A nested parallel experiment demonstrates differences in intensity-dependence between RNA-seq and microarrays

Development of machine learning models for the prediction of the skin sensitization potential of cosmetic compounds

Background

To enhance the accuracy of allergen detection in cosmetic compounds, we developed a co-culture system that combines HaCaT keratinocytes (transfected with a luciferase plasmid driven by the AKR1C2 promoter) and THP-1 cells for machine learning applications.

Methods

Following chemical exposure, cell cytotoxicity was assessed using CCK-8 to determine appropriate stimulation concentrations. RNA-Seq was subsequently employed to analyze THP-1 cells, followed by differential expression gene (DEG) analysis and weighted gene co-expression net-work analysis (WGCNA). Using two data preprocessing methods and three feature extraction techniques, we constructed and validated models with eight machine learning algorithms.

Results

Our results demonstrated the effectiveness of this integrated approach. The best performing models were random forest (RF) and voom-based diagonal quadratic discriminant analysis (voomDQDA), both achieving 100% accuracy. Support vector machine (SVM) and voom based nearest shrunken centroids (voomNSC) showed excellent performance with 96.7% test accuracy, followed by voom-based diagonal linear discriminant analysis (voomDLDA) at 95.2%. Nearest shrunken centroids (NSC), Poisson linear discriminant analysis (PLDA) and negative binomial linear discriminant analysis (NBLDA) achieved 90.5% and 90.2% accuracy, respectively. K-nearest neighbors (KNN) showed the lowest accuracy at 85.7%.

Conclusion

This study highlights the potential of integrating co-culture systems, RNA-Seq, and machine learning to develop more accurate and comprehensive in vitro methods for skin sensitization testing. Our findings contribute to the advancement of cosmetic safety assessments, potentially reducing the reliance on animal testing.

Identification of a Four-Gene Signature With Prognostic Significance in Endometrial Cancer Using Weighted-Gene Correlation Network Analysis

Endometrial hyperplasia (EH) is a precursor for endometrial cancer (EC). However, biomarkers for the progression from EH to EC and standard prognostic biomarkers for EC have not been identified. In this study, we aimed to identify key genes with prognostic significance for the progression from EH to EC. Weighted-gene correlation network analysis (WGCNA) was used to identify hub genes utilizing microarray data (GSE106191) downloaded from the Gene Expression Omnibus database. Differentially expressed genes (DEGs) were identified from the Uterine Corpus Endometrial Carcinoma (UCEC) dataset of The Cancer Genome Atlas database. The Limma-Voom R package was applied to detect differentially expressed genes (DEGs; mRNAs) between cancer and normal samples. Genes with |log2 (fold change [FC])| > 1.0 and p < 0.05 were considered as DEGs. Univariate and multivariate Cox regression and survival analyses were performed to identify potential prognostic genes using hub genes overlapping in the two datasets. All analyses were conducted using R Bioconductor and related packages. Through WGCNA and overlapping genes in hub modules with DEGs in the UCEC dataset, we identified 42 hub genes. The results of the univariate and multivariate Cox regression analyses revealed that four hub genes, BUB1B, NDC80, TPX2, and TTK, were independently associated with the prognosis of EC (Hazard ratio [95% confidence interval]: 0.591 [0.382–0.912], p = 0.017; 0.605 [0.371–0.986], p = 0.044; 1.678 [1.132–2.488], p = 0.01; 2.428 [1.372–4.29], p = 0.02, respectively). A nomogram was established with a risk score calculated using the four genes’ coefficients in the multivariate analysis, and tumor grade and stage had a favorable predictive value for the prognosis of EC. The survival analysis showed that the high-risk group had an unfavorable prognosis compared with the low-risk group (p < 0.0001). The receiver operating characteristic curves also indicated that the risk model had a potential predictive value of prognosis with area under the curve 0.807 at 2 years, 0.783 at 3 years, and 0.786 at 5 years. We established a four-gene signature with prognostic significance in EC using WGCNA and established a nomogram to predict the prognosis of EC.

