[Predicting tumor drug sensitivity with multi-omics data].

Area of Science:

• Computational Biology
• Genomics
• Pharmacogenomics

Background:

• Accurate prediction of tumor drug sensitivity is crucial for personalized cancer treatment.
• Existing models often rely on single data types, limiting predictive power.
• Integrating diverse omics data offers a more comprehensive approach to drug sensitivity prediction.

Purpose of the Study:

• To develop and validate a multi-omics data-driven model for predicting cancer drug sensitivity using Stacking ensemble learning.
• To assess the model's performance compared to single-omics and existing multi-omics approaches.
• To explore potential drug resistance mechanisms through feature gene analysis for clinical interpretability.

Main Methods:

• Utilized multi-omics data (gene expression, mutation, copy number variation) from the Genomics of Drug Sensitivity in Cancer (GDSC) database.
• Applied Stacking ensemble learning, integrating six primary learners and one secondary learner.
• Implemented multiple feature selection methods for dimensionality reduction and validated the model using 5-fold cross-validation.

Main Results:

• The Stacking multi-omics model demonstrated superior accuracy and stability in predicting drug sensitivity compared to single-omics models.
• High Area Under the Curve (AUC) values were achieved: 36.4% of drug models exceeded 0.9, and 49.0% ranged between 0.8-0.9.
• Functional annotation identified potential tumor resistance mechanisms to sorafenib, offering biological insights.

Conclusions:

• The developed Stacking multi-omics model significantly enhances the accuracy and reliability of cancer drug sensitivity prediction.
• Multi-omics data integration via Stacking learning provides a robust framework for personalized medicine.
• The model's interpretability supports its potential application in guiding clinical medication decisions.