Differences in innate immune ‘imprinting’ between vaccine adjuvants may mediate dissimilar effects on the quantity/quality of persisting adaptive responses. We compared antibody avidity maturation, antibody/memory B cell/CD4+ T cell response durability, and recall responses to non-adjuvanted fractional-dose antigen administered 1-year post-immunization (Day [D]360), between hepatitis B vaccines containing Adjuvant System (AS)01B, AS01E, AS03, AS04, or Alum (NCT00805389). Both the antibody and B cell levels ranked similarly (AS01B/E/AS03 > AS04 > Alum) at peak response, at D360, and following their increases post-antigen recall (D390). Proportions of high-avidity antibodies increased post-dose 2 across all groups and persisted at D360, but avidity maturation appeared to be more strongly promoted by AS vs. Alum. Post-antigen recall, frequencies of subjects with high-avidity antibodies increased only markedly in the AS groups. Among the AS, total antibody responses were lowest for AS04. However, proportions of high-avidity antibodies were similar between groups, suggesting that MPL in AS04 contributes to avidity maturation. Specific combinations of immunoenhancers in the AS, regardless of their individual nature, increase antibody persistence and avidity maturation.
Dataset drift is a common challenge in machine learning, especially for models trained on unstructured data, such as images. In this article, we propose a new approach for the detection of data drift in black-box models, which is based on Hellinger distance and feature extraction methods. The proposed approach is aimed at detecting data drift without knowing the architecture of the model to monitor, the dataset on which it was trained, or both. The article analyzes three different use cases to evaluate the effectiveness of the proposed approach, encompassing a variety of tasks including document segmentation, classification, and handwriting recognition. The use cases considered for the drift are adversarial assaults, domain shifts, and dataset biases. The experimental results show the efficacy of our drift detection approach in identifying changes in distribution under various training settings.