Knowledge-based computational methods for identifying or designing novel, non-homologous antimicrobial peptides.

We describe computational approaches for identifying promising lead candidates for the development of peptide antibiotics, in the context of quantitative structure-activity relationships (QSAR) studies for this type of molecule. A first approach deals with predicting the selectivity properties of generated antimicrobial peptide sequences in terms of measured therapeutic indices (TI) for known antimicrobial peptides (AMPs). Based on a training set of anuran AMPs, the concept of sequence moments was used to construct algorithms that could predict TIs for a second test set of natural AMPs and could also predict the effect of point mutations on TI values. This approach was then used to design peptide antibiotics (adepantins) not homologous to known natural or synthetic AMPs. In a second approach, many novel putative AMPs were identified from DNA sequences in EST databases, using the observation that, as a rule, specific subclasses of highly conserved signal peptides are associated exclusively with AMPs. Both anuran and teleost sequences were used to elucidate this observation and its implications. The predicted therapeutic indices of identified sequences could then be used to identify new types of selective putative AMPs for future experimental verification.