OpenTox 2022 Virtual Conference
A Meta Database for Feeding the Data Hungry AI-based Knowledge Hub to Improve In Silico Chemical Safety Assessment
J Marusczyk, B Hobocienski, C Yang
MN-AM, Nürnberg Germany; Columbus, OH, USA
The knowledge hub (KH) concept introduced by COSMOS Next Generation (NG) is designed to enable users to access remote systems sharing a common set of communication rules. The KH concept requires technical infrastructure support within a cooperative framework. For example, COSMOS NG was designed to interface with external in silico tools and databases. COSMOS NG originated from a SEURAT-1 project and we are expanding the concept to the new ONTOX project to leverage resources from multiple European Projects. One of the goals of the ONTOX project is to enable the KH with artificial intelligence (AI) approaches for risk assessment where data from new approach methodologies (NAMs) can empower users along with traditional in vivo data. AI approaches are data hungry for all types of information at all granularity while capable of extracting relational information on demand. The efficiency of AI-driven methodologies can often be improved by incorporating prior knowledge such as known relationships between data, and specific information about computational models. In this paradigm, we are designing a meta database that will not only house the relational information, but also comprehensive model metadata. As a simple example, our concept of a meta database will be used to transform the QSAR database, created at JRC under the principles of OECD guidelines, into an actionable knowledge generator. The ONTOX/COSMOS KH will house detailed documentation for every model (e.g., parameters, algorithms, training sets, validation, software system(s) used, etc.). The meta database can be interfaced with modeling platforms so that models can query the meta database for model-specific information. The first prototype deliverable of the ONTOX project is the demonstration how these two worlds are coordinated by the KH. We will use ChemTunes∙Express and VEGA models within the ONTOX meta database as the proof-of-concept. Also planned in the meta database is qAOP (quantitative adverse outcome pathways) methodology using a well-established endpoint such steatosis as our first model. Again, the goal is for the KH to execute these methods supported by the meta database housing the model parameters while the KH system is able to run the containerized software engines from the backend. The design of these systems will truly move our in silico chemical safety assessment toward making maximum use of current knowledge.
CV: Dr. Jörg Marusczyk is CTO of MN-AM and for software sustainability at ONTOX. He is involved in the development of ChemTunes·ToxGPS, a unique chemoinformatics platform and toxicity knowledgebase to support the chemical safety and risk assessment process of human health and regulatory-relevant endpoints. His main scientific interests are related to novel approaches in chemoinformatics deep learning.