A comparative study of discriminating human heart failure etiology using gene expression profiles

Abstract

BackgroundHuman heart failure is a complex disease that manifests from multiple genetic and environmental factors. Although ischemic and non-ischemic heart disease present clinically with many similar decreases in ventricular function, emerging work suggests that they are distinct diseases with different responses to therapy. The ability to distinguish between ischemic and non-ischemic heart failure may be essential to guide appropriate therapy and determine prognosis for successful treatment. In this paper we consider discriminating the etiologies of heart failure using gene expression libraries from two separate institutions.ResultsWe apply five new statistical methods, including partial least squares, penalized partial least squares, LASSO, nearest shrunken centroids and random forest, to two real datasets and compare their performance for multiclass classification. It is found that the five statistical methods perform similarly on each of the two datasets: it is difficult to correctly distinguish the etiologies of heart failure in one dataset whereas it is easy for the other one. In a simulation study, it is confirmed that the five methods tend to have close performance, though the random forest seems to have a slight edge.ConclusionsFor some gene expression data, several recently developed discriminant methods may perform similarly. More importantly, one must remain cautious when assessing the discriminating performance using gene expression profiles based on a small dataset; our analysis suggests the importance of utilizing multiple or larger datasets.

Meta‐learning approach to gene expression data classification

“…One of the earliest applications of ML for classification of microarray data were the successful design of a classifier system to distinguish patients of related types of leukemia (Golub et al , 1999). Following this approach, various studies confirmed the applicability of ML algorithms to data from gene expression domains (Dudoit et al , 2002; Romualdi et al , 2003; Man et al , 2004; Lee et al , 2005; Statnikov et al 2005; Huang et al , 2005). An analysis of their results indicates that no single algorithm performs much better than the others in all cases.…”

“…Statnikov et al (2005) compared three multi‐class classification methods, named multi‐class SVMs, KNN, and neural networks, on 11 datasets and concluded that SVMs outperformed their competitors. Finally, Huang et al (2005) compared the performance of five statistical methods on two datasets and concluded that the algorithms obtain similar results. As a whole, the aforementioned studies suggest that there is no obvious winning algorithm.…”

Evaluating Microarray-based Classifiers: An Overview

“…This summary should be considered with caution, since not detailing the used variants of the considered methods. Dudoit et al (2002) 3 data sets MCCV 2:1

Included: LDA, DLDA, DQDA, Fisher, k NN, trees, tree-based ensembles

Variable selection: F-statistic

Conclusion : DLDA and k NN perform best

Romualdi et al (2003) 2 data sets CV

Included: DLDA, trees, neural networks SVM, k NN, PAM combined with:

Variable selection/dimension reduction: PLS, PCA, soft thresholding, GA/ k NN

Conclusion : PLS transformation is recommendable, No classifier uniformly better than the other

Man et al (2004) 6 data sets LOOCV, bootstrap

Included: k NN, PCA+LDA, PLS-DA, neural networks, random forests, SVM

Variable selection: F-statistic

Conclusion : PLS-DA and SVM perform best

Lee et al (2005) 7 data sets LOOCV, MCCV 2:1

Included: 21 methods (e.g. tree ensembles, SVM, LDA, DLDA, QDA, Fisher, PAM)

Variable selection: F-statistic, rank-based score, soft thresholding Conclusion : No classifier uniformly better than the other, rank-based variable selection performs best Statnikov et al (2005) 11 data sets LOOCV, 10-fold CVIncluded: SVM, k NN, probabilistic neural networks, backpropagation neural networksVariable selection: BSS/WSS, Golub et al (1999), Kruskal-Wallis test Conclusion : SVM performs best Huang et al (2005) 2 data sets LOOCVIncluded: PLS, penalized PLS, LASSO, PAM, random forestsVariable selection: F-statisticRandom forests perform slightly better Conclusion : No classifier uniformly better than the other…”

An introduction to recursive partitioning: Rationale, application, and characteristics of classification and regression trees, bagging, and random forests.

“…In the past years, recursive partitioning methods have gained popularity as a means of multivariate data exploration in various scientific fields, including, for example, the analysis of microarray data, DNA sequencing, and many other applications in genetics, epidemiology, and medicine (cf., e.g., Bureau et al, 2005; Diaz-Uriarte & Alvarez de Andrés, 2006; Gunther, Stone, Gerwien, Bento, & Heyes, 2003; Huang et al, 2005; Lunetta, Hayward, Segal, & Eerdewegh, 2004; Qi, Bar-Joseph, & Klein-Seetharaman, 2006; Segal, Barbour, & Grant, 2004; Shih, Seligson, Belldegrun, Palotie, & Horvath, 2005; Ward, Pajevic, Dreyfuss, & Malley, 2006).…”