Overview of the workflow. (a) We first prepared the multiple genomic datasets and the geoclimate dataset. (b) HCCA is then applied to combine the information for finding a joint feature representation by coprojection of the datasets in a hierarchy learned by the analysis of the conditional numbers. (c) Optionally, the PPI network can also be integrated with HCCA to learn a more joint feature representation with better functional coherence. (d) The coprojection of the integrated data provides the joint feature representation of the integrated datasets. (e) The joint feature representation is then used by support vector regressor for phenotype prediction. (f) The joint feature representation is also analyzed for the association between genomic features and the geoclimatic features.