Hierarchical Feature Selection Incorporating Known and Novel Biological Information: Identifying Genomic Features Predictive of Cancer Recurrence

• Activity Number: 224
• Type: Topic Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Yize Zhao*+ and Matthias Chung and Brent A. Johnson and Carlos Moreno and Qi Long
• Companies: Emory University and Virginia Tech and Emory University and Emory University and Emory University
• Keywords: Biological information ; Feature selection ; Imputation ; Missing data ; Regularization

Abstract

Our work is motivated by a prostate cancer study aimed at identifying mRNA and miRNA biomarkers that are predictive of cancer recurrence. It has been shown that incorporating biological information improves feature selection. However, biological information is often not fully known, for example, the role of miRNAs in regulating gene expression is not fully understood. To this end, we treat unknown biological information as missing data and propose a new concept of imputing unknown biological information based on observed data. In addition, we propose a hierarchical group penalty to encourage sparsity at both the pathway level and the within-pathway level, which, combined with the imputation step, allows for incorporation of known and novel biological information. We develop the proposed approach in the context of semiparametric AFT models motivated by our data example. Application to the motivating data shows that the incorporation of novel biological information improves prediction and the proposed penalty outperforms the extensions of existing ones. Our simulations further show the superiority of the proposed penalty over the others in both feature selection and risk prediction.