ACSNI: An unsupervised machine-learning tool for prediction of tissue-specific pathway components using gene expression profiles

Determining the tissue- and disease-specific circuit of biological pathways remains a fundamental goal of molecular biology. Many components of these biological pathways still remain unknown, hindering the full and accurate characterization of biological processes of interest. Here we describe ACSNI, an algorithm that combines prior knowledge of biological processes with a deep neural network to effectively decompose gene expression profiles (GEPs) into multi-variable pathway activities and identify unknown pathway components. Experiments on public GEP data show that ACSNI predicts cogent components of mTOR, ATF2, and HOTAIRM1 signaling that recapitulate regulatory information from genetic perturbation and transcription factor binding datasets. Our framework provides a fast and easy-to-use method to identify components of signaling pathways as a tool for molecular mechanism discovery and to prioritize genes for designing future targeted experiments (https://github.com/caanene1/ACSNI).

Identification of lncRNA biomarkers for lung cancer through integrative cross-platform data analyses

This study was designed to identify lncRNA biomarker candidates using lung cancer data from RNA-Seq and microarray platforms separately.Lung cancer datasets were obtained from the Gene Expression Omnibus (GEO, n = 287) and The Cancer Genome Atlas (TCGA, n = 216) repositories, only common lncRNAs were used. Differentially expressed (DE) lncRNAs in tumors with respect to normal were selected from the Affymetrix and TCGA datasets. A training model consisting of the top 20 DE Affymetrix lncRNAs was used for validation in the TCGA and Agilent datasets. A second similar training model was generated using the TCGA dataset.First, a model using the top 20 DE lncRNAs from Affymetrix for training and validated using TCGA and Agilent, achieved high prediction accuracy for both training (98.5% AUC for Affymetrix) and validation (99.2% AUC for TCGA and 92.8% AUC for Agilent). A similar model using the top 20 DE lncRNAs from TCGA for training and validated using Affymetrix and Agilent, also achieved high prediction accuracy for both training (97.7% AUC for TCGA) and validation (96.5% AUC for Affymetrix and 80.9% AUC for Agilent). Eight lncRNAs were found to be overlapped from these two lists.

Aging induces aberrant state transition kinetics in murine muscle stem cells

Murine muscle stem cells (MuSCs) experience a transition from quiescence to activation that is required for regeneration, but it remains unknown if the trajectory and dynamics of activation change with age. Here, we use time-lapse imaging and single cell RNA-seq to measure activation trajectories and rates in young and aged MuSCs. We find that the activation trajectory is conserved in aged cells, and we develop effective machine-learning classifiers for cell age. Using cell-behavior analysis and RNA velocity, we find that activation kinetics are delayed in aged MuSCs, suggesting that changes in stem cell dynamics may contribute to impaired stem cell function with age. Intriguingly, we also find that stem cell activation appears to be a random walk-like process, with frequent reversals, rather than a continuous linear progression. These results support a view of the aged stem cell phenotype as a combination of differences in the location of stable cell states and differences in transition rates between them.

Murine single-cell RNA-seq reveals cell-identity- and tissue-specific trajectories of aging

Aging is a pleiotropic process affecting many aspects of mammalian physiology. Mammals are composed of distinct cell type identities and tissue environments, but the influence of these cell identities and environments on the trajectory of aging in individual cells remains unclear. Here, we performed single-cell RNA-seq on >50,000 individual cells across three tissues in young and old mice to allow for direct comparison of aging phenotypes across cell types. We found transcriptional features of aging common across many cell types, as well as features of aging unique to each type. Leveraging matrix factorization and optimal transport methods, we found that both cell identities and tissue environments exert influence on the trajectory and magnitude of aging, with cell identity influence predominating. These results suggest that aging manifests with unique directionality and magnitude across the diverse cell identities in mammals.

