Abstract
Our group aims to understand and describe the organisation principles of cell signalling that allow the diverse and context-specific cell responses and phenotypes.
It is well established that signalling responses happen through complex networks. However, most signalling research still uses linear pathways as the ground truth. Moreover, signalling responses are highly dependent on context, such as tissue type, genetic background etc and therefore these static pathways are not always suitable. There is also a high bias in the literature towards kinases and pathways for which reagents and prior knowledge is readily available. This leaves a huge dark space in our understanding of cell signalling and significantly hinders studies of its general principles. Data-driven methods to study cell signalling using phosphoproteomics have typically been developed on transcriptomics data. Due to fundamental differences of these data types, including the sparsity of phosphoproteomics, these methods perform poorly on signalling network inference. Finally, large-scale essentiality datasets that can shed light on the context in which different signalling-related genes and pathways are crucial have thus far been severely underused in largescale omics data integrative efforts, likely due to their inaccessibility to non-expert users.
In this talk I will present two projects where we try to mitigate some of the above issues. For the first one I will present a method for data-driven machine-learning-based approach that takes advantage of global phosphoproteomics datasets to predict kinase regulatory networks, including their direction and sign of regulation. Our predicted kinase regulatory network is able to recapitulate known signalling pathways and agrees with orthogonal validation datasets. We were able to provide predictions for a large fraction of the understudied kinase space and found that kinase regulatory networks are denser than previously suspected.
For the second one I will present CEN-tools, an integrative webserver and python package, that allows users to navigate the contexts of different gene essentialities. I will demonstrate examples of its use in discovering new gene-gene relationships and important putative signalling targets for different cancers.
Junior talk
Ping-Han Hsieh, PhD candidate in Marieke Kuijjer's group at NCMM, UiO, will present his work on "Adjustment of Spurious Correlations Derived from Quantile Normalization in RNA-Seq Co-expression Data."
Abstract: Read count obtained from RNA-Seq experiments are affected by different gene compositions as well as library size. Because of this, between-sample normalization is an essential step before comparing the similarity between genes in terms of expression pattern. Although the performance of many normalization methods has been largely evaluated in the context of differential gene expression, few have focused on the impact of normalization methods on co-expression analysis. In this study, we use tissue-specific genes and Y chromosomal genes to showcase how quantile-based normalization can give rise to spurious correlations for genes with the low read count. We dissected the problem by analyzing the implementations of the normalization methods and showed how these spurious correlations are introduced. To address this problem, we propose an adjustment method that removes the artificial linear relations introduced by the normalization method. The results show that the spurious correlations can be effectively removed after applying our adjustment method.