Skip to main content

Table 3 Performance on HARD CASP10 targets using neighbourhoods (δ)

From: A study and benchmark of DNcon: a method for protein residue-residue contact prediction using deep networks

  

Acc. Top L/10(SE)

Acc. Top L/5 (SE)

Method

d

Long

Medium

Long

Medium

DNcon (222)

1

0.484 (0.067)

0.612 (0.069)

0.438 (0.055)

0.559 (0.057)

IBGteam [DL] (305)

1

0.450 (0.103)

0.546 (0.104)

0.395 (0.086)

0.507 (0.081)

RandomForest (396)

1

0.412 (0.081)

0.505 (0.082)

0.385 (0.067)

0.481 (0.058)

RandomForest (257)

1

0.377 (0.078)

0.505 (0.082)

0.365 (0.066)

0.481 (0.068)

RaptorX-Roll (358)

1

0.349 (0.067)

0.626 (0.084)

0.378 (0.062)

0.591 (0.075)

SVM (81)

1

0.251 (0.050)

0.464 (0.087)

0.243 (0.051)

0.440 (0.073)

DNcon (222)

2

0.619 (0.081)

0.726 (0.058)

0.563 (0.062)

0.674 (0.052)

IBGteam [DL] (305)

2

0.500 (0.112)

0.635 (0.093)

0.427 (0.097)

0.592 (0.074)

RandomForest (396)

2

0.527 (0.079)

0.591 (0.080)

0.486 (0.065)

0.569 (0.058)

RaptorX-Roll (358)

2

0.464 (0.075)

0.692 (0.081)

0.471 (0.068)

0.672 (0.072)

RandomForest (257)

2

0.470 (0.078)

0.591 (0.080)

0.456 (0.066)

0.569 (0.058)

SVM (81)

2

0.409 (0.082)

0.566 (0.086)

0.371 (0.071)

0.537 (0.074)

  1. Accuracies for the top L/10 and L/5 medium and long range contact predictions for 13 hard targets. L is the length of the protein. Estimates for standard error are provided in parenthesises. δ is the size of the neighbourhood.