Search

Abstract The early and highly accurate prediction of COVID-19 based on medical images can speed up the diagnostic process and thereby mitigate disease spread; therefore, developing AI-based models is an inevitable endeavor. The presented work, to our knowledge, is the first to expand the model space and identify a better performing model among 10,000 constructed deep transfer learning (DTL) models as follows. First, we downloaded and processed 4481 CT and X-ray images pertaining to COVID-19 and non-COVID-19 patients, obtained from the Kaggle repository. Second, we provide processed images as inputs to four pre-trained deep learning models (ConvNeXt, EfficientNetV2, DenseNet121, and ResNet34) on more than a million images from the ImageNet database, in which we froze the convolutional and pooling layers pertaining to the feature extraction part while unfreezing and training the densely connected classifier with the Adam optimizer. Third, we generate and take a majority vote of two, three, and four combinations from the four DTL models, resulting in $$\sum\nolimits_{r = 2}^{4} {\left( {\begin{array}{*{20}c} 4 \\ r \\ \end{array} } \right)} = 11$$ DTL models. Then, we combine the 11 DTL models, followed by consecutively generating and taking the majority vote of $$\sum\nolimits_{r = 2}^{11} {\left( {\begin{array}{*{20}c} {11} \\ r \\ \end{array} } \right)} = 2036$$ DTL models. Finally, we select $$7953$$ DTL models from $$\left( {\begin{array}{*{20}c} {2036} \\ 2 \\ \end{array} } \right).$$ Experimental results from the whole datasets using five-fold cross-validation demonstrate that the best generated DTL model, named HC, achieving the best AUC of 0.909 when applied to the CT dataset, while ConvNeXt yielded a higher marginal AUC of 0.933 compared to 0.93 for HX when considering the X-ray dataset. These promising results set the foundation for promoting the large generation of models (LGM) in AI.

Abstract 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.

Understanding breast cancer drug response mechanisms can play a crucial role in improving treatment outcomes and survival rates. Existing bioinformatics-based approaches are far from perfect and do not adopt computational methods based on advanced artificial intelligence concepts. Therefore, we introduce a novel computational framework based on an efficient support vector machine (esvm) working as follows: First, we downloaded and processed three gene expression datasets related to breast cancer responding and non-responding to treatments from the gene expression omnibus (GEO) according to the following GEO accession numbers: GSE130787, GSE140494, and GSE196093. Our method esvm is formulated as a constrained optimization problem in its dual form as a function of λ. We recover the importance of each gene as a function of λ, y, and x. Then, we select p genes out of n, which are provided as input to enrichment analysis tools, Enrichr and Metascape. Compared to existing baseline methods, including deep learning, results demonstrate the superiority and efficiency of esvm, achieving high-performance results and having more expressed genes in well-established breast cancer cell lines, including MD-MB231, MCF7, and HS578T. Moreover, esvm is able to identify (1) various drugs, including clinically approved ones (e.g., tamoxifen and erlotinib); (2) seventy-four unique genes (including tumor suppression genes such as TP53 and BRCA1); and (3) thirty-six unique TFs (including SP1 and RELA). These results have been reported to be linked to breast cancer drug response mechanisms, progression, and metastasizing. Our method is available publicly on the maGENEgerZ web server.

Transmission electron microscopy imaging provides a unique opportunity to inspect the detailed structure of infected lung cells with SARS-CoV-2. Unlike previous studies, this novel study aims to investigate COVID-19 classification at the lung cellular level in response to SARS-CoV-2. Particularly, differentiating between healthy and infected human alveolar type II (hAT2) cells with SARS-CoV-2. Hence, we explore the feasibility of deep transfer learning (DTL) and introduce a highly accurate approach that works as follows: First, we downloaded and processed 286 images pertaining to healthy and infected hAT2 cells obtained from the electron microscopy public image archive. Second, we provided processed images to two DTL computations to induce ten DTL models. The first DTL computation employs five pre-trained models (including DenseNet201 and ResNet152V2) trained on more than one million images from the ImageNet database to extract features from hAT2 images. Then, it flattens and provides the output feature vectors to a trained, densely connected classifier with the Adam optimizer. The second DTL computation works in a similar manner, with a minor difference in that we freeze the first layers for feature extraction in pre-trained models while unfreezing and jointly training the next layers. The results using five-fold cross-validation demonstrated that TFeDenseNet201 is 12.37× faster and superior, yielding the highest average ACC of 0.993 (F1 of 0.992 and MCC of 0.986) with statistical significance (P<2.2×10−16 from a t-test) compared to an average ACC of 0.937 (F1 of 0.938 and MCC of 0.877) for the counterpart (TFtDenseNet201), showing no significance results (P=0.093 from a t-test).

Abstract 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.

Abstract 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.

MotivationTensor 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.ResultsWe 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 implementationTDbasedUFE 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.

Melanoma skin cancer is one of the most dangerous types of skin cancer, which, if not diagnosed early, may lead to death. Therefore, an accurate diagnosis is needed to detect melanoma. Traditionally, a dermatologist utilizes a microscope to inspect and then provide a report on a biopsy for diagnosis; however, this diagnosis process is not easy and requires experience. Hence, there is a need to facilitate the diagnosis process while still yielding an accurate diagnosis. For this purpose, artificial intelligence techniques can assist the dermatologist in carrying out diagnosis. In this study, we considered the detection of melanoma through deep learning based on cutaneous image processing. For this purpose, we tested several convolutional neural network (CNN) architectures, including DenseNet201, MobileNetV2, ResNet50V2, ResNet152V2, Xception, VGG16, VGG19, and GoogleNet, and evaluated the associated deep learning models on graphical processing units (GPUs). A dataset consisting of 7146 images was processed using these models, and we compared the obtained results. The experimental results showed that GoogleNet can obtain the highest performance accuracy on both the training and test sets (74.91% and 76.08%, respectively).

