Fig. 4
SHAP summary plot for the trained convolutional neural network (CNN) model (see Supplementary Fig. 1 for architecture details). The model’s input consists of three CRISPRspec-derived energy terms, six experimental epigenetic scores (bolded), and 13 computed nucleosome organization-related scores. The three CRISPRspec-derived energy terms are \(E_{\text {RNA-DNA}}\), \(E_{\text {RNA-DNA}}^{\text {corr}}\) and EgRNAfold. The six experimental epigenetic scores are CTCF, DNase I, DRIP, H3K4me3, MNase and RRBS. The 13 computed nucleosome organization-related scores are GC147 [51], W/S scheme, YR scheme [52, 53], Strong-Weak BDM, Nucleotide BDM [48, 49], NuPoP (Occupancy), NuPoP (Affinity), NuPoP (Viterbi) [54], nuCpos (Occupancy), nuCpos (Affinity), nuCpos (Viterbi) [55], VanDerHeijden [56] and LeNup (H3Q85C) [57]. The base pair-resolved SHAP contributions for each data point are summed for each computed nucleosome organization-related score