Skip to main content

Table 3 Coverage and accuracy of the various functional site prediction schemes across all 34 functional sites.

From: Assessing the ability of sequence-based methods to provide functional insight within membrane integral proteins: a case study analyzing the neurotransmitter/Na+ symporter family

Method

Coverage1

Absolute Accuracy2

Relative Accuracy3

Overall Performance4

(a.) Unique methods

Phylogenetic motif

0.62 (21)

0.24 (89)

0.55 (38)

0.40

False positive expectation

0.53 (18)

0.35 (51)

0.90 (20)

0.43

Site conservation

0.59 (20)

0.35 (58)

---

0.45

Rate4Site

0.50 (17)

0.37 (46)

---

0.43

ET

0.44 (15)

0.27 (56)

---

0.34

SDPpred

0.12 (4)

0.27 (15)

---

0.19

(b.) Hybrid methods

Union (PM + FPE)

0.77 (26)

0.22 (116)

0.45 (58)

0.42

Intersect_25

0.71 (24)

0.29 (84)

---

0.46

Intersect_35

0.56 (19)

0.44 (43)

---

0.50

Intersect_45

0.32 (11)

0.50 (22)

---

0.40

Intersect_55

0.18 (6)

0.67 (9)

---

0.37

  1. 1 Coverage describes the fraction of correctly predicted functional sites from Table 2; the raw count is provided in parentheses. 2 Absolute accuracy is defined as the ratio of correct predictions to total alignment positions predicted (provided in parentheses). 3 Because the PM, FPE, and Union methods are based on alignment windows (versus individual sites within the alignment), relative accuracy is also calculated. Relative accuracy is similar to absolute accuracy, but describes the ratio of correct predictions to the total number of alignment windows (also provided in parentheses). 4 Overall performance is calculated as the Cartesian distance between (coverage, accuracy) of each method and that of a perfect method (coverage = 1.00, accuracy = 1.00). The distance is normalized such that a method with 0.00 coverage and accuracy would have a value of unity, which is then subtracted from one in order to stay consistent with the coverage and accuracy scales. 5 The Intersect predictions describe a hybrid approach composed of the unique prediction strategies. Whenever the number of predictions for a particular site are greater than the intersect value, that site is put forth as a prediction (e.g., Intersect_2 represents the intersection of any two unique methods). Due to poor performance, the SDPpred method is not considered within the Intersect results.