Unsupervised multiple kernel learning for heterogeneous data integration.

The diverse microarray datasets that have become available over the past several years represent a rich opportunity and challenge for biological data mining. Many supervised and unsupervised methods have been developed for the analysis of individual microarray datasets. However, integrated analysis of multiple datasets can provide a broader insight into genetic regulation of specific biological pathways under a variety of conditions. To aid in the analysis of such large compendia of microarray experiments, we present Microarray Experiment Functional Integration Technology (MEFIT), a scalable Bayesian framework for predicting functional relationships from integrated microarray datasets. Furthermore, MEFIT predicts these functional relationships within the context of specific biological processes. All results are provided in the context of one or more specific biological functions, which can be provided by a biologist or drawn automatically from catalogs such as the Gene Ontology (GO). Using MEFIT, we integrated 40 Saccharomyces cerevisiae microarray datasets spanning 712 unique conditions. In tests based on 110 biological functions drawn from the GO biological process ontology, MEFIT provided a 5% or greater performance increase for 54 functions, with a 5% or more decrease in performance in only two functions.

BackgroundGene expression profiles based on microarray data have been suggested by many studies as potential molecular prognostic indexes of breast cancer. However, due to the confounding effect of clinical background, independent studies often obtained inconsistent results. The current study investigated the possibility to improve the quality and generality of expression profiles by integrated analysis of multiple datasets. Profiles of recurrence outcome were derived from two independent datasets and validated by a third dataset.ResultsThe clinical background of patients significantly influenced the content and performance of expression profiles when the training samples were unbalanced. The integrated profiling of two independent datasets lead to higher classification accuracy (71.11% vs. 70.59%) and larger ROC curve area (0.789 vs. 0.767) of the testing samples. Cell cycle, especially M phase mitosis, was significantly overrepresented by the 60-gene profile obtained from integrated analysis (p < 0.0001). This profiles significantly differentiated poor and good prognosis in a third patient cohort (p = 0.003). Simulation procedures demonstrated that the change of profile specificity had more instant influence on the performance of expression profiles than the change of profile sensitivity.ConclusionThe current study confirmed that the gene expression profile generated by integrated analysis of multiple datasets achieved better prediction of breast cancer recurrence. However, the content and performance of profiles was confounded by clinical background of training patients. In future studies, prognostic profile applicable to the general population should be derived from more diversified and balanced patient cohorts in larger scale.

Single-cell RNA-seq (scRNA-seq) has been widely used to resolve cellular heterogeneity. After collecting scRNA-seq data, the natural next step is to integrate the accumulated data to achieve a common ontology of cell types and states. Thus, an effective and efficient cell-type identification method is urgently needed. Meanwhile, high-quality reference data remain a necessity for precise annotation. However, such tailored reference data are always lacking in practice. To address this, we aggregated multiple datasets into a meta-dataset on which annotation is conducted. Existing supervised or semi-supervised annotation methods suffer from batch effects caused by different sequencing platforms, the effect of which increases in severity with multiple reference datasets. Herein, a robust deep learning-based single-cell Multiple Reference Annotator (scMRA) is introduced. In scMRA, a knowledge graph is constructed to represent the characteristics of cell types in different datasets, and a graphic convolutional network serves as a discriminator based on this graph. scMRA keeps intra-cell-type closeness and the relative position of cell types across datasets. scMRA is remarkably powerful at transferring knowledge from multiple reference datasets, to the unlabeled target domain, thereby gaining an advantage over other state-of-the-art annotation methods in multi-reference data experiments. Furthermore, scMRA can remove batch effects. To the best of our knowledge, this is the first attempt to use multiple insufficient reference datasets to annotate target data, and it is, comparatively, the best annotation method for multiple scRNA-seq datasets. An implementation of scMRA is available from https://github.com/ddb-qiwang/scMRA-torch. Supplementary data are available at Bioinformatics online.

Accurate disease phenotype prediction plays an important role in the treatment of heterogeneous diseases like cancer in the era of precision medicine. With the advent of high throughput technologies, more comprehensive multi-omics data is now available that can effectively link the genotype to phenotype. However, the interactive relation of multi-omics datasets makes it particularly challenging to incorporate different biological layers to discover the coherent biological signatures and predict phenotypic outcomes. In this study, we introduce omicsGAN, a generative adversarial network model to integrate two omics data and their interaction network. The model captures information from the interaction network as well as the two omics datasets and fuse them to generate synthetic data with better predictive signals. Large-scale experiments on The Cancer Genome Atlas breast cancer, lung cancer and ovarian cancer datasets validate that (i) the model can effectively integrate two omics data (e.g. mRNA and microRNA expression data) and their interaction network (e.g. microRNA-mRNA interaction network). The synthetic omics data generated by the proposed model has a better performance on cancer outcome classification and patients survival prediction compared to original omics datasets. (ii) The integrity of the interaction network plays a vital role in the generation of synthetic data with higher predictive quality. Using a random interaction network does not allow the framework to learn meaningful information from the omics datasets; therefore, results in synthetic data with weaker predictive signals. Source code is available at: https://github.com/CompbioLabUCF/omicsGAN. Supplementary data are available at Bioinformatics online.

