Department of Biostatistics Seminar/Workshop Series
Sparse Additive Index Model for Survival Prediction with Genomic Data
Sijian Wang, PhD, Associate Professor, Department of Biostatistics & Medical Informatics Department of Statistics University of Wisconsin-Madison
In this talk, motivated by genomic studies with survival outcomes, we propose a sparse additive-index model to integrate pathway information to survival models. The method simultaneously constructs an index for each pathway and estimates the corresponding link function to connect the index to the outcome. A novel constraint is proposed to solve the identifiability issue when regularization on index parameters is present. Our proposed method can not only identify important pathways, but also select important genes within selected pathways. Furthermore, the proposed method has three good properties:
1) It is flexible to model the nonlinear association between genes and survival phenotype; 2) It automatically considers the interactions among genes within the same pathway; 3) It may distinguish the effects of a gene in all of pathways it belongs to. We have studied the theoretical properties of the methods. The methods are demonstrated using simulation studies and analysis on a TCGA ovarian cancer dataset.