PREDICTD PaRallel Epigenomics Data Imputation with Cloud-based Tensor Decomposition

This dataset contains training data and the trained model parameters for PREDICTD, a 3D tensor decomposition-based model of the NIH Epigenomics Roadmap project data. The NIH Epigenomics Roadmap consortium was focused on mapping the genomic locations of many different epigenomic marks (for example histone modifications) in many different cell types. However, due to the large number of experiments that it would take to exhaustively produce such maps for all epigenomic marks and all cell types, many possible experiments were missing. The PREDICTD model used 3D tensor decomposition and reconstruction to predict the results of these missing experiments. The training data and model parameters are preserved to allow others to more easily replicate or follow up on the results in the PREDICTD publication (Durham, et al. 2018.).