PRECISE

PRECISE adapts drug response predictors trained on pre-clinical models (cell lines and patient-derived xenografts, PDXs) for application to human tumor data to improve translational fidelity of drug-response inference.

Key Features:

• Domain Adaptation Methodology: Employs a domain adaptation approach that captures shared characteristics between cell lines, PDXs, and human tumors by deriving a consensus representation.
• Predictor Training and Application: Trains drug-response predictors on pre-clinical model data within the shared representation and applies those predictors to stratify human tumors.
• Performance Evaluation: Reports a reduction in predictive performance within the pre-clinical domain while recovering known associations between independent biomarkers and their corresponding drugs when applied to human tumor data.

Scientific Applications:

• Oncology translational research: Bridges pre-clinical models and clinical tumor data to enable more reliable inference of drug responses in human tumors.
• Personalized medicine support: Supports stratification of tumors by predicted drug sensitivity to inform translational and precision oncology studies.

Methodology:

Quantifies similarities between cell lines, PDXs, and human tumors to derive a consensus, domain-invariant representation and trains drug-response predictors on pre-clinical data within that shared representation for application to human tumors.