Abstract Background Feature selection of multi-omics data analysis remains challenging owing to the size of omics datasets, comprising approximately $$10^2$$ 10 2 – $$10^5$$ 10 5 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/ .

It is difficult to identify histone modification from datasets that contain high-throughput sequencing data. Although multiple methods have been developed to identify histone modification, most of these methods are not specific to histone modification but are general methods that aim to identify protein binding to the genome. In this study, tensor decomposition (TD) and principal component analysis (PCA)-based unsupervised feature extraction with optimized standard deviation were successfully applied to gene expression and DNA methylation. The proposed method was used to identify histone modification. Histone modification along the genome is binned within the region of length L. Considering principal components (PCs) or singular value vectors (SVVs) that PCA or TD attributes to samples, we can select PCs or SVVs attributed to regions. The selected PCs and SVVs further attribute p-values to regions, and adjusted p-values are used to select regions. The proposed method identified various histone modifications successfully and outperformed various state-of-the-art methods. This method is expected to serve as a de facto standard method to identify histone modification. For reproducibility and to ensure the systematic analysis of our study is applicable to datasets from different gene expression experiments, we have made our tools publicly available for download from gitHub.

Integration of gene expression (GE) and protein–protein interaction (PPI) is not straightforward because the former is provided as a matrix, whereas the latter is provided as a network. In many cases, genes processed with GE analysis are refined further based on a PPI network or vice versa. This is hardly regarded as a true integration of GE and PPI. To address this problem, we proposed a tensor decomposition (TD)-based method that can integrate GE and PPI prior to any analyses where PPI is also formatted as a matrix to which singular value decomposition (SVD) is applied. Integrated analyses with TD improved the coincidence between vectors attributed to samples and class labels over 27 cancer types retrieved from The Cancer Genome Atlas Program (TCGA) toward five class labels. Enrichment using genes selected with this strategy was also improved with the integration using TD. The PPI network associated with the information on the strength of the PPI can improve the performance than PPI that stores only if the interaction exists in individual pairs. In addition, even restricting genes to the intersection of GE and PPI can improve coincidence and enrichment.

Overlooked diseases in agriculture severely impact crop growth, which results in significant losses for farmers. Unfortunately, manual field visits for plant disease diagnosis (PDD) are costly and time consuming. Although various methods of PDD have been proposed, many challenges have yet to be investigated, such as early stage leaf disease diagnosis, class variations in diseases, cluttered backgrounds, and computational complexity of the diagnosis system. In this paper, we propose a Convolutional Neural Network (CNN)-based PDD framework (i.e., PDD-Net), which employs data augmentation techniques and incorporates multilevel and multiscale features to create a class and scale-invariant architecture. The Flatten-T Swish (FTS) activation function is utilized to prevent gradient vanishing and exploding problems, while the focal loss function is used to mitigate the impact of class imbalance during PDD-Net training. The PDD-Net method outperforms baseline models, achieving an average precision of 92.06%, average recall of 92.71%, average F1 score of 92.36%, and accuracy of 93.79% on the PlantVillage dataset. It also achieves an average precision of 86.41%, average recall of 85.77%, average F1 score of 86.02%, and accuracy of 86.98% on the cassava leaf disease dataset. These results demonstrate the efficiency and robustness of PDD-Net in plant disease diagnosis.

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).

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.

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.

A plethora of negative behavioural activities have recently been found in social media. Incidents such as trolling and hate speech on social media, especially on Twitter, have grown considerably. Therefore, detection of hate speech on Twitter has become an area of interest among many researchers. In this paper, we present a computational framework to (1) examine out the computational challenges behind hate speech detection and (2) generate high performance results. First, we extract features from Twitter data by utilizing a count vectorizer technique. Then, we provide the labeled dataset of constructed features to adopted ensemble methods, including Bagging, AdaBoost, and Random Forest. After training, we classify new tweet examples into one of the two categories, hate speech or non-hate speech. Experimental results show (1) that Random Forest has surpassed other methods by generating 95% using accuracy performance results and (2) word cloud displays the most prominent tweets that are responsible for hateful sentiments.

Gene expression profiles of tissues treated with drugs have recently been used to infer clinical outcomes. Although this method is often successful from the application point of view, gene expression altered by drugs is rarely analyzed in detail, because of the extremely large number of genes involved. Here, we applied tensor decomposition (TD)-based unsupervised feature extraction (FE) to the gene expression profiles of 24 mouse tissues treated with 15 drugs. TD-based unsupervised FE enabled identification of the common effects of 15 drugs including an interesting universal feature: these drugs affect genes in a gene-group-wide manner and were dependent on three tissue types (neuronal, muscular, and gastroenterological). For each tissue group, TD-based unsupervised FE enabled identification of a few tens to a few hundreds of genes affected by the drug treatment. These genes are distinctly expressed between drug treatments and controls as well as between tissues in individual tissue groups and other tissues. We also validated the assignment of genes to individual tissue groups using multiple enrichment analyses. We conclude that TD-based unsupervised FE is a promising method for integrated analysis of gene expression profiles from multiple tissues treated with multiple drugs in a completely unsupervised manner.