Workshop SDS: Registration
View Submission - SDS2022
A0158
Title: Effect of bootstrapping in sparse biological data and model selection via likelihood ratio test Authors:  Mehmet Ali Kaygusuz - Middle East Technical University (Turkey) [presenting]
Vilda Purutcuoglu - Middle East Technical University (Turkey)
Abstract: The likelihood ratio test (LRT) is very useful to compare with various models since it assesses the goodness of fit of competitive models based on the ratio of their likelihoods. From previous studies, it has been shown that LRT is preferable while alternative models are nested to each other and from recent studies, it has been observed that it is more computationally efficient than other model selection criteria. On the other hand, although the reduction in the description of the full model is challenging in complex models, it can be computational efficient when the difference between the number of parameters $p$ and the number of observations $n$ is large. We propose different bootstrap schemes to fill the underlying difference between $n$ and $p$, and represent the data via a Gaussian graphical model. In model selection, we perform LRT due to its advantages in computational efficiency and high accuracy. As alternative models, we generate nested models from the results of k-means clustering so that we can evaluate comprehensively the combined effect of bootstrapping and LRT in the analyses of high-dimensional biological network data. In our assessment, we use simulated networks under different sparsity and sample sizes