Multi-modal data integration using Markov Field graphical networks predicts B cell depletion and immune responses associated with protective intravenous BCG (IV-BCG) vaccination against tuberculosis in macaques

Shu Wang, Amy J Myers, Edward B Irvine, Chuangqi Wang, Pauline Maiello, Mark A Rodgers, Jaime Tomko, Kara Kracinovsky, H. Jacob Borish, Michael C Chao, Douaa Mugahid, Patricia A. Darrah, Robert A. Seder, Mario Roederer, Charles A Scanga, Philana Ling Lin, Galit Alter, Sarah M Fortune, JoAnne L Flynn, Douglas A Lauffenburger

Paper Link: https://doi.org/10.1101/2024.04.13.589359

Link to Model Validation Datasets: https://fairdomhub.org/studies/1198

Abstract: Multi-modal datasets provide rich information that can help identify critical multi-scale interactions underlying biological systems. However, identifying associations between features and outputs can be beset by spurious connections due to indirect impacts propagating through an unmapped biological network. We applied a probabilistic graphical modeling approach, Markov Fields, to dissect correlations between immune features in a public multi-modal dataset (systems serology, cytokines, cytometry) of macaques undergoing intravenous BCG vaccination against tuberculosis. This yielded an interaction network that interprets network paths underlying vaccine efficacy, and shows how correlations between features often arise indirectly. We next conducted experimental depletion of B cells during vaccination in macaques—which did not reduce protection against tuberculosis—to validate our Markov Field model’s predictions of network-wide shifts post-depletion. Finally, we highlight immune changes predicted to strongly affect intravenous BCG vaccine efficacy, showing that probabilistic graphical models increase the interpretability of multi-modal datasets for identifying new disease targets.