Tu224 - Asthmagraph – Knowledge Graph Framework for Drug Target Prioritization in Asthma

Travis Ahn-Horst – Data Scientist, Precision Medicine and Computational Biology, Sanofi; Emanuele de Rinaldis – Global Head of Precision Medicine and Computational Biology, Precision Medicine and Computational Biology, Sanofi; Shameer Khader – Head of Computational Biology Cluster, Precision Medicine and Computational Biology, Sanofi; Hamid Mattoo – Immune Mediated Disease Lead, Precision Medicine and Computational Biology, Sanofi; Mario Pandolfi – Data/Software Engineer, RCH Solutions, Inc; Franck Rapaport – Director of PMCB Data Science, Precision Medicine and Computational Biology, Sanofi

Abstract Text: Knowledge graphs (KGs) use a graph-structured data model to store interlinked descriptions of entities – objects, events, situations or abstract concepts – with free-form semantics (unstructured data). They could be thought as of a collection of statements (e.g. “A activates B”) that are gathered into a domain knowledge and organized in a graph structure. KGs are a powerful way to efficiently store and retrieve domain findings, as well as interrogate their relationships through graph “traversing”. Importantly, KGs also allow for the inference of new relationships not present in the initial collection of statements used as data input, as well as the effect of perturbations in the graph.
Starting from a large database of pathway and network interactions (Qiagen Biological Knowledge Base), we extracted the subnetwork of entities related to asthma (AsthmaGraph), based on pathogenic cell-type, asthma related pathways and genes. Using the KG framework as a foundation, we analyzed the causative connections between genes and immune conditions related to asthma. Furthermore, using a graph embedding, we applied deep learning and infer properties of weakly characterized entities in the network.
Our results demonstrate a significant enrichment in clinically validated therapeutic targets amongst the causative connections, supporting the effectiveness of this approach in the target identification process for drug discovery and clinical development. By expanding the same approach to other immune conditions, we will also be able to reposition known drugs on other medical indications.