<p>We consider the problem of relating two high-dimensional datasets with the goal of identifying subsets of outcomes with the same regression function over a subset of covariates, allowing for nonlinear relationships. This is accomplished by specifying a mixture of Gaussian process regression models and performing variable selection using a stochastic partitioning method. The proposed method provides a flexible approach that simultaneously uncovers cluster structures and identifies linear and nonlinear relationships across two large datasets. We evaluate its performance on simulated data in terms of identification of cluster structures, variable selection, and prediction.</p>