Robert Tibshirani

Significance analysis of microarrays applied to the ionizing radiation response

Microarrays can measure the expression of thousands of genes to identify changes in expression between different biological states. Methods are needed to determine the significance of these changes while accounting for the enormous number of genes. We describe a method, Significance Analysis of Microarrays (SAM), that assigns a score to each gene on the basis of change in gene expression relative to the standard deviation of repeated measurements. For genes with scores greater than an adjustable threshold, SAM uses permutations of the repeated measurements to estimate the percentage of genes identified by chance, the false discovery rate (FDR). When the transcriptional response of human cells to ionizing radiation was measured by microarrays, SAM identified 34 genes that changed at least 1.5-fold with an estimated FDR of 12%, compared with FDRs of 60 and 84% by using conventional methods of analysis. Of the 34 genes, 19 were involved in cell cycle regulation and 3 in apoptosis. Surprisingly, four nucleotide excision repair genes were induced, suggesting that this repair pathway for UV-damaged DNA might play a previously unrecognized role in repairing DNA damaged by ionizing radiation.

Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications

The purpose of this study was to classify breast carcinomas based on variations in gene expression patterns derived from cDNA microarrays and to correlate tumor characteristics to clinical outcome. A total of 85 cDNA microarray experiments representing 78 cancers, three fibroadenomas, and four normal breast tissues were analyzed by hierarchical clustering. As reported previously, the cancers could be classified into a basal epithelial-like group, an ERBB2-overexpressing group and a normal breast-like group based on variations in gene expression. A novel finding was that the previously characterized luminal epithelial/estrogen receptor-positive group could be divided into at least two subgroups, each with a distinctive expression profile. These subtypes proved to be reasonably robust by clustering using two different gene sets: first, a set of 456 cDNA clones previously selected to reflect intrinsic properties of the tumors and, second, a gene set that highly correlated with patient outcome. Survival analyses on a subcohort of patients with locally advanced breast cancer uniformly treated in a prospective study showed significantly different outcomes for the patients belonging to the various groups, including a poor prognosis for the basal-like subtype and a significant difference in outcome for the two estrogen receptor-positive groups.

Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling.

Diffuse large B-cell lymphoma (DLBCL), the most common subtype of non-Hodgkins lymphoma, is clinically heterogeneous: 40% of patients respond well to current therapy and have prolonged survival, whereas the remainder succumb to the disease. We proposed that this variability in natural history reflects unrecognized molecular heterogeneity in the tumours. Using DNA microarrays, we have conducted a systematic characterization of gene expression in B-cell malignancies. Here we show that there is diversity in gene expression among the tumours of DLBCL patients, apparently reflecting the variation in tumour proliferation rate, host response and differentiation state of the tumour. We identified two molecularly distinct forms of DLBCL which had gene expression patterns indicative of different stages of B-cell differentiation. One type expressed genes characteristic of germinal centre B cells (‘germinal centre B-like DLBCL’); the second type expressed genes normally induced during in vitro activation of peripheral blood B cells (‘activated B-like DLBCL’). Patients with germinal centre B-like DLBCL had a significantly better overall survival than those with activated B-like DLBCL. The molecular classification of tumours on the basis of gene expression can thus identify previously undetected and clinically significant subtypes of cancer.

Statistical significance for genomewide studies

With the increase in genomewide experiments and the sequencing of multiple genomes, the analysis of large data sets has become commonplace in biology. It is often the case that thousands of features in a genomewide data set are tested against some null hypothesis, where a number of features are expected to be significant. Here we propose an approach to measuring statistical significance in these genomewide studies based on the concept of the false discovery rate. This approach offers a sensible balance between the number of true and false positives that is automatically calibrated and easily interpreted. In doing so, a measure of statistical significance called the q value is associated with each tested feature. The q value is similar to the well known p value, except it is a measure of significance in terms of the false discovery rate rather than the false positive rate. Our approach avoids a flood of false positive results, while offering a more liberal criterion than what has been used in genome scans for linkage.

