Skip to main content

Table 3 Performance differences among algorithms for the homopolymers of length 64 and 32 in long runs.

From: An adaptive bin framework search method for a beta-sheet protein homopolymer model

Method

Temperature set

Length

E avg ± sd

E med

E q 75

E q 25

E min

p – value

our MC

1.25

32

-158.7 (± 1.9)

-159

-159

-158

-161

0.0271*

our REMC

linear 1.25 to 2.75

32

-159.6 (± 1.3)

-160

-161

-158

-161

0.5471

our PHAT

linear 1.3 to 2.75

32

-158.9 (± 1.4)

-159

-159

-158

-161

0.0638

our BINMC

T MC = 1.25, T bin = 6.521

32

-160.1 (± 0.9)

-161

-161

-159

-161

 

our MC

1.25

64

-372.2 (± 2.3)

-372

-373

-371

-377

0.0005*

our REMC

linear 1.25 to 2.75

64

-376.1 (± 3.5)

-376

-378

-373

-382

0.0521

our PHAT

linear 1.3 to 2.75

64

-374.1 (± 3.8)

-374

-377

-371

-383

0.0120*

our BINMC

T MC = 1.25, T bin = 6.521

64

-379.5 (± 3.3)

-381

-382

-376

-389

 
  1. Comparison of the energy levels reached for the homopolymers of length 64 and 32 by our implementations of MC, REMC (with the linear set of temperatures), PHAT, and the new BINMC algorithm in 10 independent runs of 10 CPU hours each on our 2.4 GHz reference machine. The p-values reported in the last column were determined using the Mann-Whitney U test to test the null hypothesis that the mean energies reached by the respective algorithm and BINMC (within the same CPU cut-off time) are identical [4]; p-values marked with an asterisk (*) correspond to cases in which the null hypothesis is rejected at a standard significance level of 0.05, and therefore indicate statistically significant performance differences.