A machine learning-driven framework accurately predicts MPA exposure and supports individualized dosing in childhood-onset LN.