Skip to main content

Advertisement

Table 1 Comparison of tree-aware methods and tree-ignorant methods.

From: Detecting coevolution without phylogenetic trees? Tree-ignorant metrics of coevolution perform as well as tree-aware metrics

(A) Tetrapod Myoglobin
  Tree-aware Tree-ignorant   
  pα p > α pα p > α χ 2 pvalue
(i,i+3) w best SCA (0.9) 3 125 32 233 10.08 1.50 × 10-3
(i,i+3) w worst SCA (0.5) 3 125 23 242 5.61 1.79 × 10-2
(i,i+4) w best SCA (0.8) 14 114 128 137 52.22 4.97 × 10-13
(i,i+4) w worst SCA (0.4) 14 114 100 165 30.10 4.10 × 10-8
(B) Randomized Tetrapod Myoglobin
  Tree-aware Tree-ignorant   
  pα p > α pα p > α χ 2 pvalue
(i,i+3) w best SCA (0.9) 1 127 4 261 0.36 5.46 × 10-1
(i,i+3) w worst SCA (0.5) 1 127 9 256 2.38 1.23 × 10-1
(i,i+4) w best SCA (0.8) 0 128 1 264 0.48 4.87 × 10-1
(i,i+4) w worst SCA (0.4) 0 128 1 264 0.48 4.87 × 10-1
(C) Chordate Myosin
  Tree-aware Tree-ignorant   
  pα p > α pα p > α χ 2 pvalue
(i,i+3) w best SCA (0.4) 54 14 221 44 0.60 4.40 × 10-1
(i,i+3) w worst SCA (0.9) 54 14 198 67 0.65 4.21 × 10-1
(i,i+4) w best SCA (0.6) 58 10 209 56 1.41 2.36 × 10-1
(i,i+4) w worst SCA (0.9) 58 10 177 88 8.92 2.82 × 10-3
(D) Randomized Chordate Myosin
  Tree-aware Tree-ignorant   
  pα p > α pα p > α χ 2 pvalue
(i,i+3) w best SCA (0.4) 7 61 56 209 4.14 4.18 × 10-2
(i,i+3) w worst SCA (0.9) 7 61 55 210 3.91 4.81 × 10-2
(i,i+4) w best SCA (0.6) 0 68 0 265 n/a n/a
(i,i+4) w worst SCA (0.9) 0 68 0 265 n/a n/a
  1. χ2 goodness-of-fit tests comparing all applicable tree-aware and all tree-ignorant methods for detecting alpha helix periodicity at separations of (i, i + 3) and (i, i + 4), using all reduced-state amino acid alphabets, where applicable. Some of the methods (LnLCorr07 and GCTMPCA) were not run on myosin (see text), causing different numbers of runs on the myosin versus the myoglobin data sets. Additionally, the reduced-state alphabets were not applicable to all tree-aware methods, causing the sum of the p-value counts to differ for the tree-aware versus tree-ignorant methods. Bolded rows highlight statistically significant χ2 results, indicating a difference between the tree-aware and tree-ignorant populations. Bold counts in these rows highlight which class of methods achieved the higher ratio of significant to insignificant scores. Counts are calculated including both the empirically determined best and worst SCA cutoffs to illustrate the effect of this free parameter on method performance, and with the empirically validated optimal ε = 0.70 value for GCTMPCA. The data suggest that tree-ignorant methods perform better on the myoglobin alignment (A), and there is no statistically significant difference in the classes on the myosin rod domain alignment except when using the empirically determined worst SCA cutoff choice, when the tree-aware methods perform better (C). The myoglobin (B) and myosin (D) negative controls suggest that there is no significant difference in the Type 1 error incurred by the classes of methods. α = 0.01.