(A) Strategy for using GBR to transfer information between cell types. (B,C) Performance of predicting the locations of GM12878 (B) EP300 and (C) CTCF ChIP-seq peaks with and without using GBR to integrate information from K562. Each plot shows the fraction of elements detected as a function of the number of bases predicted (Methods). Results are shown on the test set (10 Mbp). Model training and label ordering were performed on the training set (10 Mbp). The x-axis is plotted up to 1 kb times the total number of elements. These plots can be interpreted, for example, in the context of an enhancer validation experiment, in which case it shows how many sequences would need to be tested in order to discover a certain number of enhancers.
