Skip to main content

Table 6 Results of increase the number of negative samples on SHS27K and SHS148k

From: Long-distance dependency combined multi-hop graph neural networks for protein–protein interactions prediction

Positive:negative sample rate

SHS27k

SHS148k

 

Random

BFS

DFS

Random

BFS

DFS

1:1

89.34 \(\pm \; 0.44\)

74.56 \(\pm \; 3.03\)

78.20 \(\pm \;2.69\)

92.38 \(\pm \; 0.08\)

73.98 \(\pm \; 5.51\)

83.79 \(\pm \; 0.95\)

1:3

\(86.25 \pm 0.21\)

\(71.43 \pm 1.56\)

\(76.14 \pm 1.66\)

\(89.31 \pm 0.98\)

\(71.72 \pm 2.34\)

\(78.97 \pm 0.75\)

1:5

\(82.74 \pm 0.55\)

\(69.29 \pm 2.71\)

\(72.94 \pm 2.81\)

\(82.10 \pm 2.39\)

\(68.21 \pm 4.84\)

\(72.64 \pm 0.93\)

1:10

\(77.94 \pm 0.75\)

\(65.62 \pm 2.47\)

\(69.84 \pm 1.28\)

\(75.55 \pm 0.52\)

\(61.36 \pm 5.41\)

\(66.92 \pm 1.85\)

  1. Bold number represents the best value for that partition scheme