Understanding the Bacterial Response to Mycotoxins: The Transcriptomic Analysis of Deoxynivalenol-Induced Changes in Devosia mutans 17-2-E-8

Deoxynivalenol (DON) is a major fusarium toxin widely detected in cereal grains. The inadvertent exposure to this fungal secondary-metabolite gives rise to a myriad of adverse health effects including appetite loss, emesis, and suppression of the immune system. While most of the attention this mycotoxin has gained in the past four decades was related to its eukaryotic toxicity (monogastric animals and plants more precisely), recent studies have begun to reveal its negative influence on prokaryotes. Recently presented evidence indicates that DON can negatively affect many bacterial species, raising the possibility of DON-induced imbalances within the microbiota of the human and animal gut, in addition to other environmental niches. This in turn has led to a greater interest in understanding bacterial responses toward DON, and the involved mechanism(s) and metabolic pathways, in order to build a more comprehensive picture of DON-induced changes in both prokaryotes and eukaryotes alike. This study reveals the transcriptomic profiling of Devosia mutans strain 17-2-E-8 after the inclusion of DON within its growth medium. The results highlight three adaptive mechanisms involved in the response of D. mutans 17-2-E-8 to this mycotoxin, which include: (a) activation of adenosine 5’-triphosphate-binding cassette transporters; (b) engagement of a toxin-specific pyrroloquinoline quinone-dependent detoxification pathway; and finally (c) the upregulation of auxiliary coping proteins such as porins, glutathione S-transferases, and phosphotransferases. Some of the identified mechanisms are universal in nature and are shared with other bacterial genera and species.

Population genomics of parallel evolution in gene expression and gene sequence during ecological adaptation

Natural selection often produces parallel phenotypic changes in response to a similar adaptive challenge. However, the extent to which parallel gene expression differences and genomic divergence underlie parallel phenotypic traits and whether they are decoupled or not remains largely unexplored. We performed a population genomic study of parallel ecological adaptation among replicate ecotype pairs of the rough periwinkle (Littorina saxatilis) at a regional geographical scale (NW Spain). We show that genomic changes underlying parallel phenotypic divergence followed a complex pattern of both repeatable differences and of differences unique to specific ecotype pairs, in which parallel changes in expression or sequence are restricted to a limited set of genes. Yet, the majority of divergent genes were divergent either for gene expression or coding sequence, but not for both simultaneously. Overall, our findings suggest that divergent selection significantly contributed to the process of parallel molecular differentiation among ecotype pairs, and that changes in expression and gene sequence underlying phenotypic divergence could, at least to a certain extent, be considered decoupled processes.

Comparison of alternative approaches for analysing multi-level RNA-seq data

RNA sequencing (RNA-seq) is widely used for RNA quantification in the environmental, biological and medical sciences. It enables the description of genome-wide patterns of expression and the identification of regulatory interactions and networks. The aim of RNA-seq data analyses is to achieve rigorous quantification of genes/transcripts to allow a reliable prediction of differential expression (DE), despite variation in levels of noise and inherent biases in sequencing data. This can be especially challenging for datasets in which gene expression differences are subtle, as in the behavioural transcriptomics test dataset from D. melanogaster that we used here. We investigated the power of existing approaches for quality checking mRNA-seq data and explored additional, quantitative quality checks. To accommodate nested, multi-level experimental designs, we incorporated sample layout into our analyses. We employed a subsampling without replacement-based normalization and an identification of DE that accounted for the hierarchy and amplitude of effect sizes within samples, then evaluated the resulting differential expression call in comparison to existing approaches. In a final step to test for broader applicability, we applied our approaches to a published set of H. sapiens mRNA-seq samples, The dataset-tailored methods improved sample comparability and delivered a robust prediction of subtle gene expression changes. The proposed approaches have the potential to improve key steps in the analysis of RNA-seq data by incorporating the structure and characteristics of biological experiments.