In studies that aim to model the relationship between an outcome variable and multiple omics datasets, it is often desirable to reduce the dimensionality of these datasets or to represent one omics dataset in terms of another. Several approaches exist for this purpose, including univariate methods such as polygenic scores, and multivariate methods. Multivariate approaches offer advantages by producing lower-dimensional integrative scores, capturing joint structures across datasets, and filtering out dataset-specific noise. In this paper, we describe one univariate and two multivariate methods, and evaluate their performance through simulations involving two correlated multivariate normally distributed omics datasets, as well as a combination of one multivariate normal and one fixed categorical dataset. We assess method performance using the root mean squared error (RMSE) when modelling the outcome variable as a function of the reduced omics representations. Multivariate methods generally perform well, particularly when a slightly higher number of components is used for integration. They outperform the univariate method in scenarios involving two normally distributed omics datasets and perform comparably in settings with one normal and one categorical dataset. In real data applications, including two metabolomics datasets from TwinsUK and a metabolomics-genetic dataset from ORCADES, all methods show similar performance in modelling body mass index. Multivariate methods provide a valuable framework for summarizing multi-omics datasets into low-dimensional components suitable for outcome modelling. Even in the presence of non-normal data, these methods offer a promising alternative to high-dimensional univariate approaches.

In recent years, rapidly decreasing sequencing costs have lead to an exponential growth in the availability of both bulk and single-cell transcriptomic data. Compilation of datasets generated from different research groups would allow for many additional or improved analyses, both due to increased computational power and accessibility. However, the process of accurately and effectively integrating datasets in an easily interpretable manner is challenging, and there is an unmet need for additional tools that incorporate multiple datasets in an effective and intuitive manner. Pancreatic ductal adenocarcinoma (PDA) is the third most common cause of cancer mortality, and there are several web portals, including cBioportal or TCGA, that integrate and analyze multi-omics data from multiple large-scale projects. However, these tools fail to integrate datasets generated by independent laboratories and thus leave a large portion of the available PDA data untapped. An tool that incorporates these data into a single easy-to-use source is not only useful but necessary to the PDA research community. Here we present PADCAS (PAncreatic Ductal Cancer AnalysiS), an intuitive tool that allows researchers and clinicians rapid access to multiple datasets related to PDA. Using a friendly visualization and easy-to-use exploratory panels, PADCAS allows the user to perform various statistical analyses investigating the expression of one or multiple genes of interest in normal tissue, primary tumor, or metastatic samples in both bulk and single-cell RNA sequencing data. The tool integrates four public single-cell data sets where users can analyze gene(s) of interest in different context, such as across cell types (fibroblasts vs. macrophages vs. epithelial cells, for example) or tissue conditions. Additionally, regulatory network analyses have been performed and incorporated into the tool for a subset of the datasets, allowing users to investigate and visualize the activity of key proteins in both the ductal and stroma compartments. A complementary novel dataset generated from 23 PDA cell lines is also included with similar analysis options at the gene expression or protein activity level. PADCAS also incorporates bulk RNAseq data generated using laser capture microdissection (LCM-RNA-Seq) at Columbia University. LCM-RNA-Seq dataset includes human PDA, and low grade samples PanIN-1 and IPMN-adenomas), with cross-condition analyses performed at the gene expression level including survival analysis. Finally, all plots generated by PADCAS can be easily downloaded in different formats and sizes. The tool has been designed to be easily maintained and extended as long as new datasets and functionalities are required. Thus, PADCAS represents the first public, free-of-programming resource compilating multiple omics datasets, including single cell RNAseq, for exploratory analysis in PDAC Citation Format: Alvaro Curiel-Garcia, Lorenzo Tomassoni, Lukas J. Vlahos, Carlo H. Maurer, Andrea Califano, Kenneth P. Olive. PADCAS: A friendly web page for exploratory analysis of pancreatic ductal adenocarcinoma datasets [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 1902.

