Multilevel regularized regression for simultaneous taxa selection and network construction with metagenomic count data

Clustering Single-Cell RNA-Seq Data with Regularized Gaussian Graphical Model

Single-cell RNA-seq (scRNA-seq) is a powerful tool to measure the expression patterns of individual cells and discover heterogeneity and functional diversity among cell populations. Due to variability, it is challenging to analyze such data efficiently. Many clustering methods have been developed using at least one free parameter. Different choices for free parameters may lead to substantially different visualizations and clusters. Tuning free parameters is also time consuming. Thus there is need for a simple, robust, and efficient clustering method. In this paper, we propose a new regularized Gaussian graphical clustering (RGGC) method for scRNA-seq data. RGGC is based on high-order (partial) correlations and subspace learning, and is robust over a wide-range of a regularized parameter λ. Therefore, we can simply set λ=2 or λ=log(p) for AIC (Akaike information criterion) or BIC (Bayesian information criterion) without cross-validation. Cell subpopulations are discovered by the Louvain community detection algorithm that determines the number of clusters automatically. There is no free parameter to be tuned with RGGC. When evaluated with simulated and benchmark scRNA-seq data sets against widely used methods, RGGC is computationally efficient and one of the top performers. It can detect inter-sample cell heterogeneity, when applied to glioblastoma scRNA-seq data.

Sparse Treatment-Effect Model for Taxon Identification with High-Dimensional Metagenomic Data

To identify disease-associated taxa is an important task in metagenomics. To date, many methods have been proposed for feature selection and prediction. However, those proposed methods are either using univariate (generalized) regression approaches to get the corresponding P-values without considering the interactions among taxa, or using lasso or L₀ type sparse modeling approaches to identify taxa with best predictions without providing P-values. To the best of our knowledge, there are no available methods that consider taxon interactions and also generate P-values.In this paper, we propose a treatment-effect model for identifying taxa (STEMIT) and performing statistical inference with high-dimensional metagenomic data. STEMIT will provide a P-value for a taxon through a two-step treatment-effect maximization. It will provide causal inference if the study is a clinical trial. We first identify taxa associated with the treatment-effect variable and the targeting feature with sparse modeling, and then estimate the P-value of the targeting gene with ordinary least square (OLS) regression. We demonstrate that the proposed method is efficient and can identify biologically important taxa with a real metagenomic data set. The software for L₀ sparse modeling can be downloaded at https://cran.r-project.org/web/packages/l0ara/ .

Interstitial Cystitis-Associated Urinary Metabolites Identified by Mass-Spectrometry Based Metabolomics Analysis

This study on interstitial cystitis (IC) aims to identify a unique urine metabolomic profile associated with IC, which can be defined as an unpleasant sensation including pain and discomfort related to the urinary bladder, without infection or other identifiable causes. Although the burden of IC on the American public is immense in both human and financial terms, there is no clear diagnostic test for IC, but rather it is a disease of exclusion. Very little is known about the clinically useful urinary biomarkers of IC, which are desperately needed. Untargeted comprehensive metabolomic profiling was performed using gas-chromatography/mass-spectrometry to compare urine specimens of IC patients or health donors. The study profiled 200 known and 290 unknown metabolites. The majority of the thirty significantly changed metabolites before false discovery rate correction were unknown compounds. Partial least square discriminant analysis clearly separated IC patients from controls. The high number of unknown compounds hinders useful biological interpretation of such predictive models. Given that urine analyses have great potential to be adapted in clinical practice, research has to be focused on the identification of unknown compounds to uncover important clues about underlying disease mechanisms.

Efficient Regularized Regression with L0 Penalty for Variable Selection and Network Construction

Variable selections for regression with high-dimensional big data have found many applications in bioinformatics and computational biology. One appealing approach is the L₀ regularized regression which penalizes the number of nonzero features in the model directly. However, it is well known that L₀ optimization is NP-hard and computationally challenging. In this paper, we propose efficient EM (L₀EM) and dual L₀EM (DL₀EM) algorithms that directly approximate the L₀ optimization problem. While L₀EM is efficient with large sample size, DL₀EM is efficient with high-dimensional (n ≪ m) data. They also provide a natural solution to all L_p   p ∈ [0,2] problems, including lasso with p = 1 and elastic net with p ∈ [1,2]. The regularized parameter λ can be determined through cross validation or AIC and BIC. We demonstrate our methods through simulation and high-dimensional genomic data. The results indicate that L₀ has better performance than lasso, SCAD, and MC+, and L₀ with AIC or BIC has similar performance as computationally intensive cross validation. The proposed algorithms are efficient in identifying the nonzero variables with less bias and constructing biologically important networks with high-dimensional big data.

