Fig. 1

Integrative model configuration. A and B. Heatmap showing the signal difference of TFs and HMs in accessible and non-accessible regions and the enrichment of TF motifs in 895,695 accessible regions in GM12878 (A) and 883,018 accessible regions in HepG2 (B). The signal was calculated by taking the average of all TFs/HMs binding signals. C. The chromatin accessibility of GM12878 and HepG2 data were acquired from ENCODE database. The DNA-sequence in the accessible region was extracted and embedded into the CNN model to convert the DNA sequence information to the probability of accessibility (Layer 1 model). The resulting probability was next integrated with ChIP-seq signal of HMs and TFs and the TFs motif occurrence in the accessible regions to train a Random Forest model