Skip to main content

Advertisement

Table 1 Top-1, top-3, top-5 accuracy in identifying matching motif for TF binding (out of 422 experiments) with similarity score (p-value) smaller than 0.5 and 0.05, and the paired/unpaired rank tests of the proposed ranks of best matching filters between our method and the input-dependent methods

From: Visualizing complex feature interactions and feature sharing in genomic deep neural networks

  Top 1 Top 3 Top 5 Ranksum Wilcoxon
TOMTOM P-value 0.5 0.05 0.5 0.05 0.5 0.05 p-values p-values
DeepResolve (ours) 412 407 421 418 422 421 N/A N/A
DeepLIFT 418 418 420 420 421 421 0.784 0.412
Saliency*activation 404 393 419 419 420 420 7.478×10−7 1.049×10−9
Saliency 388 377 417 416 420 420 4.63×10−7 4.63×10−15
enhanced DeepMotif 217 89 310 123 343 147 N/A N/A