A Joint Bayesian Model for Integrating Microarray and RNA Sequencing Transcriptomic Data

As the sequencing cost continued to drop in the past decade, RNA sequencing (RNA-seq) has replaced microarray to become the standard high-throughput experimental tool to analyze transcriptomic profile. As more and more datasets are generated and accumulated in the public domain, meta-analysis to combine multiple transcriptomic studies to increase statistical power has received increasing popularity. In this article, we propose a Bayesian hierarchical model to jointly integrate microarray and RNA-seq studies. Since systematic fold change differences across RNA-seq and microarray for detecting differentially expressed genes have been previously reported, we replicated this finding in several real datasets and showed that incorporation of a normalization procedure to account for the bias improves the detection accuracy and power. We compared our method with the popular two-stage Fisher's method using simulations and two real applications in a histological subtype (invasive lobular carcinoma) of breast cancer comparing PR+ versus PR- and early-stage versus late-stage patients. The result showed improved detection power and more significant and interpretable pathways enriched in the detected biomarkers from the proposed Bayesian model.

Systems-level analysis of mechanisms regulating yeast metabolic flux

Cellular metabolic fluxes are determined by enzyme activities and metabolite abundances. Biochemical approaches reveal the impact of specific substrates or regulators on enzyme kinetics but do not capture the extent to which metabolite and enzyme concentrations vary across physiological states and, therefore, how cellular reactions are regulated. We measured enzyme and metabolite concentrations and metabolic fluxes across 25 steady-state yeast cultures. We then assessed the extent to which flux can be explained by a Michaelis-Menten relationship between enzyme, substrate, product, and potential regulator concentrations. This revealed three previously unrecognized instances of cross-pathway regulation, which we biochemically verified. One of these involved inhibition of pyruvate kinase by citrate, which accumulated and thereby curtailed glycolytic outflow in nitrogen-limited yeast. Overall, substrate concentrations were the strongest driver of the net rates of cellular metabolic reactions, with metabolite concentrations collectively having more than double the physiological impact of enzymes.

A comparative analytical assay of gene regulatory networks inferred using microarray and RNA-seq datasets

A Gene Regulatory Network (GRN) is a collection of interactions between molecular regulators and their targets in cells governing gene expression level. Omics data explosion generated from high-throughput genomic assays such as microarray and RNA-Seq technologies and the emergence of a number of pre-processing methods demands suitable guidelines to determine the impact of transcript data platforms and normalization procedures on describing associations in GRNs. In this study exploiting publically available microarray and RNA-Seq datasets and a gold standard of transcriptional interactions in Arabidopsis, we performed a comparison between six GRNs derived by RNA-Seq and microarray data and different normalization procedures. As a result we observed that compared algorithms were highly data-specific and Networks reconstructed by RNA-Seq data revealed a considerable accuracy against corresponding networks captured by microarrays. Topological analysis showed that GRNs inferred from two platforms were similar in several of topological features although we observed more connectivity in RNA-Seq derived genes network. Taken together transcriptional regulatory networks obtained by Robust Multiarray Averaging (RMA) and Variance-Stabilizing Transformed (VST) normalized data demonstrated predicting higher rate of true edges over the rest of methods used in this comparison.

Development of a RNA-Seq Based Prognostic Signature in Lung Adenocarcinoma

Background

Precision therapy for lung cancer will require comprehensive genomic testing to identify actionable targets as well as ascertain disease prognosis. RNA-seq is a robust platform that meets these requirements, but microarray-derived prognostic signatures are not optimal for RNA-seq data. Thus, we undertook the first prognostic analysis of lung adenocarcinoma RNA-seq data and generated a prognostic signature.

Methods

Lung adenocarcinoma RNA-seq and clinical data from The Cancer Genome Atlas (TCGA) were divided chronologically into training (n = 255) and validation (n = 157) cohorts. In the training cohort, prognostic association was assessed by univariate Cox analysis. A prognostic signature was built with stepwise multivariable Cox analysis. Outcomes by risk group, stage, and mutation status were analyzed with Kaplan-Meier and multivariable Cox analyses. All the statistical tests were two-sided.

