Exploring Selective Edge Pruning to Integrate Domain Knowledge in Biological Networks
This is the support page for the poster presentation at the Complex Network conference (2023). You can find and download the PDF version.
Project Context
The aim of my PhD project is to integrate multiple data types to improve the subtyping of Muscle-Invasive Bladder Cancer (MIBC). The standard approach is to apply various computational methods separately on each data type and then integrate multiple strands of information at the computational level. The integration is expected to lead to more precisely defined groups.
Poster Overview
As the title suggests, the work presented explores how selective edge pruning affects the network structure and communities. The network is constructed from a Spearman correlation matrix of the gene expression from non-cancerous samples (88) from which the top 5000 most varied genes are kept. Transcription Factors (TFs) regulate the expression of other genes, thus the aggressive pruning strategy was relaxed for this subset of genes (325/500), allowing a higher minimum degree than the standard genes.
This exploration started by selecting a minimum degree of 50 for TFs using the Leiden algorithm, while the communities are found with the Stochastic Block Model (SBMs), and TFs with a minimum of 6 edges. The model-specific metric, the number of communities, and how many TFs are eventually used in the MIBC stratification were used to assess the right network configuration; the result figures can be seen in the poster with the two networks which can also be interacted with below on this page.
Broadly, this work is part of the more fundamental question which I raise in my doctoral dissertation: What is the optimal network representation of molecular biology? My use of molecular biology is as an umbrella term for multiple data types - gene expression, mutations, TFs, and domain knowledge.
Decree Corrected Stochastic Block Model (DC-SBM) - 6TF
Leiden with Modularity Maximisation - 50TF
Acknowledgments
My participation to the conference wouldn’t have been possible without the support from the department of Physics, Engineering and Technology (PET) and York Against Cancer or YAC. The latter is an amazing local charity that funds cancer research in the Jack Birch Unit (JBU).