CAMP |
SVMRandom Forests |
91.50%93.2% |
64 (after recursive feature elimination on initial set of 257) physicochemical properties (composition), dipeptide & tripeptide frequencies, distribution & transition of some features along sequences |
2578 experimentally validated CAMP peptides |
4011 random proteins from UniProt, synthesized sequences using random numbers, experimentally verified non-antimicrobial peptides (25) |
30% of positive & negative sets |
Fjell et al |
Quantitative structure-activity relationships (QSAR) |
80.00% |
44 QSAR descriptors |
1433 synthesized peptides, 9 amino-acids long(antibacterial acitivity measured experimentally) |
∼100000 synthesized peptides |
Torrent et al |
ANNSVM |
90%75% |
8 physicochemical & structural properties (50 hidden neurons) |
1157 CAMP antimicrobial peptides |
991 randomly selected UniProt protein fragments |
290 antimicrobial peptides from CAMEL and RANDOM databases |
Porto et al |
SVM |
83.02% |
4 physicochemical properties |
199 peptides from APD |
199 proteins predicted to be transmembrane |
106 sequences from positive & negative training sets |
Wang et al |
BLASTP & Nearest-Neighbour Algorithm (NNA) |
93.31% |
25 composition & pseudo-amino acid composition features from initial set of 270 (for NNA) |
870 peptides from CAMP (including some predicted) |
8661 protein fragments randomly selected from UniProt |
1136 predicted peptides from CAMP |