PDXGEM

PDXGEM builds multi-gene predictive models by integrating gene expression and drug sensitivity data from patient-derived xenograft (PDX) models to identify biomarkers that predict therapeutic response in cancer patients.

Key Features:

• Multi-gene Expression Model: Constructs predictive models by integrating gene expression data with drug sensitivity information from PDX models and identifies biomarkers correlating with therapeutic outcomes.
• Random-Forest Algorithm: Implements a random-forest algorithm to build prediction models and selects biomarkers based on concordant co-expression patterns between PDX models and pre-treatment cancer patient tumors.
• Broad Applicability: Applies to various anti-cancer agents, including cytotoxic chemotherapies and targeted therapies, across cancer types such as breast, pancreatic, colorectal, and non-small cell lung cancers.
• Validation and Accuracy: Validated through independent tests on diverse cancer patient datasets from retrospective observational studies and prospective clinical trials, demonstrating significant accuracy in predicting pathological responses or survival outcomes.
• Clinical Translation Potential: Utilizes molecular profiles and drug activity data from PDX tumors to develop predictive biomarkers that can aid in tailoring cancer treatments to individual patient profiles.

Scientific Applications:

• Molecular mechanism insights: Provides insights into drug response mechanisms at the molecular level using PDX-derived gene expression and drug activity data.
• Therapeutic outcome prediction: Predicts therapeutic outcomes from gene expression data to support treatment selection, aiming to optimize strategies, reduce adverse effects, and improve patient care.

Methodology:

Associates gene expression levels with post-treatment tumor volume changes in PDX models, identifies drug sensitivity biomarkers, integrates those biomarkers with pre-treatment cancer patient tumor data, and uses a random-forest algorithm selecting biomarkers based on concordant co-expression patterns between PDX and patient tumors to construct predictive models.