Semisupervised Learning for Molecular Profiling

Abstract

Class prediction and feature selection are two learning tasks that are strictly paired in the search of molecular profiles from microarray data. Researchers have become aware how easy it is to incur a selection bias effect, and complex validation setups are required to avoid overly optimistic estimates of the predictive accuracy of the models and incorrect gene selections. This paper describes a semisupervised pattern discovery approach that uses the by-products of complete validation studies on experimental setups for gene profiling. In particular, we introduce the study of the patterns of single sample responses (sample-tracking profiles) to the gene selection process induced by typical supervised learning tasks in microarray studies. We originate sample-tracking profiles as the aggregated off-training evaluation of SVM models of increasing gene panel sizes. Genes are ranked by E-RFE, an entropy-based variant of the recursive feature elimination for support vector machines (RFE-SVM). A Dynamic Time Warping (DTW) algorithm is then applied to define a metric between sample-tracking profiles. An unsupervised clustering based on the DTW metric allows automating the discovery of outliers and of subtypes of different molecular profiles. Applications are described on synthetic data and in two gene expression studies.

Gene expression profiling identifies potential relevant genes in alveolar rhabdomyosarcoma pathogenesis and discriminates PAX3‐FKHR positive and negative tumors

“…This was confirmed with the blind analysis of 4 additional ARMS biopsies that were correctly assigned to either subgroups and remarkably with a standard complete validation procedure implemented to control selection bias in predictive classification expression studies. 22,23 It is difficult to establish whether among the deregulated genes that characterize the translocation-positive ARMS samples, there are some specific transcripts that could be directly related to this primary genetic event. Further functional studies are needed to investigate this issue and some possible candidate genes are presently under study.…”

“…For this randomized label experiment, the ATE over the 1,000 runs was close to the 50% no-information error rate, and significantly different from the results with true labels. Furthermore, to investigate the stability of the classifiers with respect to the single sample level, the 14 sampletracking profiles ( Supplementary Information, s-Figure 1) were computed, 23 Finally, genes' signature was assessed using the multiplicity occurrence in each of the 441 lists generated by the feature selection as a ranking: genes with higher occurrence are considered as strong discriminants between ARMS positive and negative for t(2;13). Table II shows the list of these top 50 genes.…”

“…This candidate molecular signature has been verified with a multiple resampling schema, implementing a complete validation protocol. 22,23 Furthermore, we have identified some novel deregulated genes possibly implicated in ARMS pathogenesis.…”

Toxicogenomic markers for corticosteroid treatment in beef cattle: Integrated analysis of transcriptomic data

“…2, left panel). This is an approach broadly detailed and used in the literature (Ambroise and McLachlan, 2002;Di Camillo et al, 2012;Furlanello et al, 2005).…”

“…applying the bootstrap externally to the selection process, has been proved to be an effective countermeasure against unwanted selection bias effects able to (i) realistically estimate the classification accuracy on external, previously unseen samples (Ambroise and McLachlan, 2002); (ii) select biomarkers in a precise and reproducible way Furlanello et al, 2005).…”