Tissue classification with gene expression profiles.

Constantly improving gene expression profiling technologies are expected to provide understanding and insight into cancer-related cellular processes. Gene expression data is also expected to significantly aid in the development of efficient cancer diagnosis and classification platforms. In this work we examine three sets of gene expression data measured across sets of tumor(s) and normal clinical samples: The first set consists of 2,000 genes, measured in 62 epithelial colon samples (Alon et al., 1999). The second consists of approximately equal to 100,000 clones, measured in 32 ovarian samples (unpublished extension of data set described in Schummer et al. (1999)). The third set consists of approximately equal to 7,100 genes, measured in 72 bone marrow and peripheral blood samples (Golub et al, 1999). We examine the use of scoring methods, measuring separation of tissue type (e.g., tumors from normals) using individual gene expression levels. These are then coupled with high-dimensional classification methods to assess the classification power of complete expression profiles. We present results of performing leave-one-out cross validation (LOOCV) experiments on the three data sets, employing nearest neighbor classifier, SVM (Cortes and Vapnik, 1995), AdaBoost (Freund and Schapire, 1997) and a novel clustering-based classification technique. As tumor samples can differ from normal samples in their cell-type composition, we also perform LOOCV experiments using appropriately modified sets of genes, attempting to eliminate the resulting bias. We demonstrate success rate of at least 90% in tumor versus normal classification, using sets of selected genes, with, as well as without, cellular-contamination-related members. These results are insensitive to the exact selection mechanism, over a certain range.     

Deriving quantitative conclusions from microarray expression data

MOTIVATION: The last few years have seen the development of DNA microarray technology that allows simultaneous measurement of the expression levels of thousands of genes. While many methods have been developed to analyze such data, most have been visualization-based. Methods that yield quantitative conclusions have been diverse and complex. RESULTS: We present two straightforward methods for identifying specific genes whose expression is linked with a phenotype or outcome variable as well as for systematically predicting sample class membership: (1) a conservative, permutation-based approach to identifying differentially expressed genes; (2) an augmentation of K-nearest-neighbor pattern classification. Our analyses replicate the quantitative conclusions of Golub et al. (1999; Science, 286, 531-537) on leukemia data, with better classification results, using far simpler methods. With the breast tumor data of Perou et al. (2000; Nature, 406, 747-752), the methods lend rigorous quantitative support to the conclusions of the original paper. In the case of the lymphoma data in Alizadeh et al. (2000; Nature, 403, 503-511), our analyses only partially support the conclusions of the original authors. AVAILABILITY: The software and supplementary information are available freely to researchers at academic and non-profit institutions at http://cc.ucsf.edu/jain/public     

Classification of multiple cancer types by multicategory support vector machines using gene expression data

MOTIVATION: High-density DNA microarray measures the activities of several thousand genes simultaneously and the gene expression profiles have been used for the cancer classification recently. This new approach promises to give better therapeutic measurements to cancer patients by diagnosing cancer types with improved accuracy. The Support Vector Machine (SVM) is one of the classification methods successfully applied to the cancer diagnosis problems. However, its optimal extension to more than two classes was not obvious, which might impose limitations in its application to multiple tumor types. We briefly introduce the Multicategory SVM, which is a recently proposed extension of the binary SVM, and apply it to multiclass cancer diagnosis problems. RESULTS: Its applicability is demonstrated on the leukemia data (Golub et al., 1999) and the small round blue cell tumors of childhood data (Khan et al., 2001). Comparable classification accuracy shown in the applications and its flexibility render the MSVM a viable alternative to other classification methods. SUPPLEMENTARY INFORMATION: http://www.stat.ohio-state.edu/~yklee/msvm.htm     

Microarray gene expression data with linked survival phenotypes: diffuse large-B-cell lymphoma revisited

