Predictive modeling of co-infection in lupus nephritis using multiple machine learning algorithms.

Area of Science:

• Immunology
• Medical Informatics
• Nephrology

Background:

• Lupus nephritis (LN) patients exhibit altered immune profiles, increasing susceptibility to infections.
• Understanding peripheral blood lymphocyte subset changes is crucial for managing LN complications.

Purpose of the Study:

• To analyze peripheral blood lymphocyte subsets in lupus nephritis (LN) patients.
• To develop a machine learning (ML) algorithm for predicting co-infection in LN patients.

Main Methods:

• Compared lymphocyte subsets (T, B, helper T, suppressor T, NK, Tregs) in 111 non-infected LN, 72 infected LN patients, and 206 healthy controls (HCs).
• Evaluated eight ML algorithms (Logistic Regression, Decision Tree, KNN, SVM, MLP, Random Forest, Ada boost, XGBoost) for predictive accuracy.
• Validated the best-performing ML model on a separate test group.

Main Results:

• Infected LN patients showed significantly lower levels of T, B, helper T, suppressor T, and natural killer cells compared to non-infected LN patients and HCs.
• Regulatory T cells (Tregs) were significantly lower in LN patients than HCs, with the lowest counts in infected LN patients.
• Extreme Gradient Boosting (XGBoost) achieved the highest accuracy and precision in predicting LN co-infections.

Conclusions:

• Disruptions in innate and adaptive immunity are evident in LN patients.
• Monitoring lymphocyte subsets aids in the prevention and treatment of infections in LN.
• The XGBoost algorithm is a promising tool for predicting co-infection in lupus nephritis.