The application of sufficient dimension reduction in the prediction of lung cancer

Ruiling Liu, The University of Texas School of Public Health

Abstract

Using patients' genetic variants to predict the risk of lung cancer is a challenging task. In this study, we applied the sparse sufficient dimension reduction (sparse SDR) method to identify reduced genetic variant set that contributes to lung cancer susceptibility. Lung cancer data downloaded from dbGaP was used to build the predictive models using sparse SDR and to evaluate the performance. However, the performance of the sparse SDR on this task is just slightly better than random prediction. We also evaluated other methods, such as GWAS and sparse logistic regression, which showed similar results comparing to the sparse SDR. According to our results, germline mutation might not be the major cause of lung cancer and GWAS data might not be useful in the prediction of lung cancer. In the future, we will combine clinical features and genetic variants with reduced dimension identified by sparse SDR for better predictive modeling. We will also test the sparse SDR method using other cancer data.^

Subject Area

Biostatistics

Recommended Citation

Liu, Ruiling, "The application of sufficient dimension reduction in the prediction of lung cancer" (2015). Texas Medical Center Dissertations (via ProQuest). AAI1597538.
http://digitalcommons.library.tmc.edu/dissertations/AAI1597538

Share

COinS