Diffuse large-B-cell lymphoma (DLBCL) is an aggressive malignancy of mature B lymphocytes and is the most common type of lymphoma in adults. While treatment advances have been substantial in what was formerly a fatal disease, less than 50% of patients achieve lasting remission. In an effort to predict treatment success and explain disease heterogeneity clinical features have been employed for prognostic purposes, but have yielded only modest predictive performance. This has spawned a series of high-profile microarray-based gene expression studies of DLBCL, in the hope that molecular-level information could be used to refine prognosis. The intent of this paper is to reevaluate these microarray-based prognostic assessments, and extend the statistical methodology that has been used in this context. Methodological challenges arise in using patients' gene expression profiles to predict survival endpoints on account of the large number of genes and their complex interdependence. We initially focus on the Lymphochip data and analysis of Rosenwald et al. (2002). After describing relationships between the analyses performed and gene harvesting (Hastie et al., 2001a), we argue for the utility of penalized approaches, in particular least angle regression-least absolute shrinkage and selection operator (Efron et al., 2004). While these techniques have been extended to the proportional hazards/partial likelihood framework, the resultant algorithms are computationally burdensome. We develop residual-based approximations that eliminate this burden yet perform similarly. Comparisons of predictive accuracy across both methods and studies are effected using time-dependent receiver operating characteristic curves. These indicate that gene expression data, in turn, only delivers modest predictions of posttherapy DLBCL survival. We conclude by outlining possibilities for further work.     

Linear regression and two-class classification with gene expression data

MOTIVATION: Using gene expression data to classify (or predict) tumor types has received much research attention recently. Due to some special features of gene expression data, several new methods have been proposed, including the weighted voting scheme of Golub et al., the compound covariate method of Hedenfalk et al. (originally proposed by Tukey), and the shrunken centroids method of Tibshirani et al. These methods look different and are more or less ad hoc. RESULTS: We point out a close connection of the three methods with a linear regression model. Casting the classification problem in the general framework of linear regression naturally leads to new alternatives, such as partial least squares (PLS) methods and penalized PLS (PPLS) methods. Using two real data sets, we show the competitive performance of our new methods when compared with the other three methods.     

Detecting differentially expressed genes by relative entropy

DNA microarray experiments have generated large amount of gene expression measurements across different conditions. One crucial step in the analysis of these data is to detect differentially expressed genes. Some parametric methods, including the two-sample t-test (T-test) and variations of it, have been used. Alternatively, a class of non-parametric algorithms, such as the Wilcoxon rank sum test (WRST), significance analysis of microarrays (SAM) of Tusher et al. (2001), the empirical Bayesian (EB) method of Efron et al. (2001), etc., have been proposed. Most available popular methods are based on t-statistic. Due to the quality of the statistic that they used to describe the difference between groups of data, there are situations when these methods are inefficient, especially when the data follows multi-modal distributions. For example, some genes may display different expression patterns in the same cell type, say, tumor or normal, to form some subtypes. Most available methods are likely to miss these genes. We developed a new non-parametric method for selecting differentially expressed genes by relative entropy, called SDEGRE, to detect differentially expressed genes by combining relative entropy and kernel density estimation, which can detect all types of differences between two groups of samples. The significance of whether a gene is differentially expressed or not can be estimated by resampling-based permutations. We illustrate our method on two data sets from Golub et al. (1999) and Alon et al. (1999). Comparing the results with those of the T-test, the WRST and the SAM, we identified novel differentially expressed genes which are of biological significance through previous biological studies while they were not detected by the other three methods. The results also show that the genes selected by SDEGRE have a better capability to distinguish the two cell types.     

Molecular classification of multiple tumor types

Using gene expression data to classify tumor types is a very promising tool in cancer diagnosis. Previous works show several pairs of tumor types can be successfully distinguished by their gene expression patterns (Golub et al. 1999, Ben-Dor et al. 2000, Alizadeh et al. 2000). However, the simultaneous classification across a heterogeneous set of tumor types has not been well studied yet. We obtained 190 samples from 14 tumor classes and generated a combined expression dataset containing 16063 genes for each of those samples. We performed multi-class classification by combining the outputs of binary classifiers. Three binary classifiers (k-nearest neighbors, weighted voting, and support vector machines) were applied in conjunction with three combination scenarios (one-vs-all, all-pairs, hierarchical partitioning). We achieved the best cross validation error rate of 18.75% and the best test error rate of 21.74% by using the one-vs-all support vector machine algorithm. The results demonstrate the feasibility of performing clinically useful classification from samples of multiple tumor types.     

