ExposoGraph: An Interactive Platform for Carcinogen Bioactivation and Detoxification Pathway Visualization

ExposoGraph is an interactive knowledge-graph platform that unifies data from major carcinogen and pharmacogenomic resources to visualize and explore the complex pathways linking carcinogenic exposures, metabolic activation/detoxification, DNA damage, and genetic variations for improved gene-environment interaction research.

The Gist

Imagine your body is a bustling, high-tech city. Every day, this city is bombarded by "visitors" from the outside world—some are harmless tourists, but others are troublemakers known as carcinogens (chemicals that can cause cancer). These troublemakers include things like pollution, certain chemicals in food, or even hormones produced naturally by your body.

The problem is that these troublemakers don't just walk in and cause chaos immediately. They usually need to be "unlocked" or "activated" by the city's own security guards (enzymes) before they can break into the city's most important building: your DNA.

The Problem: A Scattered Map

Until now, scientists had a lot of information about these troublemakers and the guards, but the data was scattered across different libraries.

• One library listed the bad chemicals.
• Another listed the security guards (enzymes).
• A third listed the genetic "blueprints" that tell us if a guard is fast, slow, or broken.

If a researcher wanted to know, "How does a specific chemical affect a person with a specific genetic blueprint?", they had to manually visit all these different libraries, copy-paste notes, and try to draw a map by hand. It was slow, confusing, and easy to miss connections.

The Solution: ExposoGraph

The authors of this paper, Julhash U. Kazi and Kenneth J. Pienta, have built a new tool called ExposoGraph.

Think of ExposoGraph as a live, interactive GPS map for your body's chemical defense system. Instead of static books, it's a digital playground where you can see exactly how a chemical travels through your body, who tries to stop it, who accidentally helps it, and where it might cause damage.

How the Map Works (The Journey of a Chemical)

The map visualizes a chemical's journey in four distinct "zones":

1. The Activation Zone (Phase I):
   Imagine a troublemaker (like a carcinogen) arrives. Some security guards (enzymes like CYP1A1) try to "wake it up" to make it easier to deal with, but sometimes they accidentally turn it into a super-villain that can attack DNA. On the map, you can see this "activation" step clearly.
2. The Detox Zone (Phase II):
   Other guards (like GSTM1) try to neutralize the villain by wrapping it in a protective bubble (conjugation) so it can be safely thrown out of the city.
3. The Exit Zone (Phase III):
   Special transport trucks (ABC transporters) carry the neutralized waste out of the cell.
4. The Repair Crew (DNA Repair):
   If the villain manages to break into the DNA building, a repair crew (enzymes like OGG1) tries to fix the damage before it becomes a permanent mutation (cancer).

The "Genetic Twist"

Here is the coolest part: Not everyone's city has the same security guards.

Some people have "super guards" that work very fast. Others have "lazy guards" that work slowly. Some might be missing a guard entirely (like the GSTM1-null deletion mentioned in the paper).

• The Analogy: Imagine two people are exposed to the same amount of cigarette smoke (the troublemaker).
  • Person A has a genetic blueprint that makes their "detox guards" very slow. The troublemaker stays active longer, attacks the DNA, and causes damage.
  • Person B has "super guards" that neutralize the troublemaker instantly. They stay safe.

ExposoGraph lets you toggle these genetic settings. You can click on a chemical, then filter the map to show only people with "slow guards," and instantly see how the risk changes.

Why This Matters

The paper highlights a few key features:

• It connects the dots: It links the chemical (the exposure), the enzyme (the biology), and the DNA damage (the result) in one picture.
• It's interactive: You can search for a specific chemical (like "Benzene") or a specific gene (like "CYP1A1") and watch the relevant part of the map light up, while the rest fades away.
• It reveals hidden connections: For example, the map shows how testosterone (a male hormone) can sometimes be converted by the body into a form that damages DNA, linking hormone metabolism directly to cancer risk in a way that wasn't easy to visualize before.

The Bottom Line

ExposoGraph is a research tool, not a medical diagnosis app you can use at home today. However, it is a massive leap forward for scientists.

It turns a messy pile of data into a clear, navigable story. It helps researchers ask better questions like, "Why do some people get cancer from this chemical while others don't?" and gives them a visual map to find the answer. By understanding these interactions, we can eventually move toward personalized medicine, where we can tell individuals exactly how their unique genetics might make them more or less vulnerable to the chemicals in their environment.

Technical Summary

1. Problem Statement

Despite the existence of extensive resources cataloging carcinogenic exposures (e.g., IARC Monographs) and pharmacogenomic variations (e.g., PharmVar, CPIC), there is a critical gap in unified platforms that integrate these domains. Current tools lack a single interactive framework that connects:

• Exposure: Specific chemical carcinogens.
• Metabolism: Bioactivation (Phase I) and detoxification (Phase II/III) pathways.
• Genetic Modulation: Pharmacogenomic variants affecting enzyme activity.
• Outcome: DNA damage (adducts) and repair mechanisms.

Researchers currently must manually synthesize data across disparate databases (CTD, KEGG, PharmCAT, etc.) to understand gene-environment interactions (GxE), a process that hinders hypothesis generation, risk assessment, and clinical translation.

2. Methodology

The authors developed ExposoGraph, a Carcino-Genomic Knowledge Graph implemented as an interactive, web-based visualization platform.

