W149 - Using Machine Learning to Harness the Complexities of Inflammasome Biology for Novel Drug Discovery

Elisa Cambronero; Wendy Cousin; Ben Kamens; Ben Komalo; Lauren Nicolaisen; Dat Nguyen; Tempest Plott; Will Van Trump; Michael Wiest; Daniel Chen; Christian Elabd

Abstract Text: Inflammasomes are critical components of our innate immune system that convert internal and external danger stimuli into pro-inflammatory signals. Dysregulation of these complex pathways leads to hyperinflammation and tissue damage, linking them to pathogenesis in many human diseases. Traditional readouts of inflammasome activation cannot sufficiently distinguish the effects of complex or convergent pathways, necessitating tools that can resolve differences between various cellular states of health, activation, and inhibition. Machine learning and advanced multidimensional data analysis can strengthen inflammasome drug discovery screening efforts by facilitating analysis of complex, physiologically relevant data.

Using primary human PBMCs under physiological inflammasome activation conditions we developed novel machine learning-based analytical tools for fluorescence microscopy images. Every image was analyzed using a mixture of automated quantification of targeted cell features, unbiased discovery of cell subpopulations, compound similarity metrics, and deep learning signatures specific to validated inflammasome cellular states. In parallel, we performed multidimensional cytokine analyses on supernatants from these same plates. We then scaled these experiments into a high content imaging compound screen using advanced lab automation. Multiple scoring rubrics were designed to categorize compounds into novel inflammasome inhibitor classes whose mechanisms of action can be uniquely matched to relevant human pathologies. Using this approach, we screened over 12,000 compounds and identified compounds from five mechanistic classes that could be used to treat inflammasome-mediated diseases, such as sepsis and gout. This approach could also be applied to other, similarly complex and disease-relevant biological processes to drive multiple clinical paths forward from a single compound screen.