BACKGROUND: Current methods in machine learning provide approaches for solving challenging, multiple constraint design problems. While deep learning and related neural networking methods have state-of-the-art performance, their vulnerability in decision making processes leading to irrational outcomes is a major concern for their implementation. With the rising antibiotic resistance, antimicrobial peptides (AMPs) have increasingly gained attention as novel therapeutic agents. This challenging design problem requires peptides which meet the multiple constraints of limiting drug-resistance in bacteria, preventing secondary infections from imbalanced microbial flora, and avoiding immune system suppression. AMPs offer a promising, bioinspired design space to targeting antimicrobial activity, but their versatility also requires the curated selection from a combinatorial sequence space. This space is too large for brute-force methods or currently known rational design approaches outside of machine learning. While there has been progress in using the design space to more effectively target AMP activity, a widely applicable approach has been elusive. The lack of transparency in machine learning has limited the advancement of scientific knowledge of how AMPs are related among each other, and the lack of general applicability for fully rational approaches has limited a broader understanding of the design space. METHODS: Here we combined an evolutionary method with rough set theory, a transparent machine learning approach, for designing antimicrobial peptides (AMPs). Our method achieves the customization of AMPs using supervised learning boundaries. Our system employs in vitro bacterial assays to measure fitness, codon-representation of peptides to gain flexibility of sequence selection in DNA-space with a genetic algorithm and machine learning to further accelerate the process. RESULTS: We use supervised machine learning and a genetic algorithm to find a peptide active against S. epidermidis, a common bacterial strain for implant infections, with an improved aggregation propensity average for an improved ease of synthesis. CONCLUSIONS: Our results demonstrate that AMP design can be customized to maintain activity and simplify production. To our knowledge, this is the first time when codon-based genetic algorithms combined with rough set theory methods is used for computational search on peptide sequences.

BACKGROUND: Antimicrobial peptides attract considerable interest as novel agents to combat infections. Their long-time potency across bacteria, viruses and fungi as part of diverse innate immune systems offers a solution to overcome the rising concerns from antibiotic resistance. With the rapid increase of antimicrobial peptides reported in the databases, peptide selection becomes a challenge. We propose similarity analyses to describe key properties that distinguish between active and non-active peptide sequences building upon the physicochemical properties of antimicrobial peptides. We used an iterative supervised machine learning approach to classify active peptides from inactive peptides with low false discovery rates in a relatively short computational search time. RESULTS: By generating explicit boundaries, our method defines new categories of active and inactive peptides based on their physicochemical properties. Consequently, it describes physicochemical characteristics of similarity among active peptides and the physicochemical boundaries between active and inactive peptides in a single process. To build the similarity boundaries, we used the rough set theory approach; to our knowledge, this is the first time that this approach has been used to classify peptides. The modified rough set theory method limits the number of values describing a boundary to a user-defined limit. Our method is optimized for specificity over selectivity. Noting that false positives increase activity assays while false negatives only increase computational search time, our method provided a low false discovery rate. Published datasets were used to compare our rough set theory method to other published classification methods and based on this comparison, we achieved high selectivity and comparable sensitivity to currently available methods. CONCLUSIONS: We developed rule sets that define physicochemical boundaries which allow us to directly classify the active sequences from inactive peptides. Existing classification methods are either sequence-order insensitive or length-dependent, whereas our method generates the rule sets that combine order-sensitive descriptors with length-independent descriptors. The method provides comparable or improved performance to currently available methods. Discovering the boundaries of physicochemical properties may lead to a new understanding of peptide similarity.

MOTIVATION: The discovery of solid-binding peptide sequences is accelerating along with their practical applications in biotechnology and materials sciences. A better understanding of the relationships between the peptide sequences and their binding affinities or specificities will enable further design of novel peptides with selected properties of interest both in engineering and medicine. RESULTS: A bioinformatics approach was developed to classify peptides selected by in vivo techniques according to their inorganic solid-binding properties. Our approach performs all-against-all comparisons of experimentally selected peptides with short amino acid sequences that were categorized for their binding affinity and scores the alignments using sequence similarity scoring matrices. We generated novel scoring matrices that optimize the similarities within the strong-binding peptide sequences and the differences between the strong- and weak-binding peptide sequences. Using the scoring matrices thus generated, a given peptide is classified based on the sequence similarity to a set of experimentally selected peptides. We demonstrate the new approach by classifying experimentally characterized quartz-binding peptides and computationally designing new sequences with specific affinities. Experimental verifications of binding of these computationally designed peptides confirm our predictions with high accuracy. We further show that our approach is a general one and can be used to design new sequences that bind to a given inorganic solid with predictable and enhanced affinity.