Leveraging TCGA gene expression data to build predictive models for cancer drug response

• Evan A. Clayton  |  John F. McDonald and Peng Qiu  |  Toyya A. Pujol

• Integrated Cancer Research Center Georgia Institute of Technology USA

Predictive models may aid oncologists with making critical treatment decisions. A machine learning model is built using gene expression data from patients’ primary tumor tissues to predict whether a patient will respond positively or negatively to two chemotherapeutics: 5-Fluorouracil and Gemcitabine. It was also found that the genes most informative for predicting drug response were enriched in well-known cancer signaling pathways and their potential significance in chemotherapy prognosis was highlighted. As Fluorouracil and Gemcitabine provide the largest numbers of patients within TCGA they were chosen for this model. Normalized gene expression data were clustered and used as the input features. Matching clinical trial data was used to ascertain the response of the patients via multiple classification methods. Multiple clustering and classification methods were compared for prediction accuracy of drug response. Clara and random forest were found to be the best clustering and classification methods,respectively. The models predict with up to 86% accuracy.

Input variables : Patients’ clinical trial data and transcriptomic data from patients’ primary tumor samples
Output Variables : Patient's response to two chemotherapeutics: 5-Fluorouracil and Gemcitabine (Positive/ Negative)

Visit Model : github.com

Additional links : researchgate.net