Prediction potential of candidate biomarker sets identified and validated on gene expression data from multiple datasets

Abstract

BackgroundIndependently derived expression profiles of the same biological condition often have few genes in common. In this study, we created populations of expression profiles from publicly available microarray datasets of cancer (breast, lymphoma and renal) samples linked to clinical information with an iterative machine learning algorithm. ROC curves were used to assess the prediction error of each profile for classification. We compared the prediction error of profiles correlated with molecular phenotype against profiles correlated with relapse-free status. Prediction error of profiles identified with supervised univariate feature selection algorithms were compared to profiles selected randomly from a) all genes on the microarray platform and b) a list of known disease-related genes (a priori selection). We also determined the relevance of expression profiles on test arrays from independent datasets, measured on either the same or different microarray platforms.ResultsHighly discriminative expression profiles were produced on both simulated gene expression data and expression data from breast cancer and lymphoma datasets on the basis of ER and BCL-6 expression, respectively. Use of relapse-free status to identify profiles for prognosis prediction resulted in poorly discriminative decision rules. Supervised feature selection resulted in more accurate classifications than random or a priori selection, however, the difference in prediction error decreased as the number of features increased. These results held when decision rules were applied across-datasets to samples profiled on the same microarray platform.ConclusionOur results show that many gene sets predict molecular phenotypes accurately. Given this, expression profiles identified using different training datasets should be expected to show little agreement. In addition, we demonstrate the difficulty in predicting relapse directly from microarray data using supervised machine learning approaches. These findings are relevant to the use of molecular profiling for the identification of candidate biomarker panels.

Meta-analysis of gene expression profiles related to relapse-free survival in 1,079 breast cancer patients

“…For example, a recent study on microarray data related to breast cancer, renal tumors and lymphoma and including clinical information compared the prediction errors using different training sets. The results suggested that expression profiles established in this way showed little overlap [21]. We achieved significant prediction success using different gene sets established on different microarray platforms, and therefore we provide additional support to this finding.…”

Introducing Serine as Cardiovascular Disease Biomarker Candidate via Pathway Analysis

“…Usage of limited numbers of sensitive and specific biomarkers related to a certain disease in diagnosis (such as glucose in diabetes) is established from many years ago. Advances in biomarker discovery methods provided new opportunities to introduce efficient biomarkers or set of biomarkers (a panel) associated with the studied diseases [21]. Proteomics, genomics, metabolomics and other high throughput methods are used widely to identify new biomarkers.…”

Pattern recognition methods in microarray based oncology study

“…Usually microarray data class prediction problems are similar to class prediction problems in other areas, so most of the classic class prediction methods have been applied to microarray data analysis -LDA projects the samples into one dimension space to maximize the distance between classes and minimize the distance in each class at the same time [46]; k-NN predicts the test sample based on the class labels of the k samples near the test sample [37,45,46,58,60]; decision tree is a classifier in the form of a tree structure and each node either indicates the label of the test sample or specifies some test to select which sub-tree to go [67]; SVM aims to find an optimal hyper-plane to separate the two classes and maximize the distance between the hyper-plane and the closest data pointing to the hyperplane [24,47,[50][51][52]56,57]; and artificial neural networks take genes as input nodes and class label as the output node to learn the parameters connecting to nodes of different layers and predict the unknown samples [68][69][70]. Some modifications to the classic methods are also applied to this problem, such as Linder et al who proposed the subsequent artificial neural network (ANN) method, which has two levels of ANN.…”

“…where μ ki is mean value of gene i in class k, σ ki is standard deviation of gene i in class k [1,45,46], ratio between class sum of squares to within class sum of squares [46], correlation between gene expression G i and class label Y [42,47], entropy-based method [48] are applied in feature selection process. Because most of the filter methods only consider one gene at a time, they will fail to identify the combination effect of several genes.…”