ASCCT Award Winners Webinar
Thursday, January 19, 2023
10:00 -11:00 a.m. EST
Facilitating Global Connections through the Microphysiological Systems for COVID Research (MPSCoRe) Working Group
Amber Daniel, M.S.
The emergence and global spread of the coronavirus disease 2019 (COVID-19) emphasizes the need for effective approaches to prevent, control, and treat infectious diseases. Although animal models have historically been used to address such challenges, human cell-based in vitro platforms known as microphysiological systems (MPS) have the potential to more efficiently and effectively model the human lung and other organ systems affected by COVID-19. However, absent a venue for coordination, widespread application of MPS to COVID-19 research presents a risk of overlapping investigations and duplication of efforts among researchers. The MPS for COVID Research (MPSCoRe) working group was organized to reduce this risk by globally connecting key MPS stakeholders from the research, method development, drug and vaccine manufacturing, and regulatory sectors. The working group facilitates open communication among stakeholders to maximize the impact of MPS technologies in understanding disease mechanisms and treatments, and reducing animal use while improving human health. In this way, the group aims to promote adoption of MPS for studying COVID-19 and future emerging infectious diseases. Activities currently supported by MPSCoRe include a proof-of-concept study to evaluate MPS models for testing the safety and efficacy of novel COVID-19 therapeutics. Additionally, MPSCoRe is supporting development of a web-based repository for sharing COVID-19 experimental data and other MPS resources. These efforts will accelerate the development and adoption of MPS in infectious disease research, thereby reducing the reliance on animal models in this space. This project was funded with federal funds from NIEHS, NIH under Contract No. HHSN273201500010C.
Applying Deep Learning Toxicity Models Across the Chemical Universe
Chemicals are ubiquitously found in many exposure routes in day-to-day life. It is important that toxicity testing guides maximally informed decisions on chemical use and regulation to best protect human health. Traditional toxicity testing generally relies on in vivo or in vitro methods which are time-consuming, resource-intensive, and frequently results in flawed generalizations to humans. Collaborators at the FDA/NCTR developed two novel deep learning models called DeepCarc and DeepDILI to predict endpoints for carcinogenicity and drug-induced liver injury, respectively, outperforming state-of-the-art methods. These deep learning models incorporate an ensemble approach of five conventional machine learning algorithms integrated into a neural network to generate probabilistic predictions for carcinogenicity and liver injury. Here, DeepCarc and DeepDILI were reproduced and refined, and the externally validated models were applied to screen the carcinogenicity and liver injury potential of 7176 compounds in the Tox21 database. The majority of the Tox21 compounds were predicted to have low carcinogenicity concern with only 88 of the compounds possessing a carcinogenicity probability of greater than 0.9. DeepDILI predictions suggested that the compounds in Tox21 generally have a higher risk for liver injury with many compounds predicted to have a liver injury probability of greater than 0.9. Compounds with high toxicity probability predictions were run through Integrated Chemical Environment Characterization workflows to investigate the distribution of physicochemical properties, presence in consumer products, and bioactivity profiles from high-throughput screening assays. Such computational models can be used to rapidly screen large chemical libraries to prioritize potentially hazardous substances for further examination.