Utilizing stabl and MOFA to uncover predictive features in vaccine response from multi-omics data 3746

Abstract Description Massive biological datasets pose significant analytical challenges due to their heterogeneity. Recent advancements, such as the development of algorithms like Stabl and MOFA, provide effective solutions for identifying key biological features that explain individual responses or cell subset activity. MOFA employs dimensionality reduction to uncover critical features and is robust to missing values, while Stabl requires imputation but incorporates outcome information to prioritize relevant predictors. In this study, we applied Stabl and MOFA to two vaccine response datasets: FluPRINT (influenza) and CMI-PB (pertussis). For FluPRINT, MOFA identified IL-10 CD8+CD45RA+ pSTAT1 as the most significant feature for explaining response, defined as a ≥ 4-fold antibody titer increase from day 0 to day 28. Stabl highlighted CD161+CD45RA+ Tregs as the top predictive feature, achieving an AUROC of 0.75 when combined with two additional features, consistent with findings from the SIMON study (Tomic et al., 2019). For CMI-PB, both Stabl and MOFA identified baseline expression of VSTM1 and METRNL as key predictors of antibody titer fold change from day 0 to day 14. A regression model using these and other features achieved a Spearman coefficient of 0.8 for predicting the rank of antibody titer fold change. These results underscore the strength of Stabl and MOFA in uncovering predictive features, highlighting their utility for vaccine research and the analysis of complex multi-omics datasets. Funding Sources Supported by NIH 2U19AI057229 Topic Categories Computational and Systems Immunology (COMP)