(175bi) Navigating Biological Systems with PBPK, Text Mining, and AI: In silico NAMS for the Development of Reliable and Robust Qaops

In this research, we present a computational approach that starts with an environmental exposure and culminates in quantitatively linking it to disease through the construction of qAOPs. Our methodology integrates various tools and techniques spanning natural language processing (NLP), artificial intelligence (AI), physiologically based toxicokinetic (PBTK) models, and in silico systems biology. Herein, human biomonitoring (HBM) data is utilized to calibrate the PBTK model and translate external exposure into internal exposure. Furthermore, both in vitro and in experiments are conducted to generate -omics data, and consequently implementing joint pathway analysis to identify predominantly perturbed biological pathways. We introduce a text-mining toolbox to convert metabolic pathways into mechanistic systems biology models, resulting in a heterogeneous model comprising over 1300 differential equations. Model parameterization is carried out using data from the BRENDA and SABIORK databases, and further supplemented by AI models to estimate relevant enzyme properties, with data from the HMDB database utilized for model initialization. Additionally, we devise a methodology that employs machine learning (ML) and Generative Adversarial Networks to support model initialization, generating an ML model for each endogenous metabolite, regardless of whether its concentration is known, for validation purposes. The model undergoes two executions, integrating fold change results from -omics in the second run, thereby identifying metabolites with significant concentration changes and prompting the collection of publicly available data to scrutinize concentration variations and identify potential biomarkers in individuals affected by a disease. The resulting mathematical equation describing the endogenous metabolite provides a comprehensive network of interactions involving metabolites and genes. For adverse outcome pathway (AOP) development, a bottom-up approach is employed, which leverages our knowledge of the AO and available information about the molecular initiating event (MIE), with transcriptomics, network analysis, and literature review utilized to determine the precise MIE and establish the AOP using NLP tools.