Pattern identification and classification in gene expression data using an autoassociative neural network model

The application of DNA microarray technology for analysis of gene expression creates enormous opportunities to accelerate the pace in understanding living systems and identification of target genes and pathways for drug development and therapeutic intervention. Parallel monitoring of the expression profiles of thousands of genes seems particularly promising for a deeper understanding of cancer biology and the identification of molecular signatures supporting the histological classification schemes of neoplastic specimens. However, the increasing volume of data generated by microarray experiments poses the challenge of developing equally efficient methods and analysis procedures to extract, interpret, and upgrade the information content of these databases. Herein, a computational procedure for pattern identification, feature extraction, and classification of gene expression data through the analysis of an autoassociative neural network model is described. The identified patterns and features contain critical information about gene-phenotype relationships observed during changes in cell physiology. They represent a rational and dimensionally reduced base for understanding the basic biology of the onset of diseases, defining targets of therapeutic intervention, and developing diagnostic tools for the identification and classification of pathological states. The proposed method has been tested on two different microarray datasets-Golub's analysis of acute human leukemia [Golub et al. (1999) Science 286:531-537], and the human colon adenocarcinoma study presented by Alon et al. [1999; Proc Natl Acad Sci USA 97:10101-10106]. The analysis of the neural network internal structure allows the identification of specific phenotype markers and the extraction of peculiar associations among genes and physiological states. At the same time, the neural network outputs provide assignment to multiple classes, such as different pathological conditions or tissue samples, for previously unseen instances.     

On predicting medulloblastoma metastasis by gene expression profiling

Accurately predicting clinical outcome or metastatic status from gene expression profiles remains one of the biggest hurdles facing the adoption of predictive medicine. Recently, MacDonald et al. (Nat. Genet. 2001, 29, 143-152) used gene expression profiles, from samples taken at diagnosis, to distinguish between clinically designated metastatic and nonmetastatic primary medulloblastomas, helping to elucidate the genetic mechanisms underlying metastasis and suggesting novel therapeutic targets. The obtained accuracy of predicting metastatic status does not, however, reach statistical significance on Fisher's exact test, although 22 training samples were used to make each prediction via leave-one-out testing. This paper introduces readily implemented nonlinear filters to transform sequences of gene expression levels into output signals that are significantly easier to classify and predict metastasis. It is shown that when only 3 exemplars each from the metastatic and nonmetastatic classes were assumed known, a predictor was constructed whose accuracy is statistically significant over the remaining profiles set aside as a test set. The predictor was as effective in recognizing metastatic as nonmetastatic medulloblastomas, and may be helpful in deciding which patients require more aggressive therapy. The same predictor was similarly effective on an independent set of 5 nonmetastatic tumors and 3 metastatic cell lines also used by MacDonald et al.     

Google goes cancer: improving outcome prediction for cancer patients by network-based ranking of marker genes

Predicting the clinical outcome of cancer patients based on the expression of marker genes in their tumors has received increasing interest in the past decade. Accurate predictors of outcome and response to therapy could be used to personalize and thereby improve therapy. However, state of the art methods used so far often found marker genes with limited prediction accuracy, limited reproducibility, and unclear biological relevance. To address this problem, we developed a novel computational approach to identify genes prognostic for outcome that couples gene expression measurements from primary tumor samples with a network of known relationships between the genes. Our approach ranks genes according to their prognostic relevance using both expression and network information in a manner similar to Google's PageRank. We applied this method to gene expression profiles which we obtained from 30 patients with pancreatic cancer, and identified seven candidate marker genes prognostic for outcome. Compared to genes found with state of the art methods, such as Pearson correlation of gene expression with survival time, we improve the prediction accuracy by up to 7%. Accuracies were assessed using support vector machine classifiers and Monte Carlo cross-validation. We then validated the prognostic value of our seven candidate markers using immunohistochemistry on an independent set of 412 pancreatic cancer samples. Notably, signatures derived from our candidate markers were independently predictive of outcome and superior to established clinical prognostic factors such as grade, tumor size, and nodal status. As the amount of genomic data of individual tumors grows rapidly, our algorithm meets the need for powerful computational approaches that are key to exploit these data for personalized cancer therapies in clinical practice.     