ObjectiveHuman endogenous retroviruses (HERVs) make up 8% of the human genome. HERVs are biologically active elements related to multiple diseases. HERV-K, a subfamily of HERVs, has been associated with certain types of cancer and suggested as an immunologic target in some tumors. The expression levels of HERV-K in breast cancer (BCa) have been studied as biomarkers and immunologic therapeutic targets. However, HERV-K has multiple copies in the human genome, and few studies determined the transcriptional profile of HERV-K copies across the human genome for BCa.MethodsNinety-one HERV-K indexes with entire proviral sequences were used as the reference database. Nine raw sequencing datasets with 243 BCa and 137 control samples were mapped to this database by Salmon software. The differential proviral expression across several groups was analyzed by DESeq2 software.ResultsFirst, the clustering of each dataset demonstrated that these 91 HERV-K proviruses could well cluster the BCa and control samples when the normal controls were normal cells or healthy donor tissues. Second, several common HERV-K proviruses that are closely related with BCa risk were significantly differentially expressed (padj < 0.05 and absolute log2FC > 1.5) in the tissues and cell lines. Additionally, almost all the HERV-K proviruses had higher expression in BCa tissue than in healthy donor tissue. Notably, we first found the expression of 17p13.1 provirus that located with TP53 should regulate TP53 expression in ER+ and HER2+ BCa.ConclusionThe expression profiling of these 91 HERV-K proviruses can be used as biomarkers to distinguish individuals with BCa and healthy controls. Some proviruses, especially 17p13.1, were strongly associated with BCa risk. The results suggest that HERV-K expression profiles may be appropriate biomarkers and targets for BCa.

Chromatin immunoprecipitation followed by next-generation sequencing (ChIP-Seq) has opened new avenues of research in the genome-wide characterization of regulatory DNA-protein interactions at the genetic and epigenetic level. As a consequence, it has become the de facto standard for studies on the regulation of transcription, and literally thousands of data sets for transcription factors and cofactors in different conditions and species are now available to the scientific community. However, while pipelines and best practices have been established for the analysis of a single experiment, there is still no consensus on the best way to perform an integrated analysis of multiple datasets in the same condition, in order to identify the most relevant and widespread regulatory modules composed by different transcription factors and cofactors. We present here a computational pipeline for this task, that integrates peak summit colocalization, a novel statistical framework for the evaluation of its significance, and motif enrichment analysis. We show examples of its application to ENCODE data, that led to the identification of relevant regulatory modules composed of different factors, as well as the organization on DNA of the binding motifs responsible for their recruitment.

Granulosa cells (GCs) play crucial roles in oocyte maturation. Through gap junctions and extracellular vesicles, they mediate the exchange of molecules such as microRNAs and messenger RNAs. Different ovarian cell types exhibit unique gene expression profiles, reflecting their specialized functions and stages. By combining RNA-seq data from various cell types forming the follicle, we aimed at capturing a wide range of expression patterns, offering insights into the functional diversity and complexity of the transcriptome regulation across GCs. Herein, we performed an integrated bioinformatics analysis of RNA sequencing datasets present in public databases, with a unique and standardized workflow., By combining the data from different studies, we successfully increased the robustness and reliability of our findings and discovered novel genes, miRNAs, and signaling pathways associated with GCs function and oocyte maturation. Moreover, our results provide a valuable resource for further wet-lab research on GCs biology and their impact on oocyte development and competence.

Conventional kernel principal component analysis (KPCA) may not function well for nonlinear process monitoring, since the Gaussian assumption of the method may be violated through nonlinear and kernel transformation of the original process data. To overcome this deficiency, a statistical local approach has been incorporated into KPCA. Through this method, a new score variable which was called improved residual in the statistical local approach is constructed. This new variable approximately follows Gaussian distribution, in spite of which distribution the original data follow. Like the traditional method, two statistics can be constructed for process monitoring, with their corresponding confidence limits determined by a $${{\chi }^{2}}$$ distribution. Besides of the improvement made on KPCA, the joint local approach-KPCA method also shows superiority on detection sensitivity, especially for small faults slow changes of the process.KeywordsKernal principal component analysisNonlinear process monitoringNon-GaussianLocal approach

The integration of multiple omics datasets measured on the same samples is a challenging task: data come from heterogeneous sources and vary in signal quality. In addition, some omics data are inherently compositional, e.g. sequence count data. Most integrative methods are limited in their ability to handle covariates, missing values, compositional structure and heteroscedasticity. In this article we introduce a flexible model-based approach to data integration to address these current limitations: COMBI. We combine concepts, such as compositional biplots and log-ratio link functions with latent variable models, and propose an attractive visualization through multiplots to improve interpretation. Using real data examples and simulations, we illustrate and compare our method with other data integration techniques. Our algorithm is available in the R-package combi.

R-loop-driven molecular subtypes reveal prognostic and immunogenomic features in uterine corpus endometrial carcinoma.

