MOTIVATION: Recurrent DNA breakpoints in cancer genomes indicate the presence of critical functional elements for tumor development. Identifying them can help determine new therapeutic targets. High-dimensional DNA microarray experiments like arrayCGH afford the identification of DNA copy number breakpoints with high precision, offering a solid basis for computational estimation of recurrent breakpoint locations. RESULTS: We introduce a method for identification of recurrent breakpoints (consensus breakpoints) from copy number aberration datasets. The method is based on weighted kernel counting of breakpoints around genomic locations. Counts larger than expected by chance are considered significant. We show that the consensus breakpoints facilitate consensus segmentation of the samples. We apply our method to three arrayCGH datasets and show that by using consensus segmentation we achieve significant dimension reduction, which is useful for the task of prediction of tumor phenotype based on copy number data. We use our approach for classification of neuroblastoma tumors from different age groups and confirm the recent recommendation for the choice of age cut-off for differential treatment of 18 months. We also investigate the (epi)genetic properties at consensus breakpoint locations for seven datasets and show enrichment in overlap with important functional genomic regions. AVAILABILITY: Implementation in R of our approach can be found at http://www.mpi-inf.mpg.de/ ∼laura/FeatureGrouping.html. CONTACT: [email protected]. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

In excess of 12% of human cancer incidents have a viral cofactor. Epidemiological studies of idiopathic human cancers indicate that additional tumor viruses remain to be discovered. Recent advances in sequencing technology have enabled systematic screenings of human tumor transcriptomes for viral transcripts. However, technical problems such as low abundances of viral transcripts in large volumes of sequencing data, viral sequence divergence, and homology between viral and human factors significantly confound identification of tumor viruses. We have developed a novel computational approach for detecting viral transcripts in human cancers that takes the aforementioned confounding factors into account and is applicable to a wide variety of viruses and tumors. We apply the approach to conducting the first systematic search for viruses in neuroblastoma, the most common cancer in infancy. The diverse clinical progression of this disease as well as related epidemiological and virological findings are highly suggestive of a pathogenic cofactor. However, a viral etiology of neuroblastoma is currently contested. We mapped 14 transcriptomes of neuroblastoma as well as positive and negative controls to the human and all known viral genomes in order to detect both known and unknown viruses. Analysis of controls, comparisons with related methods, and statistical estimates demonstrate the high sensitivity of our approach. Detailed investigation of putative viral transcripts within neuroblastoma samples did not provide evidence for the existence of any known human viruses. Likewise, de-novo assembly and analysis of chimeric transcripts did not result in expression signatures associated with novel human pathogens. While confounding factors such as sample dilution or viral clearance in progressed tumors may mask viral cofactors in the data, in principle, this is rendered less likely by the high sensitivity of our approach and the number of biological replicates analyzed. Therefore, our results suggest that frequent viral cofactors of metastatic neuroblastoma are unlikely.

MOTIVATION: Classification and feature selection of genomics or transcriptomics data is often hampered by the large number of features as compared with the small number of samples available. Moreover, features represented by probes that either have similar molecular functions (gene expression analysis) or genomic locations (DNA copy number analysis) are highly correlated. Classical model selection methods such as penalized logistic regression or random forest become unstable in the presence of high feature correlations. Sophisticated penalties such as group Lasso or fused Lasso can force the models to assign similar weights to correlated features and thus improve model stability and interpretability. In this article, we show that the measures of feature relevance corresponding to the above-mentioned methods are biased such that the weights of the features belonging to groups of correlated features decrease as the sizes of the groups increase, which leads to incorrect model interpretation and misleading feature ranking. RESULTS: With simulation experiments, we demonstrate that Lasso logistic regression, fused support vector machine, group Lasso and random forest models suffer from correlation bias. Using simulations, we show that two related methods for group selection based on feature clustering can be used for correcting the correlation bias. These techniques also improve the stability and the accuracy of the baseline models. We apply all methods investigated to a breast cancer and a bladder cancer arrayCGH dataset and in order to identify copy number aberrations predictive of tumor phenotype. AVAILABILITY: R code can be found at: http://www.mpi-inf.mpg.de/~laura/Clustering.r.

MOTIVATION: In life sciences, interpretability of machine learning models is as important as their prediction accuracy. Linear models are probably the most frequently used methods for assessing feature relevance, despite their relative inflexibility. However, in the past years effective estimators of feature relevance have been derived for highly complex or non-parametric models such as support vector machines and RandomForest (RF) models. Recently, it has been observed that RF models are biased in such a way that categorical variables with a large number of categories are preferred. RESULTS: In this work, we introduce a heuristic for normalizing feature importance measures that can correct the feature importance bias. The method is based on repeated permutations of the outcome vector for estimating the distribution of measured importance for each variable in a non-informative setting. The P-value of the observed importance provides a corrected measure of feature importance. We apply our method to simulated data and demonstrate that (i) non-informative predictors do not receive significant P-values, (ii) informative variables can successfully be recovered among non-informative variables and (iii) P-values computed with permutation importance (PIMP) are very helpful for deciding the significance of variables, and therefore improve model interpretability. Furthermore, PIMP was used to correct RF-based importance measures for two real-world case studies. We propose an improved RF model that uses the significant variables with respect to the PIMP measure and show that its prediction accuracy is superior to that of other existing models. AVAILABILITY: R code for the method presented in this article is available at http://www.mpi-inf.mpg.de/ approximately altmann/download/PIMP.R CONTACT: [email protected], [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.