Automating Predictive Toxicology Using ComptoxAI

Abstract

ComptoxAI is a new data infrastructure for computational and artificial intelligence research in predictive toxicology. Here, we describe and showcase ComptoxAI's graph-structured knowledge base in the context of three real-world use-cases, demonstrating that it can rapidly answer complex questions about toxicology that are infeasible using previous technologies and data resources. These use-cases each demonstrate a tool for information retrieval from the knowledge base being used to solve a specific task: The "shortest path" module is used to identify mechanistic links between perfluorooctanoic acid (PFOA) exposure and nonalcoholic fatty liver disease; the "expand network" module identifies communities that are linked to dioxin toxicity; and the quantitative structure-activity relationship (QSAR) dataset generator predicts pregnane X receptor agonism in a set of 4,021 pesticide ingredients. The contents of ComptoxAI's source data are rigorously aggregated from a diverse array of public third-party databases, and ComptoxAI is designed as a free, public, and open-source toolkit to enable diverse classes of users including biomedical researchers, public health and regulatory officials, and the general public to predict toxicology of unknowns and modes of action.

Figure 1.

ComptoxAI homepage, showing navbar links to major features (data browsing tools, installation guide, documentation, etc.), a hierarchy diagram of the major entity types included in the knowledge base, and additional introductory text. The website is available at https://comptox.ai.

Figure 2.

Interactive data portal for ComptoxAI, demonstrating how individual nodes (entities) in the knowledge base can be queried. The gene shown is found by searching for the gene symbol CYP2E1, and details about the gene are represented in the “Node details” panel. The identified gene can be sent to other data browsing utilities using the buttons at the bottom of the panel. Out of view are components for displaying relationships and network paths.

Figure 3.

ComptoxAI’s Neo4j graph database browser. Users can manually expand and manipulate subgraphs and can also use the powerful Cypher query language to submit database queries or call specific subroutines. In this interface, the left panel allows users to query a random subset of specific node labels (entity types) or relationship types, and the right portion of the network view panel lists the properties annotated to the currently selected node in the network.

Figure 4.

Network diagram of shortest paths joining perfluorooctanoic acid (PFOA) with nonalcoholic fatty liver disease (NAFLD) in ComptoxAI. The network indicates 14 genes associated with NAFLD, 7 of which are downregulated and 8 of which are upregulated (1 is both) by PFOA. Gene nodes are arranged (via interactive drag-and-drop) to group downregulated (top) and upregulated (bottom) genes. Full-resolution image available on FigShare.

Figure 5.

Adverse outcome pathway network generated by expanding a network around the node representing dioxin (2,3,7,8-tetrachlorodibenzo-p-dioxin), with network communities labeled in red. Communities roughly correspond to the following adverse outcomes: (A.) liver steatosis, (B.) impaired fertility and bile acid regulation, (C.) long-term risk of liver cancer, and (D.) other liver cancer, especially hepatocellular carcinoma. Dioxin (2,3,7,8-tetrachlorodibenzo-p-dioxin) plays a central role as a “hub” in the network. Node types are chemicals (red), AOPs (green), molecular initiating events (orange), and key events (purple). Full-resolution image available on Figshare.

Figure 6.

Screenshot image of the “Build QSAR Dataset” module from the ComptoxAI website. Users specify an assay endpoint, (optional) list of related chemicals to filter by, whether to include a “discovery” dataset of chemicals with no known assay measurements for making novel predictions, and format of the dataset. The dataset is generated dynamically and returned within several seconds of submitting the query.