GEE-TGDR: A Longitudinal Feature Selection Algorithm and Its Application to lncRNA Expression Profiles for Psoriasis Patients Treated with Immune Therapies

Table 2

Comparison between the GEE-TGDR method and two competing algorithms.

Method	Size	Predictive error
GEE-TGDR	9	30%
GEE-based screening	50	40%
Linear mixed model-based screening	27	33.33%

The predictive errors were calculated on the basis of 10-fold crossvalidations. Here, the response status, i.e., if the PASI score experienced a reduction of 75% from the baseline affected skin after week 12 or later. Size: the number of identified lncRNAs by a specific method; here, the sizes trained on the whole dataset were given; in crossvalidations, these numbers were subject to changes since the training sets were a subset of the whole dataset. For GEE-TGDR- and GEE-based screening, only unstructured working correlation matrix was considered.