|
Abstract:
|
In network detection, the semiparametric odds ratio model (ORM) has several advantages: flexible in handling both discrete and continuous data and including interactions; avoiding the problem of model incompatibility; and invariant to biased sampling design. However, for high-dimensional data, the complexity of this model induces computational burden comparing with parametric approaches, e.g., Gaussian graphical model. In this paper, we adapt three novel methods for ORM: using ORM after marginal correlation screening; stepwise selection using partial odds ratio coefficient based on PC-simple algorithm; and using ORM with Lasso after marginal correlation screening. The theoretical supports of these methods are developed, and simulations are conducted to assess the performance of our methods.
|
