Table 3 Effect of sequence redundancy on algorithm cross-validation performance. HMM models were constructed using the 186 bootstrap sequences used to train HMM266, then tested for accuracy and coverage against non-overlapping positive and negative test sets. "Max. sequence similarity" refers to the maximum number of amino acid position matches allowed for the sequences in a given row, either within or between test and training sets. Jack-knife (leave-one-out) testing for each row was performed against the training set described in that row.
From: Predicting N-terminal myristoylation sites in plant proteins
Model Name | Max. sequence similarity | Number train seqs. | Number positive test seqs. | Number negative test seqs. | Accuracy (TP+TN)/TOTAL | Coverage TP/(TP+FN) | Jack-knife Detection |
|---|---|---|---|---|---|---|---|
HMM186B | 24/25 residues (96%) | 186 | 80 | 185 | 96.6% | 96.3% | 98.4% |
HMM162B | 20/25 residues (80%) | 162 | 53 | 128 | 96.1% | 92.5% | 96.9% |
HMM151B | 15/25 residues (60%) | 151 | 42 | 102 | 98.6% | 95.2% | 96.7% |
HMM127B | 10/25 residues (40%) | 127 | 25 | 94 | 97.5% | 96.0% | 96.1% |