Mixtures of Generalized Latent Trait Analyzers for biclustering multivariate data

A novel model-based biclustering approach for multivariate data is introduced exploiting a finite mixture of generalized latent trait models. The proposed model clusters units into distinct subsets, called components. Within each component, subsets of variables, called seg- ments, are identified by specifying the linear predictor in terms of a row-stochastic vector. The continuous latent trait integrated into the model allows us to account for the residual dependence between mul- tivariate outcomes from the same unit. We employ an EM algorithm for maximum likelihood estimation of model parameters, with Gauss- Hermite quadrature utilized to approximate multidimensional integrals where closed-form solutions are not available.