In immuno-oncology, quantitative systems pharmacology (QSP) models are widely used to describe complex immune–tumor interactions and support drug development. However, modern drug development generates large biomarker datasets, making it challenging to determine which biomarkers should be incorporated into mechanistic modeling frameworks. We developed an artificial intelligence (AI)–enhanced pharmacometrics workflow for biomarker-driven modeling. Longitudinal biomarker trajectories were transformed into AI training–ready features (baseline, early change, slope, and area under the curve) and used as inputs for tree-based machine learning algorithms, including Random Forest (RF) and Extreme Gradient Boosting (XGB), to predict clinical tumor response. SHapley Additive exPlanations (SHAP) analysis was applied to quantify biomarker contributions and interpret model predictions. The RF model identified key biomarkers and features linked to treatment response, which may serve as key drivers for semi-mechanistic PK/PD or QSP models describing major immune signaling pathways. In parallel, the ML model framework itself functions as an AI-based platform capable of predicting clinical tumor response and patient outcomes. This integrated approach combining explainable machine learning model with pharmacometrics modeling enables large-scale biomarker screening and response prediction, thereby accelerating AI/ML-driven development for immuno-oncology therapeutics.