A recursive framework for predicting the time-course of drug sensitivity.

Area of Science:

• Pharmacogenomics
• Computational Biology
• Systems Biology

Background:

• Drug mechanisms involve complex, dynamic biological processes.
• Time-course gene expression data offers insights but is underutilized in current drug response prediction models.
• Existing methods often analyze gene expression at limited time points, inadequately capturing dynamic gene-drug interactions, especially in long-term treatments.

Purpose of the Study:

• To develop a novel framework, REcursive Prediction (REP), for predicting drug response using time-course gene expression data.
• To leverage temporal gene expression patterns for predicting drug response at any stage of long-term treatment.
• To enhance the accuracy and robustness of drug response prediction by integrating past response values and handling noisy or missing data.

Main Methods:

• Developed a REcursive Prediction (REP) framework utilizing a recursive structure to exploit the temporal nature of gene expression data.
• Integrated tensor completion within the REP framework to manage noise, impute missing gene expression levels (GEXs), and predict unseen GEXs.
• Utilized historical drug response data as input for subsequent predictions, capturing the dynamic interplay between genes and drug effects.

Main Results:

• The REP framework demonstrated effectiveness in predicting drug response at various time points during long-term treatment.
• Tensor completion effectively mitigated the impact of noise and missing data in gene expression profiles.
• The framework successfully predicted drug response from initial gene expression measurements, showcasing its utility for ongoing therapy monitoring.

Conclusions:

• The REcursive Prediction (REP) framework offers a powerful approach for predicting drug response by fully utilizing time-course gene expression data.
• REP's recursive structure and tensor completion capabilities enable robust modeling of dynamic gene-drug interactions, crucial for personalized long-term therapies.
• The framework's ability to predict drug response across treatment duration holds significant potential for optimizing therapeutic strategies and patient outcomes.