Table 4.
Summary of Results for Prediction Tools
| Ng and Henikoff | Chasman and Adams (2001) | Sunyaev et al. (2001) | ||||
|---|---|---|---|---|---|---|
| Number of substitutions predicted on (number of proteins predicted on) | % predicted to have effect | Number of substitutions predicted on (number of proteins predicted on) | % predicted to have effect | Number of substitutions predicted on (number of proteins predicted on) | % predicted to have effect | |
| Predictions on substitutions that affect function | ||||||
| Substitutions annotated to be involved in disease from databases | 5218 (462) | 69% (3626/5218)a | — | — | 1551 (N/A) | 69% (1071/1551)b |
| Substitutions in LacI that affect function | 1764 (1) | 63% (1113/1764)cd | 733 (1) | 75% (551/733)e | — | — |
| Prediction on substitutions that do not affect function | ||||||
| Substitutions in LacI that do not give an altered phenotype | 2240 (1) | 28% (627/2240)c | 1131 (1) | 31% (345/1131)f | — | — |
| Substitutions between species | — | — | — | — | 440 (N/A) | 9% (41/440)b |
| Nonsynonymous polymorphisms | ||||||
| WI-nsSNPs | 115 (53) | 19% (22/115)a | 53 (N/A) | 15% (8/53)g | — | — |
| Validated polymorphisms from databases | — | — | — | — | 245 (N/A) | 32% (79/245)b |
| dbSNP | 3084 (1789) | 25% (757/3084)a | — | — | — | — |
From Table 1.
From Table 1 of Sunyaev et al. (2001).
The SIFT prediction accuracies for LacI were originally published in Ng and Henikoff (2001). The numbers shown here are slightly different from the original reference because they are based on SIFT version 2.
Because paralogues were included in the LacI alignment used for SIFT prediction, many of the wrongly predicted residues recognition residues not conserved among paralogues (Ng and Henikoff 2001). This tended to lower prediction accuracy.
(prediction = effect & actual = effect)/actual = effect. Values are taken from the leftmost column in Table 5 from Chasman and Adams (2001), the test set with the highest total prediction accuracy which used the structure of LacI for predictions. When the structure of the Pur repressor, a LacI paralogue, was used for prediction on the LacI dataset, only 19% (69/368) [rightmost column of Table 5 from Chasman and Adams (2001)] of the substitutions that have an effect were correctly predicted as damaging.
(prediction = effect & actual = no effect)/actual = no effect. Values are taken from the leftmost column in Table 5 from Chasman and Adams (2001), the dataset with the highest total prediction accuracy.
From Table 7b of Chasman and Adams (2001).