suggesting that these 134 genes can represent the coronavirus infectious process. We then selected compounds targeting the expression of the 134 selected genes based on a public domain database. The identified drug compounds were mainly related to known antiviral drugs, several of which were also included in those previously screened with an in silico method to identify candidate drugs for treating COVID-19.

The risk of tuberculosis infection in 410 Saudipatients receiving adalimumab therapy.

BACKGROUND: Adalimumab is a fully humanized monoclonal antibody inhibitor of tumor necrosis factor-a used to treat various autoimmune disorders. Adalimumab poses a risk for tuberculosis (TB) infection, especially in countries where TB is endemic. OBJECTIVE: Determine the rate of TB infection after adalimumab therapy in Saudi Arabia. DESIGN: Medical record review. SETTINGS: Tertiary care center in Riyadh. PATIENTS AND METHODS: Demographic and clinical data were retrieved from the electronic healthcare records of all patients who received adalimumab treatment from 2015 to 2019. MAIN OUTCOME MEASURES: Occurrence of TB after adalimumab therapy. SAMPLE SIZE: 410 patients (median ([QR] age, 37 [28], range 4-81 years), 40% males RESULTS: Rheumatoid arthritis was the most frequent indication (n=153, 37%). The patients were followed for a mean of 36 (8.9) months. No case of TB infection or reactivation was observed. An inter-feron-gamma release assay (IGRA) was requested in 353/391 (90.3%) patients, prior to initiating therapy. The IGRA was positive in 26 cases (6.6%). The IGRA-positive patients received isoniazid prophylactically. Bacterial infectious complications of adalimumab therapy occurred in 12 (2.9%) patients. Urinary tract infection was the most frequent complication (culture requested in 48 patients, positive in 8). CONCLUSION: Adalimumab treatment was not associated with a risk of TB disease or TB reactivation in our cohort over the follow-up observation period. No TB reactivation occurred with adalimumab therapy when TB prophylaxis was used. The positive IGRA rate in patients on adalimumab treatment was low (7%). LIMITATIONS: Single center and one geographical area in Saudi Arabia. CONFLICT OF INTEREST: None.

Tensor-Decomposition-Based Unsupervised Feature Extraction in Single-Cell Multiomics Data Analysis.

Analysis of single-cell multiomics datasets is a novel topic and is considerably challenging because such datasets contain a large number of features with numerous missing values. In this study, we implemented a recently proposed tensor-decomposition (TD)-based unsupervised feature extraction (FE) technique to address this difficult problem. The technique can successfully integrate single-cell multiomics data composed of gene expression, DNA methylation, and accessibility. Although the last two have large dimensions, as many as ten million, containing only a few percentage of nonzero values, TD-based unsupervised FE can integrate three omics datasets without filling in missing values. Together with UMAP, which is used frequently when embedding single-cell measurements into two-dimensional space, TD-based unsupervised FE can produce two-dimensional embedding coincident with classification when integrating single-cell omics datasets. Genes selected based on TD-based unsupervised FE are also significantly related to reasonable biological roles.

Unsupervised tensor decomposition-based method to extract candidate transcription factors as histone modification bookmarks in post-mitotic transcriptional reactivation.

The histone group added to a gene sequence must be removed during mitosis to halt transcription during the DNA replication stage of the cell cycle. However, the detailed mechanism of this transcription regulation remains unclear. In particular, it is not realistic to reconstruct all appropriate histone modifications throughout the genome from scratch after mitosis. Thus, it is reasonable to assume that there might be a type of "bookmark" that retains the positions of histone modifications, which can be readily restored after mitosis. We developed a novel computational approach comprising tensor decomposition (TD)-based unsupervised feature extraction (FE) to identify transcription factors (TFs) that bind to genes associated with reactivated histone modifications as candidate histone bookmarks. To the best of our knowledge, this is the first application of TD-based unsupervised FE to the cell division context and phases pertaining to the cell cycle in general. The candidate TFs identified with this approach were functionally related to cell division, suggesting the suitability of this method and the potential of the identified TFs as bookmarks for histone modification during mitosis.

Discriminating the single-cell gene regulatory networks of human pancreatic islets: A novel deep learning application.

