Detecting multivariate differentially expressed genes

Divergence of cochlear transcriptomics between reference‑based and reference‑free transcriptome analyses among Rhinolophus ferrumequinum populations

Differences in gene expression within tissues can lead to differences in tissue function. Understanding the transcriptome of a species helps elucidate the molecular mechanisms underlying phenotypic divergence. According to the presence or absence of a reference genome of for a studied species, transcriptome analyses can be divided into reference‑based and reference‑free methods, respectively. Presently, comparisons of complete transcriptome analysis results between those two methods are still rare. In this study, we compared the cochlear transcriptome analysis results of greater horseshoe bats (Rhinolophus ferrumequinum) from three lineages in China with different acoustic phenotypes using reference‑based and reference‑free methods to explore their differences in subsequent analysis. The results gained by reference-based results had lower false-positive rates and were more accurate because differentially expressed genes among the three populations obtained by this method had greater reliability and a higher annotation rate. Some phenotype-related enrichment terms, including those related to inorganic molecules and proton transmembrane channels, were also obtained only by the reference-based method. However, the reference‑based method might have the limitation of incomplete information acquisition. Thus, we believe that a combination of reference‑free and reference‑based methods is ideal for transcriptome analyses. The results of our study provided a reference for the selection of transcriptome analysis methods in the future.

A robust machine learning framework to identify signatures for frailty: a nested case-control study in four aging European cohorts

Phenotype-specific omic expression patterns in people with frailty could provide invaluable insight into the underlying multi-systemic pathological processes and targets for intervention. Classical approaches to frailty have not considered the potential for different frailty phenotypes. We characterized associations between frailty (with/without disability) and sets of omic factors (genomic, proteomic, and metabolomic) plus markers measured in routine geriatric care. This study was a prevalent case control using stored biospecimens (urine, whole blood, cells, plasma, and serum) from 1522 individuals (identified as robust (R), pre-frail (P), or frail (F)] from the Toledo Study of Healthy Aging (R=178/P=184/F=109), 3 City Bordeaux (111/269/100), Aging Multidisciplinary Investigation (157/79/54) and InCHIANTI (106/98/77) cohorts. The analysis included over 35,000 omic and routine laboratory variables from robust and frail or pre-frail (with/without disability) individuals using a machine learning framework. We identified three protective biomarkers, vitamin D3 (OR: 0.81 [95% CI: 0.68-0.98]), lutein zeaxanthin (OR: 0.82 [95% CI: 0.70-0.97]), and miRNA125b-5p (OR: 0.73, [95% CI: 0.56-0.97]) and one risk biomarker, cardiac troponin T (OR: 1.25 [95% CI: 1.23-1.27]). Excluding individuals with a disability, one protective biomarker was identified, miR125b-5p (OR: 0.85, [95% CI: 0.81-0.88]). Three risks of frailty biomarkers were detected: pro-BNP (OR: 1.47 [95% CI: 1.27-1.7]), cardiac troponin T (OR: 1.29 [95% CI: 1.21-1.38]), and sRAGE (OR: 1.26 [95% CI: 1.01-1.57]). Three key frailty biomarkers demonstrated a statistical association with frailty (oxidative stress, vitamin D, and cardiovascular system) with relationship patterns differing depending on the presence or absence of a disability.

Principles for the organization of gene-sets

A gene-set, an important concept in microarray expression analysis and systems biology, is a collection of genes and/or their products (i.e. proteins) that have some features in common. There are many different ways to construct gene-sets, but a systematic organization of these ways is lacking. Gene-sets are mainly organized ad hoc in current public-domain databases, with group header names often determined by practical reasons (such as the types of technology in obtaining the gene-sets or a balanced number of gene-sets under a header). Here we aim at providing a gene-set organization principle according to the level at which genes are connected: homology, physical map proximity, chemical interaction, biological, and phenotypic-medical levels. We also distinguish two types of connections between genes: actual connection versus sharing of a label. Actual connections denote direct biological interactions, whereas shared label connection denotes shared membership in a group. Some extensions of the framework are also addressed such as overlapping of gene-sets, modules, and the incorporation of other non-protein-coding entities such as microRNAs.

