In Silico Treatment: a computational framework for animal model selection and drug assessment

The translation of findings from animal models to human disease is a fundamental part in the field of drug development. However, only a small proportion of promising preclinical results in animals translate to human pathophysiology. This underscores the necessity for novel data analysis strategies to accurately evaluate the most suitable animal model for a specific purpose, ensuring cross-species translatability. To address this need, we present In Silico Treatment (IST), a computational method to assess translation of disease-related molecular expression patterns between animal models and humans. By simulating changes observed in animals onto humans, IST provides a holistic picture of how well animal models recapitulate key aspects of human disease, or how treatments transform pathogenic expression patterns to healthy ones. Furthermore, IST highlights particular genes that influence molecular features of pathogenesis or drug mode of action. We demonstrate the potential of IST with three applications using bulk transcriptomics data. First, we assessed two mouse models for idiopathic pulmonary fibrosis (IPF): one involving injury with intra-tubular Bleomycin exposure, and the other Adeno-associated-virus-induced, TGF{beta}1-mediated tissue transformation (AAV6.2-TGF{beta}1). Both models exhibited gene expression patterns resembling extracellular matrix derangement in human IPF, whereas differences in VEGF-driven vascularization were observed. Second, we confirmed known features of non-alcoholic steatohepatitis (NASH) mouse models, including choline-deficient, l-amino acid-defined diet (CDAA), carbon tetrachloride hepatotoxicity injury (CCl4) and bile duct ligation surgery (BDL). Overall, the three mouse models recapitulated expression changes related to fibrosis in human NASH, whereas model-specific differences were found in lipid metabolism, inflammation, and apoptosis. Third, we reproduced the strong anti-fibrotic signature and induction of the PPAR signaling observed in the Elafibranor experimental treatment for NASH in the CDAA model. We validated the contribution of known disease-related genes to the findings made with IST in the IPF and NASH applications. The complete data integration IST framework, including an interactive app to integrate and compare datasets, is made available as an open-source R package. Author summaryPreclinical testing plays a pivotal role in the drug development process, serving as a crucial evaluation phase before a new drug can be tested on humans in clinical trials. The drug must undergo a rigorous evaluation in in vivo and in vitro preclinical studies to assess its safety and efficacy. However, positive outcomes in preclinical animal models do not always translate to positive results in humans, mainly due to biological differences. Therefore, selecting an animal model that closely mirrors human disease traits and detecting and accounting for model limitations is of paramount importance. Over the last decade, the availability of gene expression data in both animals and humans has substantially increased. Gene expression states and perturbations are routinely employed as a proxy to predict and understand changes in disease states. Here, we developed In Silico Treatment, a computational method designed to overlay the gene expression changes observed in animals onto humans, quantifying the change in human disease status. We applied this method to mouse models for idiopathic pulmonary fibrosis and non-alcoholic steatohepatitis, two severe fibrotic diseases. We successfully identified known features of the disease models and provide a granular gene-level rationale behind our predictions. Consequently, our method shows promise as an effective approach to improve animal model selection and thus clinical translation.