Multi-View Learning to Unravel the Different Levels Underlying Hepatitis B Vaccine Response

Abstract

The immune system acts as an intricate apparatus that is dedicated to mounting a defense and ensures host survival from microbial threats. To engage this faceted immune response and provide protection against infectious diseases, vaccinations are a critical tool to be developed. However, vaccine responses are governed by levels that, when interrogated, separately only explain a fraction of the immune reaction. To address this knowledge gap, we conducted a feasibility study to determine if multi-view modeling could aid in gaining actionable insights on response markers shared across populations, capture the immune system's diversity, and disentangle confounders. We thus sought to assess this multi-view modeling capacity on the responsiveness to the Hepatitis B virus (HBV) vaccination. Seroconversion to vaccine-induced antibodies against the HBV surface antigen (anti-HBs) in early converters (n = 21; <2 months) and late converters (n = 9; <6 months) and was defined based on the anti-HBs titers (>10IU/L). The multi-view data encompassed bulk RNA-seq, CD4+ T-cell parameters (including T-cell receptor data), flow cytometry data, and clinical metadata (including age and gender). The modeling included testing single-view and multi-view joint dimensionality reductions. Multi-view joint dimensionality reduction outperformed single-view methods in terms of the area under the curve and balanced accuracy, confirming the increase in predictive power to be gained. The interpretation of these findings showed that age, gender, inflammation-related gene sets, and pre-existing vaccine-specific T-cells could be associated with vaccination responsiveness. This multi-view dimensionality reduction approach complements clinical seroconversion and all single modalities. Importantly, this modeling could identify what features could predict HBV vaccine response. This methodology could be extended to other vaccination trials to identify the key features regulating responsiveness.

Keywords: immunoinformatics; integrative algorithms; interpretable framework; predictive framework; systems vaccinology.

Figure 1

Age distributions reveal differences between classes. Average (range) age, in years, per seroconversion class: early converters 35.9 (21.3–50.2), late converters 44.19 (36.3–48.5), non-converters 23.53 (21.6–27.2).

Figure 2

Research methodology workflow. The color coding represents the integration philosophies (blue, orange and grey) and, in green, highlights the interpretability endpoints. Antibody (AB); Blood Transcription Module (BTM); Hepatitis B Virus (HBV); Leave-One-Out Cross-Validation (LOOCV); Machine Learning (ML); Multi-view Canonical Correlation Analysis (MCCA); Principal Component Analysis (PCA); Synthetic Minority Oversampling Technique (SMOTE). Figure created using BioRender (BioRender.com accessed on 1 February 2023).

Figure 3

The Principal Component analysis of anti-HBs titers correlated with responsiveness. (A) Antibody titers at different sampling time points on a logarithmic scale, divided across the three response classes. Lines colored by converter type (early converter = orange; later converter = green; non-converter = blue). (B) Principal component (PC) 1 and PC2 of the antibody titer data. Colored by converter type (early converter = orange circle; later converter = green diamond; non-converter = blue square).

Figure 4

Correlation matrix shows overlap of cohort features. Data represent the correlation coefficient across each non-genetic feature (see Table S1 for cohort details).

Figure 5

MCCA latent space projection provides superior class separation. Two-dimensional data projection with (A) PCA and (B) MCCA of the integrated multi-view vaccine dataset. Each scatter plot showcases the first two dimensions. Data represent the early converters (orange circle) and late converters (green diamond) after applying each model.

Figure 6

MCCA latent space projection showing inflammatory markers and characterizing part of the cohort. Two-dimensional data projection with (A) PCA and (B) MCCA after the application of a feature correlation overlay. Features significantly correlated to the MCCA-projected space revealed a distinct inflammation-associated region. Marker sizes are proportional to the age of the individual. Feature vector colors identify separate views and module functions. Vector coordinates were calculated using the Person’s correlation for the respective latent component. Patients are colored by converter type (early converter = orange; later converter = green). Granulocytes count (day 0) (GRA0); hematocrit (day 0) (HCT0); normalized ratio of vaccine-specific TCRs (day 0) (HepBTCRs); hemoglobin protein count (day 0) (HGB0); monocytes count (day 0) (MON0); red blood cells count (day 0) (RBC0).

Figure 7

Single view performance comparison showing the superior predictive power of the TCR-seq layer. Methods performance comparison for the Logistic Regression classifiers trained on single modality data in terms of (A) AUC and (B) Accuracy. The colored dots indicate the performance of each of the 20 LOOCV runs. Area under the receiver operating characteristic curve (AUC); T-Cell Receptor sequencing data (TCR); T-Cell Receptor Sequencing (TCR-seq).

Figure 8

Classifiers performance comparison showing the positive impact of joint data integration on predictions. Methods performance comparison for the (A) Logistic Regression (LR) and (B) Random Forest classifiers in terms of AUC and accuracy. The colored dots indicate the performance of each of the 20 LOOCV runs.

Figure 9

Unimodal feature importance analysis bolstering insights into biological markers tied to vaccine responsiveness. Importance of leave-one-out cross validated LR coefficients for the unimodal models: (A) Cell counts, (B) CD4+ T-cells, (C) Metadata, (D) BTMs. Positive coefficients favored early response predictions while negative coefficients favored late-response predictions. The vertical red line indicates the zero-importance threshold. BTMs with unknown functionality have been excluded from the visualization. Individual dots indicate feature importance during each of the 20 LOOCV runs.