Figure 7: ProCapNet performance and predictions are highly consistent across cell types.

(A) ProCapNet performance when trained and tested in all available cell types. (B) Comparison of measured PRO-cap signal (top right triangle) and ProCapNet counts predictions (bottom left triangle) across cell type pairs, plus same-cell ProCapNet predictions vs. measured PRO-cap (diagonal) with peaks active in other cell types included. (C) Goodness-of-fit of linear models of K562 PRO-cap data, using combinations of ProCapNet predictions and various experimental measurements of chromatin state. (D) Examples of how histone modification or DNase-seq signal is distributed across cell-union PRO-cap peaks, with respect to K562 PRO-cap and ProCapNet predictions. (E) Goodness-of-fit of linear models of PRO-cap data in one cell type, using combinations of ProCapNet predictions from another cell type and experimentally measured chromatin state signals.