Objective.To detect genes containing both rare and common variants from next-generation sequencing data with longitudinal measure of phenotypic traits, using penalized generalized estimating equations (PGEE) and penalized quadratic inference function (PQIF).Methods.We simulated longitudinal binary hypertension data derived from the GAW18 genetic data.Weighted sum statistics were constructed to collapse both common and rare variants over genes and then were included into different models:PGEE and PQIF.We examined the two penalized methods, based on the true positives (TP) rate, false positives (FP) rate and mean squared error (MSE).The two methods were also applied to the original GAW18 data.Results.Compared to the unpenalized method, the total MSEs of penalized approaches are much smaller.As the sample size increases, the estimation accuracy of PGEE and PQIF improves significantly.Both PGEE and PQIF can select the true genes such as MAP4, TNN, NRF1, FLT3, and ZFP37 genes with high accuracy.PGEE has larger TP rates than PQIF, but has higher FP rates than PQIF.In summary, the results show that PQIF performs better than PGEE does.PQIF remains relatively optimal even when the working correlation structure is miss-specified.However, PQIF is more sensitive to sample size and sometimes cannot converge when sample size is very small.Conclusion.PQIF has low false selection rate compared to PGEE.The high selection rate of PGEE with true positives might be due to its high false selection rate.Although the difference between the two is not significant under large sample size, a conservative recommendation is to apply PQIF when sample size is small to avoid false selection.In summary, the penalized methods provide a promising and powerful tool for next-generation sequencing data analysis involving longitudinal traits.