ImmunoResponse Predictor: a GUI for accurate response prediction to immunotherapy

Area of Science:

• Oncology
• Immunotherapy
• Bioinformatics

Background:

• Immune checkpoint inhibitors (ICIs) offer benefits for metastatic urothelial carcinoma (mUC) and metastatic renal cell carcinoma (mRCC), but low response rates necessitate improved patient stratification.
• Existing immune-related signatures have limitations in predicting response to specific ICIs like atezolizumab.

Purpose of the Study:

• To introduce the ImmunoResponse Predictor, a web-based framework for applying the LogitDA model in clinical settings.
• To enable reliable prediction of immunotherapy response and assess the applicability of predictions across diverse patient cohorts.

Main Methods:

• Developed a logistic regression-based predictor (LogitDA) with feature selection and domain adaptation.
• Integrated standardized transcriptomic preprocessing, interpretable predictions, and cohort-level applicability assessment.
• Evaluated the framework on four independent cohorts: PCD4989g(mUC), PCD4989g(mRCC), Moreno, and UNC-108.

Main Results:

• LogitDA achieved prediction accuracies ranging from 0.53 to 0.86 across cohorts.
• Applicability estimates varied from 48% to 76%, correctly flagging the UNC-108 cohort for cautious interpretation.
• The framework demonstrated robust performance and identified limitations in specific datasets.

Conclusions:

• The ImmunoResponse Predictor is a practical, interpretable, and safeguarded tool for predicting immunotherapy response.
• It supports cautious clinical and translational use of LogitDA for patient stratification in mUC and mRCC.
• The tool enhances the application of predictive models by assessing data suitability and prediction reliability.