Skip to main content

Table 2 Per-protein predictions: localization and membrane/globular

From: Modeling aspects of the language of life through transfer-learning protein sequences

Method

Localization

Membrane/globular

Q10 (%)

Gorodkin (MCC)

Q2

MCC

LocTree2a,b

61

0.53

  

MultiLoc2a,b

56

0.49

  

CELLOa

55

0.45

  

WoLF PSORTa

57

0.48

  

YLoca

61

0.53

  

SherLoc2a,b

58

0.51

  

iLoc-Euka,b

68

0.64

  

DeepLoca,b

78

0.73

92.3

0.844

DeepSeqVec-Loc

68 ± 1

0.61 ± 0.01

86.8 ± 1.0

0.725 ± 0.021

DeepProtVec-Loc

42 ± 1

0.19 ± 0.01

77.6 ± 1.3

0.531 ± 0.026

  1. Performance for per-protein prediction of subcellular localization and classifying proteins into membrane-bound and water-soluble. Results marked by a taken from DeepLoc [47]; the authors provided no standard errors. The results reported for SeqVec and ProtVec were based on single protein sequences, i.e. methods NOT using evolutionary information (neither during training nor testing). All methods using evolutionary information are marked by b; best in each set marked by bold numbers