ImmunoResponse Predictor: a GUI for accurate response prediction to immunotherapy

Immune checkpoint inhibitors (ICIs) have improved outcomes for subsets of patients with metastatic urothelial carcinoma (mUC) and metastatic renal cell carcinoma (mRCC), yet objective response rates remain low (∼15–25%), underscoring the need for tools that support patient stratification. We previously developed LogitDA, a logistic regression–based predictor incorporating feature selection and domain adaptation, which outperformed established immune-related signatures in predicting response to the PD-L1 inhibitor atezolizumab. Here, we present the ImmunoResponse Predictor, a web-based clinical decision-support framework that enables responsible application of LogitDA in real-world settings. The system integrates standardized transcriptomic preprocessing, interpretable individual-level predictions (including single-sample use), clinically motivated LogitDA score cutoffs that prioritize minimization of false negatives, and a cohort-level percentage of applicability with empirically derived thresholds designed to assess whether predictions can be reliably extrapolated to newly uploaded datasets. Importantly, applicability functions as a diagnostic safeguard against distributional shift rather than as a response predictor. We evaluated the framework across four independent cohorts, PCD4989g(mUC), PCD4989g(mRCC), the Moreno cohort, and the UNC-108 cohort. LogitDA achieved prediction accuracies of 0.69, 0.83, 0.86, and 0.53, with corresponding applicability estimates of 74%, 76%, 71%, and 48%, respectively, correctly identifying the UNC-108 cohort as one in which predictions warrant increased caution. Overall, the ImmunoResponse Predictor extends LogitDA into a practical, interpretable, and safeguarded tool for immunotherapy response prediction, supporting cautious clinical and translational use.