Sequence-based Characterization of Antimicrobial Activity

With growing antimicrobial resistance, it is becoming imperative to seek new novel antibiotic treatments to combat microbial infections. One interesting avenue for novel antibiotics is to build on natural templates already found in higher-order organisms. Antimicrobial Peptides (AMPs) are small innate peptides that have been found to attack bacteria and fungi through several mechanisms. However, our current understanding of what confers to these peptides their antimicrobial activity remains poor. Some large rules or features have been proposed by experimentation, and these have been tested in the context of classification in supervised machine learning to separate AMPs from non-AMPs.

Our lab focuses on finding such features and evaluating them in the context of supervised classification. Several papers in 2013 focused on evaluation of simple physicochemical features with several classifiers. More recently, in 2014 and 2015 we proposed algorithms based on genetic programming to construct complex sequence-based features that encode distal information about a peptide sequence. We demonstrated that this not only confers higher performance in the baseline AMP recognition problem but also allows to recognize specific classes of AMPs potent against Gram-positive, Gram-negative, or both classes of bacteria.

The following is an actual conference presentation of our 2013 work.

Highlight of work led by D. Veltri^g and A. Shehu^* at ACM Bioinf and Comp Biol (BCB) 2013, IEEE Intl Conf on Comp Adv in Bio and Med Sciences (ICCABS) 2013, and Intl Conf on Bioinf and Comp Biol (BiCoB) 2013.