Data Curation & Sources

The knowledge graph was constructed by systematically curating data from seven primary sources:

• IARC Monographs: For carcinogen classifications.
• KEGG Pathway Database: For metabolic routes (specifically xenobiotic metabolism, steroid biosynthesis, and DNA adduct pathways).
• PharmVar & CPIC: For allele nomenclature, functional classifications, and genotype-to-phenotype translation rules.
• CTD (Comparative Toxicogenomics Database): For chemical-gene interaction data.
• GTEx Portal: For tissue-specific enzyme expression context.
• Primary Literature (PubMed): For specific variant-function annotations.

Knowledge Graph Architecture

The graph is a directed network comprising 96 nodes and 102 edges, organized into five entity types and six relationship types:

• Node Types (5):
  1. Carcinogens: 15 nodes covering 9 classes (PAHs, heterocyclic amines, aromatic amines, nitrosamines, mycotoxins, estrogens, androgens, solvents, alkylating agents).
  2. Enzymes: 36 nodes (14 Phase I, 14 Phase II, 3 Phase III transporters, 5 DNA repair enzymes).
  3. Metabolites: 28 nodes (intermediates and ultimate reactive species).
  4. DNA Adducts: 11 nodes representing covalent DNA modifications.
  5. Pathways: 6 nodes representing KEGG reference pathways.
• Edge Types (6):
  • ACTIVATES (Phase I oxidation/reduction/hydrolysis).
  • DETOXIFIES (Phase II conjugation).
  • TRANSPORTS (Phase III export).
  • FORMS_ADDUCT (Reactive metabolite to DNA).
  • REPAIRS (Repair enzymes to adducts).
  • PATHWAY (Entity to pathway context).

Visualization Implementation

The platform utilizes a D3 HTML-based force-directed network visualization with three interfaces:

1. Plotly: Rapid interactive previews.
2. Standalone D3 HTML: Local deployment with drag-and-drop and export capabilities.
3. Dash Cytoscape: Advanced browser-based viewer supporting complex layouts (COSE, hierarchical, circular) and detailed filtering.

• Interactivity: Features include carcinogen-class filtering, node/edge-type toggling, substring search, hover tooltips, and click-to-inspect detail panels containing provenance, variant data, and tissue context.

3. Key Contributions

• Unified Framework: The first platform to integrate carcinogen metabolism, pharmacogenomic variation, and DNA damage pathways into a single, interactive knowledge graph.
• Androgen-Genotoxicity Bridge: A novel extension linking androgen metabolism to estrogen-mediated genotoxicity. It visualizes how CYP19A1 (aromatase) converts testosterone to estradiol, which is then metabolized into quinones that form DNA adducts, bridging receptor-mediated proliferation and direct DNA damage.
• Pharmacogenomic Scoring Foundation: The graph topology supports the calculation of activity scores based on enzyme position (activation vs. detoxification vs. repair), laying the groundwork for composite risk modeling.
• Open Access: The tool is freely available via a web browser (Streamlit), local installation (Docker/Python), and GitHub, with documentation provided.

4. Results

• Graph Statistics: The current reference graph covers 9 carcinogen classes, 15 index carcinogens, 36 enzymes, 28 metabolites, and 11 DNA adducts.
• Pathway Visualization: The system successfully visualized complex pathways, including:
  • Polycyclic Aromatic Hydrocarbons (PAH): Tracing Benzo[a]pyrene from CYP1A1/1B1 activation to BPDE-DNA adduct formation, highlighting the protective role of GSTM1/GSTP1.
  • Aromatic Amines: Mapping 4-aminobiphenyl metabolism through CYP1A2 and the competing roles of NAT1 (activation) and NAT2 (detoxification).
  • Mycotoxins: Visualizing Aflatoxin B1 activation by CYP3A4/1A2 and the balance between GST-mediated detoxification and XPC-mediated repair.
• Interactive Exploration: The platform allows users to filter by carcinogen class (e.g., isolating the "Androgen" branch reduces the view to 26 nodes and 19 edges) and search for specific enzymes (e.g., "CYP1") to reveal neighborhood context and variant annotations.
• Risk Scoring Prototype: A preliminary visualization demonstrated how activity scores for 13 enzymes across 8 carcinogen classes can be aggregated to show class-level activation vs. detoxification capacity.

5. Significance and Future Directions

• Hypothesis Generation: The tool enables researchers to identify gaps in GxE interaction data, facilitating the design of new epidemiological studies.
• Personalized Risk Modeling: By integrating individual genotypes (e.g., GSTM1-null, NAT2 slow acetylator) with exposure data, the platform supports the development of individualized cancer risk models.
• Education: Serves as a dynamic teaching resource for understanding chemical carcinogenesis and metabolic pathways.
• Limitations & Future Work:
  • Currently limited to a curated subset of well-characterized carcinogens; expansion to IARC Group 2A agents and other classes (heavy metals, radiation) is planned.
  • Activity scores are currently literature-derived; future work aims to standardize these with large-scale functional assays.
  • The model does not yet account for dynamic enzyme induction/inhibition by co-exposures.
  • Future iterations will incorporate machine learning (Graph Neural Networks) to predict novel carcinogen-enzyme relationships and integrate population-level genomic data (e.g., All of Us, gnomAD).

In conclusion, ExposoGraph represents a significant step forward in "carcino-genomics," providing a necessary infrastructure to move from static database queries to dynamic, interactive exploration of how genetic susceptibility modulates environmental cancer risk.