Table of contents

• About the Translator project, Team Expander Agent, and ARAX
• ARAX analyzes knowledge graphs to answer biomedical questions
• How does ARAX work?
• The Reasoners Standard Application Programming Interface
• What knowledge providers does ARAX use?
  • RTX-KG2
  • Columbia Open Health Data (COHD)
  • PubMed
  • Identifier mapping
• Team Expander Agent: who we are
  • Principal investigators
  • Team members
• What is RTX? How does it differ from ARAX?
• Organization of the ARAX/RTX software repository
  • subdirectory code
    • subdirectory code/ARAX
    • subdirectory code/ARAX/Examples
    • subdirectory code/UI/OpenAPI
    • subdirectory code/UI/Feedback
    • subdirectory code/UI/interactive
    • subdirectory code/kg2
    • subdirectory code/kg2/mediKanren
    • subdirectory code/reasoningtool/kg-construction
    • subdirectory code/reasoningtool/SemMedDB
    • subdirectory code/reasoningtool/QuestionAnswering
    • subdirectory code/reasoningtool/MLDrugRepurposing
    • subdirectory code/autocomplete
  • subdirectory data
  • Key repository branches
• License
• Disclaimer
• Credits
• Installation and dependencies
• Contact us
• Try out ARAX/RTX...
  • ...in your web browser
  • ...using our web API
  • ...by customizing a Jupyter Notebook
• Links
  • General links for the Biomedical Data Translator project
  • ARAX- and RTX-specific links

About the Translator project, Team Expander Agent, and ARAX

We are Expander Agent, a team of researchers and software experts working within a consortium effort called the Biomedical Data Translator program ("Translator program"). Initiated by the NIH National Center for Advancing Translational Sciences (NCATS), the Translator program's goal is to accelerate the development of disease therapies by harnessing artificial intelligence, Web-based distributed computing, and computationally-assisted biomedical knowledge exploration. Although the Translator program is only in its third year, Translator tools are already being used by biomedical researchers for hypothesis generation and by analysts who are supporting clinical management of rare disease cases. Key intended applications of Translator include repositioning already-approved drugs for new indications (thus accelerating time-to-market), identifying molecular targets for developing new therapeutic agents, and providing software infrastructure that could enable development of more powerful clinical decision support tools. The 18 teams on the translator project are working with NCATS to achieve this goal by building the Translator software, a modular system of Web services for biomedical knowledge exploration, reasoning, and hypothesis generation.

After a multiyear feasibility assessment (2017-2019), during which our team built a prototype Translator reasoning tool called RTX, the Translator program is moving into a prototype development phase during early 2020. In this new phase, our team is building a modular web-based software system, ARAX, that enables expressive, automatable, and reproducible exploration and analysis of biomedical knowledge graphs without requiring computer programming expertise. As we develop ARAX, we will provide up-to-date software code for ARAX and RTX to the scientific community via this software repository. While ARAX is currently under development, our team will demonstrate ARAX for NCATS in mid-March 2020; at that time we will make ARAX publicly available on the Web and we will provide demonstration web pages and notebooks that illustrate how to access and operate it.

ARAX analyzes knowledge graphs to answer biomedical questions

ARAX is a tool for querying, manipulating, filtering, and exploring biomedical knowledge graphs. It is designed to be a type of middleware—an autonomous relay agent—within the Translator system. The top-level layer of Translator (which is called the autonomous relay system) will issue structured queries to ARAX via ARAX's web application programming interface. Then, based on the query type, ARAX will determine which knowledge providers it needs to consult in order to be able to answer the query; ARAX will then query the required knowledge providers, synthesize the information that it gets from those queries, and respond to the top-level layer in a standardized structured data format. When completed, ARAX will contribute to and advance the Translator program in four key ways:

1. ARAX provides a powerful domain-specific language (DSL), called ARAXi (technical documentation on ARAXi can be found here), that is designed to enable researchers and clinicians to formulate, reuse, comprehend, and share workflows for biomedical knowledge exploration. One of the key advantages of ARAXi is that it is not a general-purpose programming language; it is purpose-built for the task of describing—in user friendly syntax—a knowledge graph manipulation workflow in terms of ARAX's modular capabilities. All of ARAX's capabilities are exposed through ARAXi. Using ARAXi, an analyst can:

• define a small query graph; query for entities (e.g., proteins, pathways, or diseases) that match the search criteria represented in the query graph
• expand a knowledge graph, pulling in concepts that are related to concepts that are already in the knowledge graph
• filter a knowledge graph, eliminating concepts or relationships that do not match a given set of search criteria
• overlay contextual information from large datasets (such as co-occurrence of terms in clinical health records or in abstracts of articles in the biomedical literature)
• resultify: enumerate and return matches of a query graph against a larger knowledge graph; as a sub-case of this step, ARAX can return as a single result, all concepts from the knowledge graph that match a given concept type and that match a given pattern of neighbor-concept-type relationships.

2. ARAX is based on a modular architecture. It provides distinct, orthogonal, and human-understandable knowledge graph manipulation and analysis capabilities via five operations (query graph, expand, overlay, filter, and resultify) that can be accessed individually or in combination by the Translator top-level layer, other Translator tools, or by individual researchers directly using ARAX. Due to this transparent and modular design, the five ARAX operations are easy-to-use in isolation and easy to compose into workflows. Unifying these modules within a single service framework (ARAX) also provides significant speed benefits for workflows that are implemented end-to-end within ARAX, because the knowledge graph is stored on the server and does not need to round-trip to the client with each operation. Through example ARAX-powered analysis vignettes linked below, we describe how an ARAX workflow using ARAXi can be much more powerful than the sum of its individual parts.

3. ARAX is a Web service that speaks the information standard—called the Reasoners Standard Application Programming Interface—that Translator has adopted for data interchange between Translator components. Team Expander Agent has been at the forefront of the development and stewardship of the Reasoners Standard API (as described below), and with this perspective, ARAX was built from the ground up to seamlessly interoperate with other Translator software components. In addition to complying with the Translator standard for inter-reasoner communication, ARAX uses knowledge sources (see below) that comply with the Biomedical Data Translator Knowledge Graph Standard (for which our team has been an active participant in the standards development process) which is based on the Biolink Model.

4. ARAX is integrated with the RTX reasoning tool's knowledge graphs and graph visualization capabilities. During the NCATS Translator feasibility assessment phase, our team built a prototype reasoning tool system called RTX, whose knowledge graphs (both the first-generation knowledge graph RTX-KG1 and the second-generation knowledge graph **RTX-