Methods, tools, and computational environment for network-based analysis of biological data

Abstract

Cancer currently affects more than 18 million persons world-wide annually. It is a leading cause of death and so far, only 60% cure rate can be reached within the most developed health care systems. The nature of cancer has been a mystery for centuries, until discoveries during recent decades shed light on the underlying molecular events. This depended on the progress in understanding cell and tissue biology, developments of molecular technologies and of -omics technologies. Cancer has then emerged as a highly heterogeneous disease, however with some very basic mechanistic features common to all cancers. To deal with the complexity of causes and consequences of pathological changes in the molecular machinery, methods and tools of network analysis can be helpful. Complexity of this task requires easy-to-use tools, which allow researchers and clinicians with no background in computer science to perform network analysis.

Paper I describes a web-based framework for network enrichment analysis (NEA), using previously developed algorithm and code. The developed platform introduces functionality for a researcher to use data pre-downloaded from various popular databases as well as own data, perform NEA and obtain statistical estimations, export results in different formats for publications or further use in research pipeline.

Paper II presents development of another web server, which provided vast opportunities for exploration and integrated analysis of multiple public cancer datasets that describe in vitro and in vivo sample collections. The web server linked molecular data at the single gene level, phenotype and pharmacological response variables, as well as pathway level variables calculated with NEA and connected to the framework presented in Paper I. Researchers can use the platform for creating multivariate models based on raw or pre-processed data from various sources, visualize created models, estimate their performance and compare them, export models for further usage in own research environments.

Paper III demonstrates NEAdriver, a practical application of NEA to probabilistic evaluation of driver roles of mutations reported in ten cancer cohorts. NEAdriver results are compared with cancer gene sets produced by other, both network analysis and network-free methods. The paper demonstrated ability of NEA to be used directly for discovering novel driver genes as well as being used in combination with other methods. In order to demonstrate benefits of using NEA, some rare cancer types and types with low mutation burden were used.

Paper IV is a manuscript evaluating performance of most representative methods of network analysis across methods’ parameters, functional ontologies and network versions. This study emphasizes discovery of novel functional associations for known genes, as opposed to previous tests dominated by a few “gold standard” genes which were well characterized previously. We performed the analysis in the context of various topological properties of networks, pathways of interest, and genes. It employed both existing, widely used topological metrics and a number of novel ones developed for this analysis.