Prediction of biologically significant components from microarray data: Independently Consistent Expression Discriminator (ICED)

MOTIVATION: Class distinction is a supervised learning approach that has been successfully employed in the analysis of high-throughput gene expression data. Identification of a set of genes that predicts differential biological states allows for the development of basic and clinical scientific approaches to the diagnosis of disease. The Independent Consistent Expression Discriminator (ICED) was designed to provide a more biologically relevant search criterion during predictor selection by embracing the inherent variability of gene expression in any biological state. The four components of ICED include (i) normalization of raw data; (ii) assignment of weights to genes from both classes; (iii) counting of votes to determine optimal number of predictor genes for class distinction; (iv) calculation of prediction strengths for classification results. The search criteria employed by ICED is designed to identify not only genes that are consistently expressed at one level in one class and at a consistently different level in another class but identify genes that are variable in one class and consistent in another. The result is a novel approach to accurately select biologically relevant predictors of differential disease states from a small number of microarray samples. RESULTS: The data described herein utilized ICED to analyze the large AML/ALL training and test data set (Golub et al., 1999, Science, 286, 531-537) in addition to a smaller data set consisting of an animal model of the childhood neurodegenerative disorder, Batten disease, generated for this study. Both of the analyses presented herein have correctly predicted biologically relevant perturbations that can be used for disease classification, irrespective of sample size. Furthermore, the results have provided candidate proteins for future study in understanding the disease process and the identification of potential targets for therapeutic intervention.     

A comparative review of statistical methods for discovering differentially expressed genes in replicated microarray experiments

MOTIVATION: A common task in analyzing microarray data is to determine which genes are differentially expressed across two kinds of tissue samples or samples obtained under two experimental conditions. Recently several statistical methods have been proposed to accomplish this goal when there are replicated samples under each condition. However, it may not be clear how these methods compare with each other. Our main goal here is to compare three methods, the t-test, a regression modeling approach (Thomas et al., Genome Res., 11, 1227-1236, 2001) and a mixture model approach (Pan et al., http://www.biostat.umn.edu/cgi-bin/rrs?print+2001,2001a,b) with particular attention to their different modeling assumptions. RESULTS: It is pointed out that all the three methods are based on using the two-sample t-statistic or its minor variation, but they differ in how to associate a statistical significance level to the corresponding statistic, leading to possibly large difference in the resulting significance levels and the numbers of genes detected. In particular, we give an explicit formula for the test statistic used in the regression approach. Using the leukemia data of Golub et al. (Science, 285, 531-537, 1999), we illustrate these points. We also briefly compare the results with those of several other methods, including the empirical Bayesian method of Efron et al. (J. Am. Stat. Assoc., to appear, 2001) and the Significance Analysis of Microarray (SAM) method of Tusher et al. (PROC: Natl Acad. Sci. USA, 98, 5116-5121, 2001).     

Monte Carlo feature selection for supervised classification

MOTIVATION: Pre-selection of informative features for supervised classification is a crucial, albeit delicate, task. It is desirable that feature selection provides the features that contribute most to the classification task per se and which should therefore be used by any classifier later used to produce classification rules. In this article, a conceptually simple but computer-intensive approach to this task is proposed. The reliability of the approach rests on multiple construction of a tree classifier for many training sets randomly chosen from the original sample set, where samples in each training set consist of only a fraction of all of the observed features. RESULTS: The resulting ranking of features may then be used to advantage for classification via a classifier of any type. The approach was validated using Golub et al. leukemia data and the Alizadeh et al. lymphoma data. Not surprisingly, we obtained a significantly different list of genes. Biological interpretation of the genes selected by our method showed that several of them are involved in precursors to different types of leukemia and lymphoma rather than being genes that are common to several forms of cancers, which is the case for the other methods. AVAILABILITY: Prototype available upon request.     

Comparing solid tumors with cell lines: implications for identifying drug resistance genes in cancer

Gene expression arrays allow researchers to profile the differences between cell lines or tissues and they may identify genetic markers of development, organ maturation, or tumor progression. Although a primary tumor that grows in a host and a tumor-cell-line derived from that primary tumor and grown in vitro share similar gene expression profiles, there are, not unexpectedly, some important differences. In fact, Stein and colleagues have found that genes that are differentially expressed in primary tumors as compared to the specific genes expressed in their cell-line derivatives are more reliably predictive of tumor tractability. Thus, sensitivity in vitro might not reflect sensitivity in vivo. Because anti-tumor compounds are largely evaluated in cell culture assays, these compounds' therapeutic utility must be judged in light of genes described by Stein et al. that better predict tractability.     

