SignNets: Fine Tuning Gene-Gene Similarity Metrics in Biological Systems

• Activity Number: 361 - SPEED: Biometrics - Methods and Application, Part 2
• Type: Contributed
• Date/Time: Tuesday, July 30, 2019 : 11:35 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Crystal Shaw* and Vinayagam Arunachalam and Jadwiga R Bienkowska
• Companies: UCLA and Pfizer, Inc. and Pfizer, Inc.
• Keywords: gene networks; drug targets; network analysis; R Package

Abstract

The key to discovering novel drug targets is understanding the gene-gene interaction network within a given biological system. Recent advancements in high throughput technologies enables generation of heterogeneous omics datasets that capture various aspects of gene-gene relationships; the challenge is finding an appropriate metric to quantify these interactions. There are a number of options for similarity metrics, the most common of which is Pearson correlation. It is known, however, that Pearson correlation is sensitive to outliers, insensitive to non-linear relationships, and produces values regardless of signal strength. We propose a non-parametric approach called “SignNets” for quantifying similarity between genes that accounts for both signal strength and the direction of association. We apply our SignNets algorithm to build gene-gene relationship networks from CRISPR gene essentiality screens of more than 500 cell lines (project Achilles data). The resulting networks are not only complementary to those generated by Pearson correlation but are more relevant to cancer biology since lineage-specific relationships are natural derivatives of the algorithm.