Antimicrobial peptides (AMPs) are anti-infective agents that may represent a novel and untapped class of biotherapeutics. In the lab of bioinformatics of IBCEB, the Database of Antimicrobial Activity and Structure of Peptides (DBAASPv.2 - accessible at http://dbaasp.org) has been developed. DBAASP provides information and analytical resources to the scientific community in order to develop antimicrobial compounds with high therapeutic index.
Quantitative structure-activity relationship (QSAR) studies for the development of predictive model for AMPs are generally based on discriminative analysis and especially machine learning methods. These methods, as a positive training set, have used a full set of antimicrobial peptide sequences, without taking into account variation in mechanisms of action, structure, mode of interaction with membrane and other differences. Contrary to available approaches, we think that strategy of prediction should be based on the fact that there are at least four kinds of AMPs for which four independent algorithms of prediction have to be developed in order to get high efficacy. We can distinguish linear cationic antimicrobial peptides (LCAP), cationic peptides stabilizing structure by intra-chain covalent bonds, proline and arginine-rich peptides, and anionic antimicrobial peptides.
Simple predictive model which can discriminate AMPs from non-AMPs has been developed for LCAP. Sequences have been taken from DBAASP. As descriptors the sequence-based physical-chemical characteristics responsible for capability of the peptide to interact with an anionic membrane were considered. On the basis of these characteristics, a new simple algorithm of prediction is developed and in silico evaluation of the efficacy of characteristics is done. The algorithm was based on the clusterization of AMPs by their physicochemical properties. The results show that descriptors relied mainly on hydrophobic and hydrophilic features allow us to predict AMPs with the high accuracy.
The developed predictive model was used to create new amino acid sequences. On the basis of designed sequences the peptides have been synthesized. Antimicrobial potency of the new peptides has been evaluated by in vitro testing of peptides activity against different pathogenic bacteria (including drug resistant strains). In vitro estimation shows that the accuracy of th developed predictive model is higher than 90%.
