Gaussian processes are kernel-based statistical methods that accommodate non-linear data structures. I will present on a two-part Gaussian process model for semicontinuous data using Bayesian methods for estimation. The two parts of the model include a probit model part for the probability of nonzero response and a conditional model for the mean response given that it is nonzero. This two-part model also performs Bayesian variable selection using spike and slab priors to determine key predictors associated with the response in high dimensional data. With kernel-based methods, there is often high computational complexity in inverting high dimensional kernel matrices and performing prediction, so to circumvent this problem I have applied the Nyström method for dimension reduction. A Markov Chain Monte Carlo algorithm is used for estimation of parameters. In my presentation, I will introduce the model and present results from simulated data.