Nonlinear-model-based analysis methods for time-course gene expression data

Microarray technology has produced a huge body of time-course gene expression data and will continue to produce more. Such gene expression data has been proved useful in genomic disease diagnosis and drug design. The challenge is how to uncover useful information from such data by proper analysis methods such as significance analysis and clustering analysis. Many statistic-based significance analysis methods and distance/correlation-based clustering analysis methods have been applied to time-course expression data. However, these techniques are unable to account for the dynamics of such data. It is the dynamics that characterizes such data and that should be considered in analysis of such data. In this paper, we employ a nonlinear model to analyse time-course gene expression data. We firstly develop an efficient method for estimating the parameters in the nonlinear model. Then we utilize this model to perform the significance analysis of individually differentially expressed genes and clustering analysis of a set of gene expression profiles. The verification with two synthetic datasets shows that our developed significance analysis method and cluster analysis method outperform some existing methods. The application to one real-life biological dataset illustrates that the analysis results of our developed methods are in agreement with the existing results.

Functional analysis: evaluation of response intensities--tailoring ANOVA for lists of expression subsets

Background

Microarray data is frequently used to characterize the expression profile of a whole genome and to compare the characteristics of that genome under several conditions. Geneset analysis methods have been described previously to analyze the expression values of several genes related by known biological criteria (metabolic pathway, pathology signature, co-regulation by a common factor, etc.) at the same time and the cost of these methods allows for the use of more values to help discover the underlying biological mechanisms.

Results

As several methods assume different null hypotheses, we propose to reformulate the main question that biologists seek to answer. To determine which genesets are associated with expression values that differ between two experiments, we focused on three ad hoc criteria: expression levels, the direction of individual gene expression changes (up or down regulation), and correlations between genes. We introduce the FAERI methodology, tailored from a two-way ANOVA to examine these criteria. The significance of the results was evaluated according to the self-contained null hypothesis, using label sampling or by inferring the null distribution from normally distributed random data. Evaluations performed on simulated data revealed that FAERI outperforms currently available methods for each type of set tested. We then applied the FAERI method to analyze three real-world datasets on hypoxia response. FAERI was able to detect more genesets than other methodologies, and the genesets selected were coherent with current knowledge of cellular response to hypoxia. Moreover, the genesets selected by FAERI were confirmed when the analysis was repeated on two additional related datasets.

Conclusions

The expression values of genesets are associated with several biological effects. The underlying mathematical structure of the genesets allows for analysis of data from several genes at the same time. Focusing on expression levels, the direction of the expression changes, and correlations, we showed that two-step data reduction allowed us to significantly improve the performance of geneset analysis using a modified two-way ANOVA procedure, and to detect genesets that current methods fail to detect.

Dynamic-model-based method for selecting significantly expressed genes from time-course expression profiles

This paper proposes a dynamic-model-based method for selecting significantly expressed (SE) genes from their time-course expression profiles. A gene is considered to be SE if its time-course expression profile is more likely time-dependent than random. The proposed method describes a time-dependent gene expression profile by a nonzero-order autoregressive (AR) model, and a time-independent gene expression profile by a zero-order AR model. Akaike information criterion (AIC) is used to compare the models and subsequently determine whether a time-course gene expression profile is time-independent or time-dependent. The performance of the proposed method is investigated on both a synthetic dataset and a real-life biological dataset in terms of the false discovery rate (FDR) and the false nondiscovery rate (FNR). The results show that the proposed method is valid for selecting SE genes from their time-course expression profiles.

Probability fold change: a robust computational approach for identifying differentially expressed gene lists

Identifying genes that are differentially expressed under different experimental conditions is a fundamental task in microarray studies. However, different ranking methods generate very different gene lists, and this could profoundly impact follow-up analyses and biological interpretation. Therefore, developing improved ranking methods are critical in microarray data analysis. We developed a new algorithm, the probabilistic fold change (PFC), which ranks genes based on a confidence interval estimate of fold change. We performed extensive testing using multiple benchmark data sources including the MicroArray Quality Control (MAQC) data sets. We corroborated our observations with MAQC data sets using qRT-PCR data sets and Latin square spike-in data sets. Along with PFC, we tested six other popular ranking algorithms including Mean Fold Change (FC), SAM, t-statistic (T), Bayesian-t (BAYT), Intensity-Conditional Fold Change (CFC), and Rank Product (RP). PFC achieved reproducibility and accuracy that are consistently among the best of the seven ranking algorithms while other ranking algorithms would show weakness in some cases. Contrary to common belief, our results demonstrated that statistical accuracy will not translate to biological reproducibility and therefore both quality aspects need to be evaluated.

