Skip to main content

Table 4 Human interactome comparison: running time and peak memory

From: SPRINT: ultrafast protein-protein interaction prediction of the entire human interactome

Dataset Program Time (s) Memory (GB)
   Preprocess Train Predict Preprocess Train Predict
Biogrid Martin 32,400 > 1,209,600 2.5 6.1
  PIPE2 312,120 N/A 1,150,675,200 2.1 N/A 18.9
  Ding 37,708 3.3 > 256
  SPRINT 105,480 N/A 6,120 11.2 N/A 3.0
HPRD Release 9 Martin 32,400 584,640 107,222,400 2.5 3.2 1.5
  PIPE2 312,120 N/A 435,628,800 2.1 N/A 18.9
  Ding 37,708 236,551 374,360 3.3 79.5 79.5
  SPRINT 105,480 N/A 1,257 11.2 N/A 3.0
Innate experim. validated Martin 32,400 > 1,209,600 2.5 5.7
  PIPE2 312,120 N/A 872,294,400 2.1 N/A 18.9
  Ding 37,708 3.3 > 256
  SPRINT 105,480 N/A 3,600 11.2 N/A 3.0
Innate manually curated Martin 32,400 26,280 30,888,000 2.5 1.9 1.5
  PIPE2 312,120 N/A 230,342,400 2.1 N/A 18.9
  Ding 37,708 55,532 285,323 3.3 25.4 25.4
  SPRINT 105,480 N/A 930 11.2 N/A 3.0
IntAct Martin 32,400 > 1,209,600 2.5 3.5
  PIPE2 312,120 N/A 616,464,000 2.1 N/A 18.9
  Ding 37,708 > 1,209,600 3.3 220
  SPRINT 105,480 N/A 2,672 11.2 N/A 3.0
MINT Martin 32,400 101,160 52,557,120 2.5 2.3 1.5
  PIPE2 312,120 N/A 372,902,400 2.1 N/A 18.9
  Ding 37,708 120,720 331,865 3.3 41.1 41.1
  SPRINT 105,480 N/A 952 11.2 N/A 3.0
  1. The predicting time for Martin’s and PIPE2 was estimated by running it for 100 h and then estimating the total time according to the number of pairs left to predict. Note that PIPE2 and SPRINT do not require training as they are not using machine learning. For the entries marked with a dash, the program ran out of (256 GB) memory or ran for more than 14 days. Times marked with a dagger are estimated