Paclitaxel Response Can Be Predicted With Interpretable Multi-Variate Classifiers Exploiting DNA-Methylation and...

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Paclitaxel is a key chemotherapy for breast cancer (BC).
• Drug resistance significantly limits paclitaxel efficacy.
• Predicting patient response is crucial for optimizing treatment.

Purpose of the Study:

• To investigate the predictability of breast cancer patient response to paclitaxel.
• To identify molecular features that can predict treatment outcomes.
• To develop machine learning classifiers for response prediction.

Main Methods:

• Utilized large-scale tumor genomic data from the National Cancer Institute's Genomic Data Commons.
• Assessed 10 machine learning algorithms on six molecular tumor profiles (DNA methylation, miRNA, etc.).
• Evaluated 60 resulting classifiers for their predictive performance, focusing on feature subset selection.

Main Results:

• DNA methylation and miRNA profiles were the most informative for predicting paclitaxel response.
• A complexity-optimized XGBoost classifier using CpG island methylation achieved an AUC of 0.74.
• A Decision Tree model using only 2 CpG sites achieved an AUC of 0.89, identifying key methylation biomarkers.

Conclusions:

• Identified specific molecular signatures from methylome and miRNome data that can predict paclitaxel response in breast cancer.
• Developed highly accurate predictive classifiers, including a Decision Tree using minimal features.
• Findings offer potential insights for optimizing paclitaxel-based therapies in clinical practice.