On reliable discovery of molecular signatures

Background

Molecular signatures are sets of genes, proteins, genetic variants or other variables that can be used as markers for a particular phenotype. Reliable signature discovery methods could yield valuable insight into cell biology and mechanisms of human disease. However, it is currently not clear how to control error rates such as the false discovery rate (FDR) in signature discovery. Moreover, signatures for cancer gene expression have been shown to be unstable, that is, difficult to replicate in independent studies, casting doubts on their reliability.

Results

We demonstrate that with modern prediction methods, signatures that yield accurate predictions may still have a high FDR. Further, we show that even signatures with low FDR may fail to replicate in independent studies due to limited statistical power. Thus, neither stability nor predictive accuracy are relevant when FDR control is the primary goal. We therefore develop a general statistical hypothesis testing framework that for the first time provides FDR control for signature discovery. Our method is demonstrated to be correct in simulation studies. When applied to five cancer data sets, the method was able to discover molecular signatures with 5% FDR in three cases, while two data sets yielded no significant findings.

Conclusion

Our approach enables reliable discovery of molecular signatures from genome-wide data with current sample sizes. The statistical framework developed herein is potentially applicable to a wide range of prediction problems in bioinformatics.

Class-specific correlations of gene expressions: identification and their effects on clustering analyses

Current microarray studies primarily focus on identifying individual genes with differential expression levels across different conditions or classes. A potential problem is that they may disregard multidimensional information hidden in gene interactions. In this study, we propose an approach to detect gene interactions related to study phenotypes through identifying gene pairs with correlations that appear to be class or condition specific. In addition, we explore the effects of ignoring class-specific correlations (CSC) on correlation-based gene-clustering analyses. Our simulation studies show that ignoring CSC can significantly decrease the accuracy of gene clustering and increase the dissimilarity within clusters. Our results from a DLBCL (distinct types of diffuse large B cell lymphoma) data set illustrate that CSC are clearly present and have great adverse effects on gene-clustering results if ignored. Meanwhile, interesting biological interpretations may be derived from studying gene pairs with CSC. This study demonstrates that our algorithm is simple and computationally efficient and has the ability to detect gene pairs with CSC that are informative for uncovering interesting regulation patterns.

Cell cycle correlated genes dictate the prognostic power of breast cancer gene lists

Background

Numerous gene lists or "classifiers" have been derived from global gene expression data that assign breast cancers to good and poor prognosis groups. A remarkable feature of these molecular signatures is that they have few genes in common, prompting speculation that they may use distinct genes to measure the same pathophysiological process(es), such as proliferation. However, this supposition has not been rigorously tested. If gene-based classifiers function by measuring a minimal number of cellular processes, we hypothesized that the informative genes for these processes could be identified and the data sets could be adjusted for the predictive contributions of those genes. Such adjustment would then attenuate the predictive function of any signature measuring that same process.

Results

We tested this hypothesis directly using a novel iterative-subtractive approach. We evaluated five gene expression data sets that sample a broad range of breast cancer subtypes. In all data sets, the dominant cluster capable of predicting metastasis was heavily populated by genes that fluctuate in concert with the cell cycle. When six well-characterized classifiers were examined, all contained a higher than expected proportion of genes that correlate with this cluster. Furthermore, when the data sets were globally adjusted for the cell cycle cluster, each classifier lost its ability to assign tumors to appropriate high and low risk groups. In contrast, adjusting for other predictive gene clusters did not impact their performance.

Conclusion

These data indicate that the discriminative ability of breast cancer classifiers is dependent upon genes that correlate with cell cycle progression.