Skip to main content

Table 3 Performances of our kernels (LLL)

From: Walk-weighted subsequence kernels for protein-protein interaction extraction

Kernel Performance
  PRE REC F-M
walk kernel[13] 72.5 83.3 77.5
dependency kernel[13] 58.6 62.9 60.7
v-walk kernel 51.4 66.6 58.0
e-walk kernel 71.9 75.9 73.8
extended dependency kernel 62.6 77.7 69.4
spectrum kernel 64.6 77.7 70.5
fixed-length subsequence kernel 72.5 83.3 77.5
gap-weighted subsequence kernel 68.4 72.2 70.2
walk-weighted subsequence kernel 79.3 85.1 82.1
  1. The performances of kernels we suggested are shown. All kernels were trained and tested over clean data set (LLL training and test) and accessed by LLL external evaluation server.