A balanced iterative random forest for gene selection from microarray data

Abstract

BackgroundThe wealth of gene expression values being generated by high throughput microarray technologies leads to complex high dimensional datasets. Moreover, many cohorts have the problem of imbalanced classes where the number of patients belonging to each class is not the same. With this kind of dataset, biologists need to identify a small number of informative genes that can be used as biomarkers for a disease.ResultsThis paper introduces a Balanced Iterative Random Forest (BIRF) algorithm to select the most relevant genes for a disease from imbalanced high-throughput gene expression microarray data. Balanced iterative random forest is applied on four cancer microarray datasets: a childhood leukaemia dataset, which represents the main target of this paper, collected from The Children’s Hospital at Westmead, NCI 60, a Colon dataset and a Lung cancer dataset. The results obtained by BIRF are compared to those of Support Vector Machine-Recursive Feature Elimination (SVM-RFE), Multi-class SVM-RFE (MSVM-RFE), Random Forest (RF) and Naive Bayes (NB) classifiers. The results of the BIRF approach outperform these state-of-the-art methods, especially in the case of imbalanced datasets. Experiments on the childhood leukaemia dataset show that a 7% ∼ 12% better accuracy is achieved by BIRF over MSVM-RFE with the ability to predict patients in the minor class. The informative biomarkers selected by the BIRF algorithm were validated by repeating training experiments three times to see whether they are globally informative, or just selected by chance. The results show that 64% of the top genes consistently appear in the three lists, and the top 20 genes remain near the top in the other three lists.ConclusionThe designed BIRF algorithm is an appropriate choice to select genes from imbalanced high-throughput gene expression microarray data. BIRF outperforms the state-of-the-art methods, especially the ability to handle the class-imbalanced data. Moreover, the analysis of the selected genes also provides a way to distinguish between the predictive genes and those that only appear to be predictive.

Evaluating the impact of single nucleotide variants on transcription factor binding

“…Since there were more non-ASB events than ASB events, non-ASB events were randomly downsampled to balance the training data set for each tree building process following the balanced random forest approach (34,35). We used a 5-fold cross-validation approach to assess the predictive power of the classifiers.…”

Feature selection and classification of urinary mRNA microarray data by iterative random forest to diagnose renal fibrosis: a two-stage study

“…Data from a previous study have shown that the OOB estimate is as accurate as using a test set of the same sample size as the training set11. Although RF was more resistant to over-fitting than a support vector machine or artificial neural network, the validation set was involved in the same process simultaneously and an “early-stop” strategy was applied to prevent “over-fitting”12. The final set of mRNAs with smallest estimated OOB was identified as important features of TIF.…”

“…Random forest (RF) methods, constructed from decision tree predictors, represent one of the most prevalent supervised machine learning methods, which was first introduced by Breiman in 200111. RF methods return measures of variable importance and have superior performance with respect to the problems that microarray data bring, making it well suited for microarray analysis12. An empirical study by Archer et al .…”

Adopting Multivariate Nonparametric Tools to Determine Genotype-Phenotype Interactions in Health and Disease

“…In this process of revisiting some of the key aspects of RF, variable importance measures were also reevaluated focusing in their performance on extreme cases of recognized limited performance of RF, such in highly unbalanced datasets [61]. With this aim, a more robust variable importance indicator was proposed, now using a variant at the variable permutation step based in the area under the curve (AUC) [30].…”