Fig. 4

DSB models improve predictions for non-model cell types. Models trained using a dataset from one cell type were used to generate predictions for a different cell type, given the matched features. The dark blue lines mark the Pearson’s correlation between the two cell types. The cell type used to train the model is indicated by the color of the bar, and the cell type on which the model is being applied is shown on the x-axis. In all cases, the random forest model greatly improves the predictions from a naïve inference, with a 1.3–1.8 fold improvement in correlation