Comparative evaluation of set-level techniques in predictive classification of gene expression samples

Abstract

BackgroundAnalysis of gene expression data in terms of a priori-defined gene sets has recently received significant attention as this approach typically yields more compact and interpretable results than those produced by traditional methods that rely on individual genes. The set-level strategy can also be adopted with similar benefits in predictive classification tasks accomplished with machine learning algorithms. Initial studies into the predictive performance of set-level classifiers have yielded rather controversial results. The goal of this study is to provide a more conclusive evaluation by testing various components of the set-level framework within a large collection of machine learning experiments.ResultsGenuine curated gene sets constitute better features for classification than sets assembled without biological relevance. For identifying the best gene sets for classification, the Global test outperforms the gene-set methods GSEA and SAM-GS as well as two generic feature selection methods. To aggregate expressions of genes into a feature value, the singular value decomposition (SVD) method as well as the SetSig technique improve on simple arithmetic averaging. Set-level classifiers learned with 10 features constituted by the Global test slightly outperform baseline gene-level classifiers learned with all original data features although they are slightly less accurate than gene-level classifiers learned with a prior feature-selection step.ConclusionSet-level classifiers do not boost predictive accuracy, however, they do achieve competitive accuracy if learned with the right combination of ingredients.AvailabilityOpen-source, publicly available software was used for classifier learning and testing. The gene expression datasets and the gene set database used are also publicly available. The full tabulation of experimental results is available at http://ida.felk.cvut.cz/CESLT.

Network-constrained forest for regularized classification of omics data

“…The probability of its occurrence increases with the decreasing path length between the pair of features in the network and increasing interaction observed in measurements. Unlike our previous efforts, we do not rely on feature extraction based on prior modules such as pathways or simple interaction subgraphs [60,61,9].…”

Ensemble Learning of Run-Time Prediction Models for Data-Intensive Scientific Workflows

“…The goal of the experiment is to compare a novel classification algorithm (GELF) and the state-of-the-art approach (baseline) [7] on their respective ability to classify unseen data.…”

“…The third task consists on consists on learning and evaluating the performance of the (GELF) classifier. GELF is a feature construction algorithm based on iterative improvement of the best solution obtained by the state-of-the-art approach [7]. Each workflow comprises 20 sub-experiments: both combinations of the GELF task with the rankers (Random and SVM-RFE) applied on the 10 dataset folds.…”

Ensemble learning of runtime prediction models for gene-expression analysis workflows

“…These ranker tasks precede the execution of the third type of tasks that consist on learning and evaluating the performance of the (GELF) classifier. GELF is a feature construction algorithm based on iterative improvement of the best solution obtained by the state-of-the-art approach [8].…”

“…Each fold involves the execution of a Random ranker task, an SVM-RFE ranker tasks and two GELF tasks giving rise to 2 sub-experiments: a «Random, GELF», and b «SVM-RFE, GELF». For each dataset 10 folds are generated algorithm (GELF)[8] on their respective ability to classify unseen gene expression samples 1 .…”