Skip to main content

Table 2 The mean values and the standard deviations of F1 scores for EP2vec and other three baseline methods in 10-fold cross-validation experiments. For FANTOM dataset, we do not evaluate TargetFinder due to lack of experimental features, and we do not evaluate SPEID since it is extremely time-consuming to run 10-fold cross validation of SPEID on so many samples

From: Prediction of enhancer-promoter interactions via natural language processing

Dataset EP2vec TargetFinder gkmSVM SPEID
K562 0.882 (0.019) 0.881 (0.014) 0.821 (0.018) 0.846 (0.024)
IMR90 0.872 (0.020) 0.863 (0.017) 0.749 (0.026) 0.825 (0.032)
GM12878 0.867 (0.014) 0.844 (0.010) 0.779 (0.015) 0.809 (0.018)
HUVEC 0.875 (0.024) 0.878 (0.022) 0.731 (0.028) 0.809 (0.023)
HeLa-S3 0.920 (0.013) 0.913 (0.014) 0.822 (0.021) 0.888 (0.023)
NHEK 0.933 (0.015) 0.922 (0.018) 0.800 (0.024) 0.900 (0.019)
FANTOM 0.841(0.004) / 0.803(0.017) /