Elucidating humoral profiles associated with Schistosomiasis pathogenesis using interpretable machine learning

Introduction:: Schistosomiasis, caused by parasitic flatworms (genus Schistosoma) is the fourth most prevalent parasitic disease, affecting over 207 million people across Africa, South America and Asia. Conventional diagnostic approaches are limited to the counting of worm eggs in patient stool/urine samples or mass drug administration. While recent studies have attempted to characterize humoral responses such as total IgG titers against soluble egg antigen (SEA), these approaches cannot provide early diagnosis of patients (before they develop acute schistosomiasis), and completely fail to distinguish between low intensity active Schistosomiasis (SEA+ Egg-) and high intensity active Schistosomiasis (SEA+ Egg+). Our study is the first characterization of deep antibody-omic profiles in a neglected tropical disease coupled to interpretable machine learning.

Materials and Methods:: To address the challenges outlined above, we use a high-throughput antigen-specific bead-based assay to profile different functional states (Fc binding, glycosylation, isotypes/subclasses) of patient antibodies from serum samples coupled to a novel interpretable machine learning method that moves beyond predictive biomarkers to putative causal latent factors of disease pathophysiology. Using our assay, we first deeply characterized (isotypes/subclasses, Fc receptor binding and glycosylation) humoral profiles corresponding to 18 different antigenic specificities resulting in a dataset of 216 measurements that provide deep insights into the corresponding humoral responses. We then analyzed these using Essential Regression (ER), a novel latent factor regression approach that comes with rigorous provable statistical guarantees regarding inference beyond prediction.


Results, Conclusions, and Discussions:: We first focused on differences between Egg+ and Egg- patients and identified 2 discriminatory latent factors (context-specific humoral signatures). The first was IgG4-centric – a hallmark anti-inflammatory/low polyfunctionality humoral response that is well known to characterize acute schistosomiasis infection. The other was a novel factor comprising crosstalk between anti-Calumenin and anti-CD63 humoral profiles. Within the SEA+ group, a targeted comparison of Egg+ and Egg- patients unveiled 4 latent factors which captured crosstalk between polyfunctional IgG3/IgA antibodies and Fc binding profiles characteristic of antibody-dependent monocyte phagocytosis (FcR2A), neutrophil phagocytosis (FcR3B) and NK activation (FcR3A).

Further, elucidation of the SEA+ Egg- patients using hybrid machine learning approaches (unsupervised approaches followed by ER) identifies two putative endotypes within these subjects – passive/past infection and active but low intensity infection. In addition to again recapitulating the canonical IgG4 signature, ER honed in on a second latent factor involving novel crosstalk between specific IgG subclass and FcR binding profiles, suggesting an ongoing polyfunctional response in the active low intensity group relative to the passive/past infection group. Interestingly, antibodies against Sm25, a protein present on the surface tegument of the parasite, were immunodominant and could be used as potential biomarkers for early diagnosis of infection. Deeper functional analyses confirmed that the active but low intensity stage is an intermediate state between passive infection and acute infection with IgG1 and FcR1 profiles characterizing this switch, opening avenues for early therapeutic intervention to prevent acute infection. We uncover novel humoral signatures that can guide biomarker discovery for early diagnosis and new therapeutic approaches.



Acknowledgements (Optional): :

References (Optional): :