Least angle regression

DISCUSSION OF “LEAST ANGLE REGRESSION” BY EFRONET AL.By Jean-Michel Loubes and Pascal MassartUniversit´e Paris-SudThe issue of model selection has drawn the attention of both applied andtheoretical statisticians for a long time. Indeed, there has been an enor-mous range of contribution in model selection proposals, including work byAkaike (1973), Mallows (1973), Foster and George (1994), Birg´e and Mas-sart (2001a) and Abramovich, Benjamini, Donoho and Johnstone (2000).Over the last decade, modern computer-driven methods have been devel-oped such as All Subsets, Forward Selection, Forward Stagewise or Lasso.Such methods are useful in the setting of the standard linear model, wherewe observe noisy data and wish to predict the response variable using onlya few covariates, since they provide automatically linear models that fit thedata. The procedure described in this paper is, on the one hand, numeri-cally very efficient and, on the other hand, very general, since, with slightmodifications, it enables us to recover the estimates given by the Lasso andStagewise.1. Estimation procedure. The “LARS” method is based on a recursiveprocedure selecting, at each step, the covariates having largest absolute cor-relation with the response y. In the case of an orthogonal design, the esti-mates can then be viewed as an lDISCUSSION OF “LEAST ANGLE REGRESSION” BY EFRONET AL.By Berwin A. TurlachUniversity of Western AustraliaI would like to begin by congratulating the authors (referred to belowas EHJT) for their interesting paper in which they propose a new variableselection method (LARS) for building linear models and show how their newmethod relates to other methods that have been proposed recently. I foundthe paper to be very stimulating and found the additional insight that itprovides about the Lasso technique to be of particular interest.My comments center around the question of how we can select linearmodels that conform with the marginality principle [Nelder (1977, 1994)and McCullagh and Nelder (1989)]; that is, the response surface is invariantunder scaling and translation of the explanatory variables in the model.Recently one of my interests was to explore whether the Lasso techniqueor the nonnegative garrote [Breiman (1995)] could be modified such that itincorporates the marginality principle. However, it does not seem to be atrivial matter to change the criteria that these techniques minimize in such away that the marginality principle is incorporated in a satisfactory manner.On the other hand, it seems to be straightforward to modify the LARStechnique to incorporate this principle. In their paper, EHJT address thisissue somewhat in passing when they suggest toward the end of Section 3that one first fit main effects only and interactions in a second step to controlthe order in which variables are allowed to enter the model. However, sucha two-step procedure may have a somewhat less than optimal behavior asthe following, admittedly artificial, example shows.Assume we have a vector of explanatory variables X =(XThe purpose of model selection algorithms such as All Subsets, Forward Selection and Backward Elimination is to choose a linear model on the basis of the same set of data to which the model will be applied. Typically we have available a large collection of possible covariates from which we hope to select a parsimonious set for the efficient prediction of a response variable. Least Angle Regression (LARS), a new model selection algorithm, is a useful and less greedy version of traditional forward selection methods. Three main properties are derived: (1) A simple modification of the LARS algorithm implements the Lasso, an attractive version of ordinary least squares that constrains the sum of the absolute regression coefficients; the LARS modification calculates all possible Lasso estimates for a given problem, using an order of magnitude less computer time than previous methods. (2) A different LARS modification efficiently implements Forward Stagewise linear regression, another promising new model selection method; this connection explains the similar numerical results previously observed for the Lasso and Stagewise, and helps us understand the properties of both methods, which are seen as constrained versions of the simpler LARS algorithm. (3) A simple approximation for the degrees of freedom of a LARS estimate is available, from which we derive a Cp estimate of prediction error; this allows a principled choice among the range of possible LARS estimates. LARS and its variants are computationally efficient: the paper describes a publicly available algorithm that requires only the same order of magnitude of computational effort as ordinary least squares applied to the full set of covariates.

