Automated construction of Boolean models using knowledge graphs
Date(s) : 21/05/2026 iCal
17h00 - 17h45
Abstract : Boolean modeling has emerged as a powerful qualitative formalism to study the dynamics of gene regulatory networks, and in particular in cancer. A Boolean model (BM) consists of a Regulatory graph (RG) — a signed directed graph capturing regulatory interactions — coupled with a set of logical rules, called network parametrization, that govern the activation state of each node. By simulating these dynamics, BMs predict the network’s evolution through various functional states in response to different environmental conditions.
However, constructing the RG remains a time-consuming and largely manual process due to the large amount of heterogeneous data available. Here, we present an automated pipeline for RG construction and BM inference.
First, a knowledge graph (KG) is assembled by integrating multiple prior knowledge databases using OntoWeaver and BioCypher: protein-protein interactions and pathways (OmniPath), drug-target interactions (OpenTargets), cancer gene biomarkers (OncoKB), and tissue-specific gene expression (Human Protein Atlas). Each database is integrated via dedicated Ontoweaver adapters, and BioCypher harmonizes the chosen ontology. Then, a context-specific subset of proteins is extracted from the KG with user-defined biological queries. Tools such as NeKo are then used for network parametrization thereby building the RG for the BM with this subset of proteins.
The approach is benchmarked against the manually constructed BM of Flobak et al. (2015) on AGS gastric cancer cells, assessing whether the automated network recapitulates the same drug synergy predictions. Automating PKN construction is a step toward personalized BM integrating patient-specific omics data.
Emplacement
I2M Luminy - TPR2, Salle de Séminaire 304-306 (3ème étage)
Catégories