Results

In the training cohort, 96 genes had prognostic association with P values of less than or equal to 1.00x10^-4, including five long noncoding RNAs (lncRNAs). Stepwise regression generated a four-gene signature, including one lncRNA. Signature high-risk cases had worse overall survival (OS) in the TCGA validation cohort (hazard ratio [HR] = 3.07, 95% confidence interval [CI] = 2.00 to 14.62) and a University of Michigan institutional cohort (n = 67; HR = 2.05, 95% CI = 1.18 to 4.55), and worse metastasis-free survival in the TCGA validation cohort (HR = 3.05, 95% CI = 2.31 to 13.37). The four-gene prognostic signature also statistically significantly stratified overall survival in important clinical subsets, including stage I (HR = 2.78, 95% CI = 1.91 to 11.13), EGFR wild-type (HR = 3.01, 95% CI = 1.73 to 14.98), and EGFR mutant (HR = 8.99, 95% CI = 62.23 to 141.44). The four-gene prognostic signature also stood out on top when compared with other prognostic signatures.

Conclusions

Here, we present the first RNA-seq prognostic signature for lung adenocarcinoma that can provide a powerful prognostic tool for precision oncology as part of an integrated RNA-seq clinical sequencing program.

Prognostic significance of PLIN1 expression in human breast cancer

Breast cancer is a heterogeneous disease associated with diverse clinical, biological and molecular features, presenting huge challenges for prognosis and treatment. Here we found that perilipin-1 (PLIN1) mRNA expression is significantly downregulated in human breast cancer. Kaplan-Meier analysis indicated that patients presenting with reduced PLIN1 expression exhibited poorer overall metastatic relapse-free survival (p = 0.03). Further Cox proportional hazard models analysis revealed that the reduced expression of PLIN1 is an independent predictor of overall survival in estrogen receptor positive (p < 0.0001, HR = 0.87, 95% CI = 0.81-0.92, N = 3,600) and luminal A-subtype (p = 0.02, HR = 0.88, 95% CI = 0.78-0.98, N = 1,469) breast cancer patients. We also demonstrated that the exogenous expression of PLIN1 in human breast cancer MCF-7 and MDA-MB-231 cells significantly inhibits cell proliferation, migration, invasion and in vivo tumorigenesis in mice. Together, these data provide novel insights into a prognostic significance of PLIN1 in human breast cancer and reveal a potentially new gene therapy target for breast cancer.

Systems Genetic Analyses Highlight a TGFβ-FOXO3 Dependent Striatal Astrocyte Network Conserved across Species and Associated with Stress, Sleep, and Huntington's Disease

Recent systems-based analyses have demonstrated that sleep and stress traits emerge from shared genetic and transcriptional networks, and clinical work has elucidated the emergence of sleep dysfunction and stress susceptibility as early symptoms of Huntington's disease. Understanding the biological bases of these early non-motor symptoms may reveal therapeutic targets that prevent disease onset or slow disease progression, but the molecular mechanisms underlying this complex clinical presentation remain largely unknown. In the present work, we specifically examine the relationship between these psychiatric traits and Huntington's disease (HD) by identifying striatal transcriptional networks shared by HD, stress, and sleep phenotypes. First, we utilize a systems-based approach to examine a large publicly available human transcriptomic dataset for HD (GSE3790 from GEO) in a novel way. We use weighted gene coexpression network analysis and differential connectivity analyses to identify transcriptional networks dysregulated in HD, and we use an unbiased ranking scheme that leverages both gene- and network-level information to identify a novel astrocyte-specific network as most relevant to HD caudate. We validate this result in an independent HD cohort. Next, we computationally predict FOXO3 as a regulator of this network, and use multiple publicly available in vitro and in vivo experimental datasets to validate that this astrocyte HD network is downstream of a signaling pathway important in adult neurogenesis (TGFβ-FOXO3). We also map this HD-relevant caudate subnetwork to striatal transcriptional networks in a large (n = 100) chronically stressed (B6xA/J)F2 mouse population that has been extensively phenotyped (328 stress- and sleep-related measurements), and we show that this striatal astrocyte network is correlated to sleep and stress traits, many of which are known to be altered in HD cohorts. We identify causal regulators of this network through Bayesian network analysis, and we highlight their relevance to motor, mood, and sleep traits through multiple in silico approaches, including an examination of their protein binding partners. Finally, we show that these causal regulators may be therapeutically viable for HD because their downstream network was partially modulated by deep brain stimulation of the subthalamic nucleus, a medical intervention thought to confer some therapeutic benefit to HD patients. In conclusion, we show that an astrocyte transcriptional network is primarily associated to HD in the caudate and provide evidence for its relationship to molecular mechanisms of neural stem cell homeostasis. Furthermore, we present a unified systems-based framework for identifying gene networks that are associated with complex non-motor traits that manifest in the earliest phases of HD. By analyzing and integrating multiple independent datasets, we identify a point of molecular convergence between sleep, stress, and HD that reflects their phenotypic comorbidity and reveals a molecular pathway involved in HD progression.

