Work

An interpretable machine learning model integrating mucosal and cellular immunity to predict SARS-CoV-2 reinfection: a multicenter cohort study

Year: 2026

Type: article

Abstract: Background: As the SARS-CoV-2 virus continues to mutate, there is a critical need for precision risk assessment. We developed and validated machine learning (ML) algorithms to assess COVID-19 reinfection risk by integrating clinical and multi-dimensional immunological data. Methods: We conducted a two-center prospective longitudinal cohort study involving 1,186 participants recruited from Wuxi and Nantong, China, between May 2024 and February 2025. Participants from the Wuxi site (n = 586) were allocated into a training cohort (n = 411) and an internal validation cohort (n = 175) using a 7:3 stratified random split. The Nantong cohort (n = 600) was designated as an independent external validation cohort. The Boruta algorithm was applied to select the most significant predictors from a pool of 21 clinical and immunological parameters. Six ML models were subsequently developed: adaptive boosting (ADB), gaussian naive bayes (GNB), K-nearest neighbors (KNN), logistic regression (LR), random forest (RF), and extreme gradient boosting (XGB). Predictive performance and clinical utility were rigorously evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA). Finally, SHAP values were utilized for model interpretation and transparency. Findings: During a median follow-up of 10 months, 128 (10.8%) reinfection cases were identified. The Boruta algorithm identified spike-specific nasal secretory immunoglobulin A (sIgA), spike-specific CD4⁺ central memory T cells (TCM), and spike-specific CD4⁺ naive T cells (TN) as the most critical predictors. Among six models, XGB demonstrated superior performance, achieving an AUC of 0.880 (95% CI: 0.782–0.977) in the internal validation and 1.000 (95% CI: 0.999–1.000) in the external validation. SHAP dependence plots further revealed that spike-specific nasal sIgA demonstrated a protective effect, spike-specific CD4+ TCM cells exhibited an optimal beneficial range, whereas elevated levels of spike-specific CD4+ TN cells were associated with an increased risk. Interpretation: We developed and validated an explainable ML model for predicting SARS-CoV-2 reinfection risk using large-sample, multicenter data. This framework supports individualized clinical decision-making and offers a scalable intelligent platform for managing future emerging infectious diseases.

Source: Emerging Microbes & Infections

DOI: 10.1080/22221751.2026.2746253