AOPxGeneNet: Cross-Linking Gene Co-expression Networks with Adverse Outcome Pathways (AOPs) for mechanistic exploration of transcriptomic data

AOPxGeneNet is an interactive, reproducible workflow and Streamlit application that connects gene co-expression networks to the Adverse Outcome Pathway (AOP) framework to help researchers move from “lists of differentially expressed genes” to mechanistically grounded hypotheses (Key Events → Adverse Outcomes). What it does AOPxGeneNet operationalizes a 3-stage pipeline (exposed as a simple UI): Build a gene co-expression network (GraphML) from expression data Runs a WGCNA-style workflow and exports a gene co-expression network in GraphML, plus an edge list and module assignments. Optionally exports one GraphML per module, so users can inspect modules independently (useful for Cytoscape/Gephi workflows and targeted downstream analyses). Map AOP knowledge to genes and compute a ranked gene–KE–AO overlapThe AOPxGeneNet mapping layer links the WGCNA output to AOP-Wiki resources in a cached, reproducible way: Downloads and quarterly-caches AOP-Wiki tables for KE components and KE relationships. Computes a topological distance from each Key Event (KE) to the nearest Adverse Outcome (AO) within the same AOP using KE–KE relationships (graph-based propagation). Joins AOP-mapped genes to genes present in the co-expression network and produces a ranked overlap table that combines (i) proximity to AO, (ii) gene-level DE signals (e.g., log2FC/padj when available), and (iii) network connectivity metrics (degree/strength/mean edge weight). Integrate and visualize the merged AOP–gene network Merges an AOP network (JSON) with a gene co-expression network (GraphML) and renders an interactive HTML network view. Supports a scalable “focus mode” for large graphs by letting users select Key Events and restrict neighbor expansion by edge-weight threshold and max-neighbors. The AOP network JSON is designed to be obtained directly from AOPGraphExplorer v2 exports (node-link JSON) Why this is strong (core value) Mechanistic anchoring: AOPxGeneNet ties co-expression structure to curated mechanistic knowledge (MIE/KE/AO), enabling interpretable hypotheses rather than purely statistical associations. Prioritization that blends multiple evidence types: it ranks candidate genes not just by DE or connectivity alone, but by integrating AOP topology (KE→AO distance), expression perturbation, and network centrality in one evidence-aware table. Interoperable outputs: the pipeline produces widely-used exchange formats (GraphML, CSV tables, and interactive HTML) so results can move easily into downstream tools (e.g., Cytoscape/Gephi/NetworkX workflows). Reproducible by design: AOP-Wiki downloads are cached (quarterly) and key inputs are hashed, reducing “silent drift” and making results easier to reproduce across runs. Who can use it (examples) Academic toxicology & systems biology: mechanistic interpretation of transcriptomics, module hubs, and KE/AO consistency checks for a phenotype of interest. Pharma & biotech: hypothesis generation for target/pathway prioritization and biomarker discovery framed in mechanistic adverse outcome logic. Chemical safety & regulatory science: support weight-of-evidence exploration by linking observed molecular changes to structured AOP knowledge. Other sectors working with stress biology/omics: cosmetics, environmental monitoring, materials safety, occupational health—any domain where mechanistic anchoring of omics signals is valuable. Current status and roadmap This Zenodo release provides the working AOP×GeneNet integration and ranking workflow plus interactive visualization. Upcoming planned improvements include: SEM/causal modeling layer to move beyond “gene flowers around KEs” into testable structural models over AOP-informed graph structures. Improved network reduction and mechanistic summarization for very large graphs (scalable subgraph extraction and reporting).