Analysis of single-cell pancreatic data can play an important role in understanding various metabolic diseases and health conditions. Due to the sparsity and noise present in such single-cell gene expression data, inference of single-cell gene regulatory networks remains a challenge. Since recent studies have reported the reliable inference of single-cell gene regulatory networks (SCGRNs), the current study focused on discriminating the SCGRNs of T2D patients from those of healthy controls. By accurately distinguishing SCGRNs of healthy pancreas from those of T2D pancreas, it would be possible to annotate, organize, visualize, and identify common patterns of SCGRNs in metabolic diseases. Such annotated SCGRNs could play an important role in accelerating the process of building large data repositories. This study aimed to contribute to the development of a novel deep learning (DL) application. First, we generated a dataset consisting of 224 SCGRNs belonging to both T2D and healthy pancreas and made it freely available. Next, we chose seven DL architectures, including VGG16, VGG19, Xception, ResNet50, ResNet101, DenseNet121, and DenseNet169, trained each of them on the dataset, and checked their prediction based on a test set. Of note, we evaluated the DL architectures on a single NVIDIA GeForce RTX 2080Ti GPU. Experimental results on the whole dataset, using several performance measures, demonstrated the superiority of VGG19 DL model in the automatic classification of SCGRNs, derived from the single-cell pancreatic data.

Novel method for the prediction of drug-drug Interaction based on gene expression profiles.

The accurate prediction of new interactions between drugs is important for avoiding unknown (mild or severe) adverse reactions to drug combinations. The development of effective in silico methods for evaluating drug interactions based on gene expression data requires an understanding of how various drugs alter gene expression. Current computational methods for the prediction of drug-drug interactions (DDIs) utilize data for known DDIs to predict unknown interactions. However, these methods are limited in the absence of known predictive DDIs. To improve DDIs interpretation, a recent study has demonstrated strong non-linear (i.e., dose-dependent) effects of DDIs. In this study, we present a new unsupervised learning approach involving tensor decomposition (TD)-based unsupervised feature extraction (FE) in 3D. We utilize our approach to reanalyze available gene expression profiles for Saccharomyces cerevisiae. We found that non-linearity is possible, even for single drugs. Thus, non-linear dose-dependence cannot always be attributed to DDIs. Our analysis provides a basis for the design of effective methods for evaluating DDIs.

Tensor-Decomposition-Based Unsupervised Feature Extraction Applied to Prostate Cancer Multiomics Data.

The large p small n problem is a challenge without a de facto standard method available to it. In this study, we propose a tensor-decomposition (TD)-based unsupervised feature extraction (FE) formalism applied to multiomics datasets, in which the number of features is more than 100,000 whereas the number of samples is as small as about 100, hence constituting a typical large p small n problem. The proposed TD-based unsupervised FE outperformed other conventional supervised feature selection methods, random forest, categorical regression (also known as analysis of variance, or ANOVA), penalized linear discriminant analysis, and two unsupervised methods, multiple non-negative matrix factorization and principal component analysis (PCA) based unsupervised FE when applied to synthetic datasets and four methods other than PCA based unsupervised FE when applied to multiomics datasets. The genes selected by TD-based unsupervised FE were enriched in genes known to be related to tissues and transcription factors measured. TD-based unsupervised FE was demonstrated to be not only the superior feature selection method but also the method that can select biologically reliable genes. To our knowledge, this is the first study in which TD-based unsupervised FE has been successfully applied to the integration of this variety of multiomics measurements.

A new advanced in silico drug discovery method for novel coronavirus (SARS-CoV-2) with tensor decomposition-based unsupervised feature extraction.

BACKGROUND: COVID-19 is a critical pandemic that has affected human communities worldwide, and there is an urgent need to develop effective drugs. Although there are a large number of candidate drug compounds that may be useful for treating COVID-19, the evaluation of these drugs is time-consuming and costly. Thus, screening to identify potentially effective drugs prior to experimental validation is necessary. METHOD: In this study, we applied the recently proposed method tensor decomposition (TD)-based unsupervised feature extraction (FE) to gene expression profiles of multiple lung cancer cell lines infected with severe acute respiratory syndrome coronavirus 2. We identified drug candidate compounds that significantly altered the expression of the 163 genes selected by TD-based unsupervised FE. RESULTS: Numerous drugs were successfully screened, including many known antiviral drug compounds such as C646, chelerythrine chloride, canertinib, BX-795, sorafenib, sorafenib, QL-X-138, radicicol, A-443654, CGP-60474, alvocidib, mitoxantrone, QL-XII-47, geldanamycin, fluticasone, atorvastatin, quercetin, motexafin gadolinium, trovafloxacin, doxycycline, meloxicam, gentamicin, and dibromochloromethane. The screen also identified ivermectin, which was first identified as an anti-parasite drug and recently the drug was included in clinical trials for SARS-CoV-2. CONCLUSIONS: The drugs screened using our strategy may be effective candidates for treating patients with COVID-19.