High-dimensional data, with many more covariates than observations, such as genomic data for example, are now commonly analyzed. In this context, it is often desirable to be able to focus on the few most relevant covariates through a variable selection procedure. High dimensional variable selection is widely documented in standard regression models, but there are still few tools to address it in the context of nonlinear mixed effects models. In this work, we approach variable selection from a Bayesian perspective and propose a selection procedure combining the use of \textit{spike-and-slab} priors and the SAEM algorithm. Similarly to Lasso regression, the set of relevant covariates is selected by exploring a grid of values for the penalization parameter. The proposed approach is much faster than a classical MCMC algorithm and shows very good selection performances on simulated data.