https://www.mdu.se/

Current cancer therapies including surgery, radiotherapy and chemotherapy are often plagued by high failure rates. Designing more targeted and personalized treatment strategies requires a detailed understanding of druggable tumor drivergenes. As a consequence, the detection of cancer driver genes has evolved to a critical scientific field integrating both high-through put experimental screens as well as computational and statistical strategies. Among such approaches, network based prediction tools have recently been accentuated and received major focus due to their potential to model various aspects of the role of cancer genes in a biological system. In this chapter, we focus on how graph centralities obtained from biological networks have been used to predict cancer genes. Specifically, we start by discussing the current problems in cancer therapy and the reasoning behind using network based cancer gene prediction, followed by an outline of biological networks, their generation and properties. Finally, we review major concepts, recent results as well as future challenges regarding the use of graph centralities in cancer gene prediction.

Graph centralities are commonly used to identify and prioritize disease genes in transcriptional regulatory networks. Studies on small networks of experimentally validated protein-protein interactions underpin the general validity of this approach and extensions of such findings have recently been proposed for networks inferred from gene expression data. However, it is largely unknown how well gene centralities are preserved between the underlying biological interactions and the networks inferred from gene expression data. Specifically, while previous studies have evaluated the performance of inference methods on synthetic gene expression, it has not been established how the choice of inference method affects individual centralities in the network. Here, we compare two gene centrality measures between reference networks and networks inferred from corresponding simulated gene expression data, using a number of commonly used network inference methods. The results indicate that the centrality of genes is only moderately conserved for all of the inference methods used. In conclusion, caution should be exercised when inspecting centralities in reverse-engineered networks and further work will be required to establish the use of such networks for prioritizing disease genes.

The prioritization of genes based on their centrality in biological networkshas emerged as a promising technique for the prediction of phenotype related genes.A number of methods have been developed to derive one such type of network, i.e.transcriptional regulatory networks, from expression data. In order to reliably prioritizegenes from such networks, it is crucial to investigate how well the inferencemethods reconstruct the centralities that exist in the true biological system. We haverecently reported that the correlation of centrality rankings between reference andinferred networks is only modest when using an unbiased inference approach. In thisstudy we extend on these results and demonstrate that the correlation remains modestalso when using a biased inference utilizing a priori information about transcriptionfactors. However, we show further that despite this lack of a strong correlation, theinferred networks still allow a signicant prediction of genes with high centralities inthe reference networks.

Motivation: Medulloblastoma (MB) is a brain cancer predominantly arising in children. Roughly 70% of patients are cured today, but survivors often suffer from severe sequelae. MB has been extensively studied by molecular profiling, but often in small and scattered cohorts. To improve cure rates and reduce treatment side effects, accurate integration of such data to increase analytical power will be important, if not essential.

Results: We have integrated 23 transcription datasets, spanning 1350 MB and 291 normal brain samples. To remove batch effects, we combined the Removal of Unwanted Variation (RUV) method with a novel pipeline for determining empirical negative control genes and a panel of metrics to evaluate normalization performance. The documented approach enabled the removal of a majority of batch effects, producing a large-scale, integrative dataset of MB and cerebellar expression data. The proposed strategy will be broadly applicable for accurate integration of data and incorporation of normal reference samples for studies of various diseases. We hope that the integrated dataset will improve current research in the field of MB by allowing more large-scale gene expression analyses.