Antimicrobial Peptide Scanner vr.1


About the Antimicrobial Peptide Scanner

The Antimicrobial Peptide Scanner is a free service to predict if a protein sequence may be an antimicrobial peptide (AMP) active against Gram-positive and/or Gram-negative bacteria (but not fungi and viruses). A user can submit one or more peptide sequences in FASTA format for prediction. The server first calculates "features" (a mix of physicochemical properties and designed sequence motifs) to numerically represent the amino acids (AA) of a peptide with respect to different functional regions: global (average of all AA), N-terminal (average of the first 10 AA), amphipathic-segment (average of the most amphipathic window, 10-18 AA in length), and C-terminal (average of the last 10 AA). Each of these regions has been shown to play an important role for AMP activity and details on the specific features considered are provided in the dissertation [1]. After a peptide sequence has been converted to a numerical vector, we submit it to two 'Activity Models' to classify it as an AMP or Non-AMP. Models are trained using a data set provided in Xiao et al. (2013)[2] and evaluated with a testing set also from [2] and one from Fernandes et al. (2012)[3] using both random forests (RF)[4] (Mean ACC: 95.5% and MCC: 0.911) and multivariate adaptive regression splines[5] (Mean ACC: 93.1% and MCC: 0.863). Sequences classified as an AMP by at least one of these predictors are further submitted to an additional model (the 'Selectivity Model') to predict if it may have better activity against the Gram-negative bacteria Escherichia coli or the Gram-positive bacteria Staphylococcus aureus (Average 10-fold cross-validation ACC: 83.1% and MCC: 0.662). The Activity Model is trained on AMPs taken from the Antimicrobial Peptide Database[6] and Non-AMPs taken from the UniProt[7]. AMP sequences used to train the Selectivity Model come from the Database of Antimicrobial Activity and Structure of Peptides[8]. Full details are provided in the dissertation.

Sources Cited:

  1. Veltri, Daniel. "Improving Recognition of Antimicrobial Peptides and Target Selectivity through Machine Learning and Genetic Programming." Doctoral dissertation (2015): 1-433. (URI: http://hdl.handle.net/1920/10178).
  2. Xiao, Xuan, et al. "iAMP-2L: a two-level multi-label classifier for identifying antimicrobial peptides and their functional types." Analytical biochemistry 436.2 (2013): 168-177.
  3. Fernandes et al. "Prediction of antimicrobial peptides based on the adaptive neuro-fuzzy inference system application." Peptide Science 98.4 (2012): 280–287.
  4. Breiman, Leo. "Random forests." Machine learning 45.1 (2001): 5-32.
  5. Friedman, Jerome H. "Multivariate adaptive regression splines." The annals of statistics (1991): 1-67.
  6. Wang, Zhe, and Guangshun Wang. "APD: the antimicrobial peptide database." Nucleic acids research 32.suppl 1 (2004): D590-D592.
  7. Wu, Cathy H., et al. "The Universal Protein Resource (UniProt): an expanding universe of protein information." Nucleic acids research 34.suppl 1 (2006): D187-D191.
  8. Gogoladze, Giorgi, et al. "DBAASP: database of antimicrobial activity and structure of peptides." FEMS microbiology letters 357.1 (2014): 63-68.