Fig. 2

Correlation of individual expert scores and comparison of plasmid-based and CRISPR-based experts. A Heatmap of mean genomewide Pearson correlations between expert model tracks clustered with hierarchical clustering, averaged over cell types. B Box plots of mean normalized score differences across cell types between experts trained on nine plasmid-based and two CRISPR-based functional characterization datasets in different ChromHMM chromatin states [13, 39]. The boxes represent quartiles and whiskers indicate maximum and minimum score differences between plasmid-based and CRISPR-based experts. Individual mean scores averaged across cell types, for each expert separately, is shown in Additional file 1: Fig. S5. The corresponding box plot distributions of means across cell types for each expert and each state is shown in Additional file 1: Fig. S6. Box colors correspond to the predefined ChromHMM imputed 25-state model colors. C Scatter plot of mean normalized expert scores for plasmid-based vs. CRISPR-based functional characterization datasets per chromatin state, averaged over cell types. Error bars indicate standard deviation of score means across cell types. Chromatin state abbreviations: active promoters (1_TssA, 2_PromU, 3_PromD1, 4_PromD2), transcribed regions (5_Tx5’, 6_Tx, 7_Tx3’, 8_TxWk), transcribed and regulatory regions (9_TxReg, 10_TxEnh5’, 11_TxEnh3’, 12_TxEnhW), active enhancers (13_EnhA1, 14_EnhA2, 15_EnhAF), weak enhancers (16_EnhW1, 17_EnhW2, 18_EnhAc), primary DNase (19_DNase), ZNF genes and repeats (20_ZNF/Rpts), heterochromatin (21_Het), poised/bivalent promoters (22_PromP, 23_PromBiv), repressed polycomb (24_ReprPC), and quiescent/low (25_Quies)