Independent research · Science fair · High school
Predicting cancer drug response from gene expression with machine learning
Can a tumor's gene-expression profile predict how strongly it responds to a cancer drug? Using publicly available cancer datasets, this project trained Random Forest models to predict drug response (IC50 values) and compared predicted against observed responses. The project qualified for the Utah State science fair.
- Utah State Qualifier
- Random Forest models
- Public cancer datasets
Question
The same drug can work well against one tumor and barely touch another. If the genes a cancer cell is expressing carry a signal about how sensitive it will be, a model could help match patients to drugs.
Approach
The project used publicly available cancer datasets pairing gene-expression measurements with drug-response data. Response was measured as IC50, the drug concentration needed to inhibit growth by half; a lower IC50 means a more sensitive cell line.
Random Forest regression models were trained to predict IC50 from gene-expression features, and their accuracy was judged by the correlation between predicted and observed responses.
Recognition
Qualified for the Utah State science fair.