Table 4 The prediction performances across cell lines.

From: Predicting CTCF cell type active binding sites in human genome

Signal type

Training cell line

Testing cell line

CNN

XGBoost

DNase I only

HSMMtube

HFF

0.973 ± 0.001

0.975 ± 0.001

HFF-Myc

0.922 ± 0.004

0.924 ± 0.002

Caco2

0.770 ± 0.001

0.729 ± 0.006

RAD21 only

K562

GM12878

0.965 ± 0.001

0.966 ± 0.001

H1-hESC

0.940 ± 0.001

0.942 ± 0.001

SK-N-SH_RA

0.956 ± 0.008

0.954 ± 0.002

SMC3 only

HeLa-S3

GM12878

0.925 ± 0.007

0.931 ± 0.001

HepG2

0.904 ± 0.002

0.902 ± 0.001

K562

0.947 ± 0.003

0.947 ± 0.003

2 TFs

K562

GM12878

0.966 ± 0.000

0.968 ± 0.001

HeLa-S3

0.963 ± 0.001

0.960 ± 0.003

HepG2

0.946 ± 0.001

0.947 ± 0.002

DNase I + 2 TFs

K562

K562

0.970 ± 0.002

0.973 ± 0.001

HeLa-S3

0.967 ± 0.001

0.966 ± 0.000

HepG2

0.945 ± 0.003

0.949 ± 0.002

  1. The “2 TFs” refers to RAD21 and SMC3.The AUPRC values are the average of three cross-validation folds with standard errors given in parentheses. The better predictions are marked in bold.
Back to article page