Abstract Text: To understand the immune system, we need to embrace and welcome its complexity. The immune system comprises multiple cell types that work together to develop an effective response. Which of myriad cell types are important in a particular response, however, is not well understood. In recent years, new technological breakthroughs allowed to obtain enormous amounts of information from a single blood draw by measuring multiple cell types simultaneously. This led to the explosion of data and we are nowadays in need of tools to help us understand data. One solution is to apply machine learning algorithms to extract meaningful patterns from such high-dimensional data. To accomplish that, we developed an approach to automate a machine learning process, named Sequential Iterative Modeling “OverNight” (SIMON). SIMON is specifically suited for clinical data collected across multiple cohorts containing inconsistent features with many missing values. Our approach runs 180 machine learning algorithms to find the ones which fit any given data distribution. Such process maximizes predictive accuracy of the generated models. SIMON was applied to data from five clinical studies across eight consecutive influenza seasons and in COVID-19 infection. Over 3,000 parameters were considered, including serological responses , serum cytokines, cell subset phenotypes, and cytokine stimulations. SIMON identified several immune cell subsets, that correlated with an effective antibody response to influenza vaccination and that can predict COVID-19 severity. Overall, SIMON is a powerful tool for data-driven research that facilitates pattern recognition and knowledge extraction from high-dimensional clinical data collected across multiple cohorts.
