Interpretable Features for the Activity Prediction of Short Antimicrobial Peptides Using Fuzzy Logic
Antimicrobial peptides are a promising class of substances for overcoming multidrug resistant bacteria. In a previous study, high-throughput screening of short peptides for antimicrobial activity against Pseudomonas aeruginosa was performed and the resulting data with 1609 peptides was analyzed using quantitative structure–activity relationship models (QSAR) with excellent prediction power. To avoid non-interpretable black-box behavior of the QSAR models, new features based on fuzzy logic and molecular descriptors were introduced. They were used for comprehensive analysis and visualization. The new features provide good interpretability and are able to differentiate between active and inactive peptides. The statistical relevance of this differentiation was shown using a Wilcoxon rank sum test. The best compromise between activity prediction and interpretability was found for fuzzy terms of the Hopp-Woods scale and the Isoelectric point. A visualization of two of these terms enables an in-depth understanding of regions with active and inactive peptides and the identification of outliers. In addition, we generated rules to explain typical amino acid distributions in active peptides. These rules can be used to increase the probability of finding active peptides in new peptide libraries, which can improve the speed of finding leading substances for drug development against resistant bacteria.
KeywordsAntimicrobial peptides Molecular descriptors QSAR Fuzzy logic Data analysis
We thank R. A. Klady for the critical proofreading of the manuscript.
- Bezdek JC (1981) Pattern recognition with fuzzy objective function algorithms. Plenum Press, New YorkGoogle Scholar
- Cherkasov A, Hilpert K, Jenssen H, Fjell CD, Waldbrook M, Mullaly SC, Volkmer R, Hancock RE (2008) Use of artificial intelligence in the design of small peptide antibiotics effective against a broad spectrum of highly antibiotic-resistant superbugs. ACS Chem Biol (in press)Google Scholar
- Hildebrand P (2006) Zur Strukturvorhersage der Membranproteine. PhD thesis, Humboldt-Universität zu Berlin, Mathematisch-Naturwissenschaftliche Fakultät IGoogle Scholar
- Hilpert K, Hancock REW (2007a) Use of luminescent bacteria for rapid screening and characterization of short cationic antimicrobial peptides synthesized on cellulose by peptide array technology. Nat Protoc 2(7):1652–1660Google Scholar
- Hilpert K, Winkler DFH, Hancock REW (2007b) Peptide arrays on cellulose support: SPOT synthesis—a time and cost efficient method for synthesis of large numbers of peptides in a parallel and addressable fashion. Nat Protoc 2(6):1333–1349Google Scholar
- Mikut R, Burmeister O, Braun S, Reischl M (2008a) The open source MATLAB toolbox Gait-CAD and its application to bioelectric signal processing. In: Proceedings of DGBMT—workshop biosignal analysis, Potsdam, pp 109–111Google Scholar
- Mikut R, Reischl M, Ulrich A, Hilpert K (2008b) Data-based activity analysis and interpretation of small antibacterial peptides. In: Proceedings of 18th workshop computational intelligence, Universitätsverlag Karlsruhe, pp 189–203Google Scholar
- Strom M, Stensen W, Svendsen J, Rekdal O (2001) Increased antibacterial activity of 15-residue murine lactoferricin derivatives. J Pept Res 57(2):127–139Google Scholar
- Zadeh L (1965) Fuzzy sets. Inf Control 8:338–353Google Scholar