Comparative Transcriptomic Approaches Exploring Contamination Stress Tolerance in Salix sp. Reveal the Importance for a Metaorganismal de Novo Assembly Approach for Nonmodel Plants

Metatranscriptomic study of nonmodel organisms requires strategies that retain the highly resolved genetic information generated from model organisms while allowing for identification of the unexpected. A real-world biological application of phytoremediation, the field growth of 10 Salix cultivars on polluted soils, was used as an exemplar nonmodel and multifaceted crop response well-disposed to the study of gene expression. Sequence reads were assembled de novo to create 10 independent transcriptomes, a global transcriptome, and were mapped against the Salix purpurea 94006 reference genome. Annotation of assembled contigs was performed without a priori assumption of the originating organism. Global transcriptome construction from 3.03 billion paired-end reads revealed 606,880 unique contigs annotated from 1588 species, often common in all 10 cultivars. Comparisons between transcriptomic and metatranscriptomic methodologies provide clear evidence that nonnative RNA can mistakenly map to reference genomes, especially to conserved regions of common housekeeping genes, such as actin, α/β-tubulin, and elongation factor 1-α. In Salix, Rubisco activase transcripts were down-regulated in contaminated trees across all 10 cultivars, whereas thiamine thizole synthase and CP12, a Calvin Cycle master regulator, were uniformly up-regulated. De novo assembly approaches, with unconstrained annotation, can improve data quality; care should be taken when exploring such plant genetics to reduce de facto data exclusion by mapping to a single reference genome alone. Salix gene expression patterns strongly suggest cultivar-wide alteration of specific photosynthetic apparatus and protection of the antenna complexes from oxidation damage in contaminated trees, providing an insight into common stress tolerance strategies in a real-world phytoremediation system.

GO-PCA: An Unsupervised Method to Explore Gene Expression Data Using Prior Knowledge

Method

Genome-wide expression profiling is a widely used approach for characterizing heterogeneous populations of cells, tissues, biopsies, or other biological specimen. The exploratory analysis of such data typically relies on generic unsupervised methods, e.g. principal component analysis (PCA) or hierarchical clustering. However, generic methods fail to exploit prior knowledge about the molecular functions of genes. Here, I introduce GO-PCA, an unsupervised method that combines PCA with nonparametric GO enrichment analysis, in order to systematically search for sets of genes that are both strongly correlated and closely functionally related. These gene sets are then used to automatically generate expression signatures with functional labels, which collectively aim to provide a readily interpretable representation of biologically relevant similarities and differences. The robustness of the results obtained can be assessed by bootstrapping.

Results

I first applied GO-PCA to datasets containing diverse hematopoietic cell types from human and mouse, respectively. In both cases, GO-PCA generated a small number of signatures that represented the majority of lineages present, and whose labels reflected their respective biological characteristics. I then applied GO-PCA to human glioblastoma (GBM) data, and recovered signatures associated with four out of five previously defined GBM subtypes. My results demonstrate that GO-PCA is a powerful and versatile exploratory method that reduces an expression matrix containing thousands of genes to a much smaller set of interpretable signatures. In this way, GO-PCA aims to facilitate hypothesis generation, design of further analyses, and functional comparisons across datasets.