Developing optimal prediction models for cancer classification using gene expression data

Microarrays can provide genome-wide expression patterns for various cancers, especially for tumor sub-types that may exhibit substantially different patient prognosis. Using such gene expression data, several approaches have been proposed to classify tumor sub-types accurately. These classification methods are not robust, and often dependent on a particular training sample for modelling, which raises issues in utilizing these methods to administer proper treatment for a future patient. We propose to construct an optimal, robust prediction model for classifying cancer sub-types using gene expression data. Our model is constructed in a step-wise fashion implementing cross-validated quadratic discriminant analysis. At each step, all identified models are validated by an independent sample of patients to develop a robust model for future data. We apply the proposed methods to two microarray data sets of cancer: the acute leukemia data by Golub et al. and the colon cancer data by Alon et al. We have found that the dimensionality of our optimal prediction models is relatively small for these cases and that our prediction models with one or two gene factors outperforms or has competing performance, especially for independent samples, to other methods based on 50 or more predictive gene factors. The methodology is implemented and developed by the procedures in R and Splus. The source code can be obtained at http://hesweb1.med.virginia.edu/bioinformatics.     

Survival trees for analyzing clinical outcome in lung adenocarcinomas based on gene expression profiles: identification of neogenin and diacylglycerol kinase alpha expression as critical factors.

We present survival trees as an exploratory tool for revealing new insights into gene expression profiles in combination with clinical patient data. Survival trees partition the patient data studied into groups with similar survival outcomes and identify characteristic genetic profiles within these groups. We demonstrate the application of survival trees in a study involving the expression profiles of 3,588 genes in 211 lung adenocarcinoma patients. The survival tree identified a group of early-stage cancer patients with relatively low survival rates and another group of advanced-stage patients with remarkably good survival outcome. For both groups, the tree identified characteristic expression profiles of genes that might play a role in cancerogenesis and disease progression, notably the genes for the netrin receptor neogenin and the Ras/Rho kinase modulator diacylglycerol kinase alpha.     

Noise does not equal bias in assessing the evolutionary history of the angiosperm flora of China: A response to Qian (2019)

In response to our paper on the evolutionary history of the Chinese flora, Qian suggests that certain features of the divergence time estimation employed might have led to biased conclusions in Lu et al (2018). Here, we consider Qian's specific criticisms, explore the extent of uncertainty in the data and demonstrate that (i) no systematic bias toward dates that are too young or too old is detected in Lu et al.; (ii) constraint of the crown age of angiosperms does not bias the generic ages estimated by Lu et al.; and (iii) ages derived from the Chinese regional phylogeny do not bias the conclusions reported by Lu et al. All these analyses confirm that the conclusions reported previously are robust. We argue that, like many large-scale biodiversity analyses, sources of noise in divergence time estimation are to be expected, but these should not be confused with bias.     

Directed indices for exploring gene expression data

MOTIVATION: Large expression studies with clinical outcome data are becoming available for analysis. An important goal is to identify genes or clusters of genes where expression is related to patient outcome. While clustering methods are useful data exploration tools, they do not directly allow one to relate the expression data to clinical outcome. Alternatively, methods which rank genes based on their univariate significance do not incorporate gene function or relationships to genes that have been previously identified. In addition, after sifting through potentially thousands of genes, summary estimates (e.g. regression coefficients or error rates) algorithms should address the potentially large bias introduced by gene selection. RESULTS: We developed a gene index technique that generalizes methods that rank genes by their univariate associations to patient outcome. Genes are ordered based on simultaneously linking their expression both to patient outcome and to a specific gene of interest. The technique can also be used to suggest profiles of gene expression related to patient outcome. A cross-validation method is shown to be important for reducing bias due to adaptive gene selection. The methods are illustrated on a recently collected gene expression data set based on 160 patients with diffuse large cell lymphoma (DLCL).