Background Lung adenocarcinoma (LUAD) is the most prevalent tumor in lung carcinoma cases and threatens human life seriously worldwide. Here we attempt to identify a prognostic biomarker and potential therapeutic target for LUAD patients. Methods Differentially expressed genes (DEGs) shared by GSE18842, GSE75037, GSE101929 and GSE19188 profiles were determined and used for protein-protein interaction analysis, enrichment analysis and clinical correlation analysis to search for the core gene, whose expression was further validated in multiple databases and LUAD cells (A549 and PC-9) by quantitative real-time PCR (qRT-PCR) and western blot analyses. Its prognostic value was estimated using the Kaplan-Meier method, meta-analysis and Cox regression analysis based on the Cancer Genome Atlas (TCGA) dataset and co-expression analysis was conducted using the Oncomine database. Gene Set Enrichment Analysis (GSEA) was performed to illuminate the potential functions of the core gene. Results A total of 115 shared DEGs were found, of which 24 DEGs were identified as candidate hub genes with potential functions associated with cell cycle and FOXM1 transcription factor network. Among these candidates, HMMR was identified as the core gene, which was highly expressed in LUAD as verified by multiple datasets and cell samples. Besides, high HMMR expression was found to independently predict poor survival in patients with LUAD. Co-expression analysis showed that HMMR was closely related to FOXM1 and was mainly involved in cell cycle as suggested by GSEA. Conclusion HMMR might be served as an independent prognostic biomarker for LUAD patients, which needs further validation in subsequent studies.

A prognostic model of bladder cancer was constructed based on costimulatory molecules, and its stability and accuracy were verified in different datasets. The expression profile of bladder cancer RNA and the corresponding clinical data in The Cancer Genome Atlas (TCGA) database were analyzed employing computational biology, and a prognostic model was constructed for costimulating molecule-related genes. The model was applied in GSE160693, GSE176307, Xiangya_Cohort, GSE13507, GSE19423, GSE31684, GSE32894, GSE48075, GSE69795 and GSE70691 in TCGA dataset and Gene Expression Omnibus database. The role of costimulating molecules in bladder cancer tumor subtypes was also explored. By consistent cluster analysis, bladder cancer in the TCGA dataset was categorized into two subtypes: C1 and C2. The C1 subtype exhibited a poor prognosis, high levels of immune cell infiltration and significant enrichment of natural killer cells, T cells and dendritic cells in the C1 subtype. In addition, the ImmuneScore calculated by the ESTIMATE algorithm differed greatly between the two subtypes, and the ImmuneScore of the C1 subtype was greater than the C2 subtype in a significant manner. This study also assessed the relationship between costimulating molecules and immunotherapy response. The high-risk group responded poorly to immunotherapy, with significant differences in the amount of most immune cells between the two groups. Further, three indices of the ESTIMATE algorithm and 22 immune cells of the CIBERSORT algorithm were significantly correlated with risk values. These findings suggest the potential value of costimulating molecules in predicting immunotherapy response. A costimulatory molecule-based prognostic model for bladder cancer was established and validated across multiple datasets. This model introduces a novel mode for tailoring treatments to each individual with bladder cancer, and offers valuable insights for informed clinical choices. Simultaneously, this research also delved into the significance of costimulating molecules within distinct bladder cancer subtypes, shedding novel insights into improving immunotherapy strategies for the treatment of bladder cancer.

Abnormal lipid metabolism often occurs under hypoxic microenvironment, which is an important energy supplement for cancer cell proliferation and metastasis. We aimed to explore the lipid metabolism characteristics and gene expression features of pancreatic ductal adenocarcinoma (PDAC) related to hypoxia and identify biomarkers for molecular classification based on hypoxic lipid metabolism that are evaluable for PDAC prognosis and therapy. The multiple datasets were analyzed integratively, including corresponding clinical information of samples. PDAC possesses a distinct metabolic profile and oxygen level compared with normal pancreatic tissues, according to the bioinformatics methods. In addition, a study on untargeted metabolomics using Ultra Performance Liquid Chromatography Tandem Mass Spectrometry(UPLC-MS) revealed lipid metabolites differences affected by oxygen. Analysis of PDAC gene expression profiling in The Cancer Genome Atlas (TCGA) revealed that the sphingolipid process correlates closely with HIF1α. According to the characters of HIF-1 and sphingolipid, samples can be clustered into three subgroups using non-negative matrix factorization clustering. In cluster2, patients had an increased survival time. Relatively high MUC16 mutation arises in cluster2 and may positively influence the cancer survival rates. This study explored the expression pattern of lipid metabolism under hypoxia microenvironment in PDAC. On the basis of metabolic signatures, we identified the prognosis subtypes linking lipid metabolism to hypoxia. The classifications may be conducive to developing personalized treatment programs targeting metabolic profiles.