A multivariate stochastic model for describing the dynamics of complex non-numerical ensembles, such as observed in Human Immunodeficiency Virus (HIV) genome, is developed. This model is based on principle component analyses for numberized variables. The model coefficients are presented in the terms of deterministic trends with correlated lags. The results indicate that we may use this model in short-term forecast of HIV evolution, for evaluation of HIV drug resistance and for testing and validation of diagnostic expert rules. The model also reproduces the specific shape of the bi-modal distribution for the mutations number.