Integrating Cellular Immune Biomarkers with Machine Learning to Identify Potential Correlates of Protection for a Trypanosoma cruzi Vaccine

<b>Background:</b> Chagas disease, caused by the protozoan parasite <i>Trypanosoma cruzi</i> (<i>T. cruzi</i>), remains a major public health concern in Latin America. No licensed vaccine exists to prevent or treat <i>T. cruzi</i> infection. Identifying correlates of protection (CoPs) could provide substitute endpoints to guide and accelerate vaccine development. Although most CoPs established to date are antibody-based, their utility has not been demonstrated in <i>T. cruzi</i> vaccine reports. Thus, this study aimed to explore alternative strategies considering the use of immune cells as potential CoPs. <b>Methods:</b> Mice were immunized with a vaccine candidate based on the <i>T. cruzi</i> trans-sialidase protein (TSf) and potentiated with 5-fluorouracil (5FU) to deplete myeloid-derived suppressor cells (MDSCs). Percentages of CD4⁺, CD8⁺, and CD11b⁺Gr-1⁺ cellular biomarkers were assessed by flow cytometry from the peripheral blood of immunized mice, which were subsequently challenged with a high dose of <i>T. cruzi.</i> A machine-learning (ML) model based on decision trees was applied to identify potential CoPs to predict survival by day 25 post-infection. <b>Results:</b> Individual biomarkers obtained from flow cytometry did not show strong predictive performance. In contrast, biomarker engineering led to a combination that integrated biomarkers rationally: summing the percentages of CD8⁺ and CD4⁺ cells and subtracting the percentage of CD11b⁺Gr-1⁺ MDSC-like cells (REB), enhanced the predictive capacity. Subsequent computational analysis and ML application led to the identification of a better and even improved potential Integrative CoP: 2 ∗ %CD8++ %CD4+ - %CD11b+ Gr1+(pICoP), which significantly improved the performance of a simple one-level decision-tree model, achieving an average accuracy of 0.86 and an average AUC-ROC of 0.87 for predicting survival in immunized and infected mice. <b>Conclusions</b>: Results presented herein provide evidence that integrating cellular immune biomarkers through rational biomarker engineering, together with ML analysis, could lead to the identification of potential CoPs for a <i>T. cruzi</i> vaccine.