Prediction of Chemosensitivity in Multiple Primary Cancer Patients Using Machine Learning

Area of Science:

• Genomics
• Computational Biology
• Pharmacogenomics

Background:

• Multiple primary cancers present a therapeutic challenge, requiring broad-spectrum or personalized treatments.
• Genomic information aids in predicting individual drug responses, crucial for personalized medicine.

Purpose of the Study:

• To identify gene sets sensitive to specific anticancer drugs.
• To compare the predictive accuracy of drug response models based on genomic features versus cancer types.

Main Methods:

• Utilized publicly available gene expression and drug sensitivity datasets for gastric and pancreatic cancers.
• Implemented five machine learning algorithms (LDA, CART, k-NN, SVM, Random Forest) to build predictive models.

Main Results:

• Genomic prediction models demonstrated higher accuracy (0.759–0.896 on testing data) compared to cancer type models (0.731–0.765).
• Specific gene models significantly outperformed general cancer type models in predicting drug sensitivity.

Conclusions:

• Predictive models based on specific gene expression profiles are superior for selecting effective cancer therapies.
• This approach allows for personalized drug selection in patients with multiple primary cancers, irrespective of cancer type.