6533b861fe1ef96bd12c446a

RESEARCH PRODUCT

MOESM7 of 7C: Computational Chromosome Conformation Capture by Correlation of ChIP-seq at CTCF motifs

subject

description

Additional file 7: Figure S4. (A) Prediction performance (auPRC) of 7C when trained and evaluated on different datasets of experimentally measured loops as gold standard. Rao_GM12878 refers to Hi-C loops from [13], Tang2015_GM12878_CTCF, and Tang2015_GM12878_RNAPII to ChIA-PET loops using CTCF or Polymerase II as the target [16]. In Union, all datasets were taken together, and in Intersection, only those CTCF motif pairs that were measured in all datasets were considered positive. (B) Prediction performance (auPRC) of 7C compared to a logistic regression model that uses only the the total coverage signal within +/− 500 bp around the motif center at both loop anchor sites separately. In both models the genomic signal are also included as input features. The performance is shown for six selected TFs in cross-validation on the GM12878 data set. (C) Prediction performance of 7C with six different TFs compared to the method by Oti et al. [40]. The figure shows from top to bottom the accuracy, precision, sensitivity, and specificity of the predictions. (D) Venn-Diagram of CTCF motif pairs overlapping experimentally determined chromatin interactions in GM12878 cells and HeLa cells by Hi-C and ChIA-PET. (E) Prediction performance as auPRC (top) and auROC (bottom) of four different models (colors) on ChIP-seq data for six selected TFs (x-axis). ‘Specific TF’ is the model fitted using the ChIP-seq data indicated on the x-axis, ‘RAD21’ is the model trained on RAD21 ChIP-seq data, ‘Avg. all TF’ is a model averaged across all 124 models of analyzed TFs, and ‘Avg. best 10 TF’ is the averaged model across the 10 best performing models.