A simulation framework for correlated count data of features subsets in high-throughput sequencing or proteomics experiments

As part of the data processing of high-throughput-sequencing experiments count data are produced representing the amount of reads that map to specific genomic regions. Count data also arise in mass spectrometric experiments for the detection of protein-protein interactions. For evaluating new computational methods for the analysis of sequencing count data or spectral count data from proteomics experiments artificial count data is thus required. Although, some methods for the generation of artificial sequencing count data have been proposed, all of them simulate single sequencing runs, omitting thus the correlation structure between the individual genomic features, or they are limited to specific structures. We propose to draw correlated data from the multivariate normal distribution and round these continuous data in order to obtain discrete counts. In our approach, the required distribution parameters can either be constructed in different ways or estimated from real count data. Because rounding affects the correlation structure we evaluate the use of shrinkage estimators that have already been used in the context of artificial expression data from DNA microarrays. Our approach turned out to be useful for the simulation of counts for defined subsets of features such as individual pathways or GO categories.

Nonparametric Regularized Regression for Phenotype-Associated Taxa Selection and Network Construction with Metagenomic Count Data

We use a metagenomic approach and network analysis to investigate the relationships between phenotypes across taxa under different environmental conditions. The network structure of taxa can be affected by the disease-associated environmental conditions. In addition, taxa abundance is differentiated under conditions. Therefore, knowing how the correlation or relative abundance changes with these factors would be of great interest to researchers. We develop a nonparametric regularized regression method to construct taxa association networks under different clinical conditions. We let the coefficients be unknown functions of the environmental variable. The varying coefficients are estimated by using regression splines. The proposed method is regularized with concave penalties, and an efficient group descent algorithm is developed for computation. We also apply the varying coefficient model to estimate taxa abundance to see how it changes across different environmental conditions. Moreover, for conducting inference, we propose a bootstrap method to construct the simultaneous confidence bands for the corresponding coefficients. We use different simulated designs and a real data set to demonstrate that our method can identify the network structures successfully under different environmental conditions. As such, the proposed method has potential applications for researchers to construct differential networks and identify taxa.

Zero-Inflated Beta Regression for Differential Abundance Analysis with Metagenomics Data

Metagenomics data have been growing rapidly due to the advances in NGS technologies. One goal of human microbial studies is to detect abundance differences across clinical conditions. Besides small sample size and high dimension, metagenomics data are usually represented as compositions (proportions) with a large number of zeros and skewed distribution. Efficient tools for handling such compositional data need to be developed. We propose a zero-inflated beta regression approach (ZIBSeq) for identifying differentially abundant features between multiple clinical conditions. The proposed method takes the sparse nature of metagenomics data into account and handle the compositional data efficiently. Compared with other available methods, the proposed approach demonstrates better performance with large AUC values for most simulation studies. When applied to a human metagenomics data, it also identifies biologically important taxa reported from previous studies. The software in R is available upon request from the first author.

Network construction and structure detection with metagenomic count data

Background

The human microbiome plays a critical role in human health. Massive amounts of metagenomic data have been generated with advances in next-generation sequencing technologies that characterize microbial communities via direct isolation and sequencing. How to extract, analyze, and transform these vast amounts of data into useful knowledge is a great challenge to bioinformaticians. Microbial biodiversity research has focused primarily on taxa composition and abundance and less on the co-occurrences among different taxa. However, taxa co-occurrences and their relationships to environmental and clinical conditions are important because network structure may help to understand how microbial taxa function together.

Results

We propose a systematic robust approach for bacteria network construction and structure detection using metagenomic count data. Pairwise similarity/distance measures between taxa are proposed by adapting distance measures for samples in ecology. We also extend the sparse inverse covariance approach to a sparse inverse of a similarity matrix from count data for network construction. Our approach is efficient for large metagenomic count data with thousands of bacterial taxa. We evaluate our method with real and simulated data. Our method identifies true and biologically significant network structures efficiently.

Conclusions

Network analysis is crucial for detecting subnetwork structures with metagenomic count data. We developed a software tool in MATLAB for network construction and biologically significant module detection. Software MetaNet can be downloaded from http://biostatistics.csmc.edu/MetaNet/.

Suboptimal cytoreduction in ovarian carcinoma is associated with molecular pathways characteristic of increased stromal activation

OBJECTIVE

Suboptimal cytoreductive surgery in advanced epithelial ovarian cancer (EOC) is associated with poor survival but it is unknown if poor outcome is due to the intrinsic biology of unresectable tumors or insufficient surgical effort resulting in residual tumor-sustaining clones. Our objective was to identify the potential molecular pathway(s) and cell type(s) that may be responsible for suboptimal surgical resection.

METHODS

By comparing gene expression in optimally and suboptimally cytoreduced patients, we identified a gene network associated with suboptimal cytoreduction and explored the biological processes and cell types associated with this gene network.

RESULTS

We show that primary tumors from suboptimally cytoreduced patients express molecular signatures that are typically present in a distinct molecular subtype of EOC characterized by increased stromal activation and lymphovascular invasion. Similar molecular pathways are present in EOC metastases, suggesting that primary tumors in suboptimally cytoreduced patients are biologically similar to metastatic tumors. We demonstrate that the suboptimal cytoreduction network genes are enriched in reactive tumor stroma cells rather than malignant tumor cells.

CONCLUSION

Our data suggest that the success of cytoreductive surgery is dictated by tumor biology, such as extensive stromal reaction and increased invasiveness, which may hinder surgical resection and ultimately lead to poor survival.