Repeated observation of breast tumor subtypes in independent gene expression data sets

Characteristic patterns of gene expression measured by DNA microarrays have been used to classify tumors into clinically relevant subgroups. In this study, we have refined the previously defined subtypes of breast tumors that could be distinguished by their distinct patterns of gene expression. A total of 115 malignant breast tumors were analyzed by hierarchical clustering based on patterns of expression of 534 “intrinsic” genes and shown to subdivide into one basal-like, one ERBB2-overexpressing, two luminal-like, and one normal breast tissue-like subgroup. The genes used for classification were selected based on their similar expression levels between pairs of consecutive samples taken from the same tumor separated by 15 weeks of neoadjuvant treatment. Similar cluster analyses of two published, independent data sets representing different patient cohorts from different laboratories, uncovered some of the same breast cancer subtypes. In the one data set that included information on time to development of distant metastasis, subtypes were associated with significant differences in this clinical feature. By including a group of tumors from BRCA1 carriers in the analysis, we found that this genotype predisposes to the basal tumor subtype. Our results strongly support the idea that many of these breast tumor subtypes represent biologically distinct disease entities.

Diagnosis of multiple cancer types by shrunken centroids of gene expression

We have devised an approach to cancer class prediction from gene expression profiling, based on an enhancement of the simple nearest prototype (centroid) classifier. We shrink the prototypes and hence obtain a classifier that is often more accurate than competing methods. Our method of “nearest shrunken centroids” identifies subsets of genes that best characterize each class. The technique is general and can be used in many other classification problems. To demonstrate its effectiveness, we show that the method was highly efficient in finding genes for classifying small round blue cell tumors and leukemias.

Missing value estimation methods for DNA microarrays

MOTIVATION Gene expression microarray experiments can generate data sets with multiple missing expression values. Unfortunately, many algorithms for gene expression analysis require a complete matrix of gene array values as input. For example, methods such as hierarchical clustering and K-means clustering are not robust to missing data, and may lose effectiveness even with a few missing values. Methods for imputing missing data are needed, therefore, to minimize the effect of incomplete data sets on analyses, and to increase the range of data sets to which these algorithms can be applied. In this report, we investigate automated methods for estimating missing data. RESULTS We present a comparative study of several methods for the estimation of missing values in gene microarray data. We implemented and evaluated three methods: a Singular Value Decomposition (SVD) based method (SVDimpute), weighted K-nearest neighbors (KNNimpute), and row average. We evaluated the methods using a variety of parameter settings and over different real data sets, and assessed the robustness of the imputation methods to the amount of missing data over the range of 1--20% missing values. We show that KNNimpute appears to provide a more robust and sensitive method for missing value estimation than SVDimpute, and both SVDimpute and KNNimpute surpass the commonly used row average method (as well as filling missing values with zeros). We report results of the comparative experiments and provide recommendations and tools for accurate estimation of missing microarray data under a variety of conditions.

Estimating the number of clusters in a data set via the gap statistic

We propose a method (the ‘gap statistic’) for estimating the number of clusters (groups) in a set of data. The technique uses the output of any clustering algorithm (e.g. K-means or hierarchical), comparing the change in within-cluster dispersion with that expected under an appropriate reference null distribution. Some theory is developed for the proposal and a simulation study shows that the gap statistic usually outperforms other methods that have been proposed in the literature.

Sparse Principal Component Analysis

Principal component analysis (PCA) is widely used in data processing and dimensionality reduction. However, PCA suffers from the fact that each principal component is a linear combination of all the original variables, thus it is often difficult to interpret the results. We introduce a new method called sparse principal component analysis (SPCA) using the lasso (elastic net) to produce modified principal components with sparse loadings. We first show that PCA can be formulated as a regression-type optimization problem; sparse loadings are then obtained by imposing the lasso (elastic net) constraint on the regression coefficients. Efficient algorithms are proposed to fit our SPCA models for both regular multivariate data and gene expression arrays. We also give a new formula to compute the total variance of modified principal components. As illustrations, SPCA is applied to real and simulated data with encouraging results.