Comparative analysis of machine learning algorithms on the microbial strain-specific AMP prediction.

The evolution of drug-resistant pathogenic microbial species is a major global health concern. Naturally occurring, antimicrobial peptides (AMPs) are considered promising candidates to address antibiotic resistance problems. A variety of computational methods have been developed to accurately predict AMPs. The majority of such methods are not microbial strain specific (MSS): they can predict whether a given peptide is active against some microbe, but cannot accurately calculate whether such peptide would be active against a particular MS. Due to insufficient data on most MS, only a few MSS predictive models have been developed so far. To overcome this problem, we developed a novel approach that allows to improve MSS predictive models (MSSPM), based on properties, computed for AMP sequences and characteristics of genomes, computed for target MS. New models can perform predictions of AMPs for MS that do not have data on peptides tested on them. We tested various types of feature engineering as well as different machine learning (ML) algorithms to compare the predictive abilities of resulting models. Among the ML algorithms, Random Forest and AdaBoost performed best. By using genome characteristics as additional features, the performance for all models increased relative to models relying on AMP sequence-based properties only. Our novel MSS AMP predictor is freely accessible as part of DBAASP database resource at http://dbaasp.org/prediction/genome.

Physicochemical Features and Peculiarities of Interaction of AMP with the Membrane.

Antimicrobial peptides (AMPs) are anti-infectives that have the potential to be used as a novel and untapped class of biotherapeutics. Modes of action of antimicrobial peptides include interaction with the cell envelope (cell wall, outer- and inner-membrane). A comprehensive understanding of the peculiarities of interaction of antimicrobial peptides with the cell envelope is necessary to perform a rational design of new biotherapeutics, against which working out resistance is hard for microbes. In order to enable de novo design with low cost and high throughput, in silico predictive models have to be invoked. To develop an efficient predictive model, a comprehensive understanding of the sequence-to-function relationship is required. This knowledge will allow us to encode amino acid sequences expressively and to adequately choose the accurate AMP classifier. A shared protective layer of microbial cells is the inner, plasmatic membrane. The interaction of AMP with a biological membrane (native and/or artificial) has been comprehensively studied. We provide a review of mechanisms and results of interactions of AMP with the cell membrane, relying on the survey of physicochemical, aggregative, and structural features of AMPs. The potency and mechanism of AMP action are presented in terms of amino acid compositions and distributions of the polar and apolar residues along the chain, that is, in terms of the physicochemical features of peptides such as hydrophobicity, hydrophilicity, and amphiphilicity. The survey of current data highlights topics that should be taken into account to come up with a comprehensive explanation of the mechanisms of action of AMP and to uncover the physicochemical faces of peptides, essential to perform their function. Many different approaches have been used to classify AMPs, including machine learning. The survey of knowledge on sequences, structures, and modes of actions of AMP allows concluding that only possessing comprehensive information on physicochemical features of AMPs enables us to develop accurate classifiers and create effective methods of prediction. Consequently, this knowledge is necessary for the development of design tools for peptide-based antibiotics.

Predictive Model of Linear Antimicrobial Peptides Active against Gram-Negative Bacteria.

Antimicrobial peptides (AMPs) have been identified as a potential new class of anti-infectives for drug development. There are a lot of computational methods that try to predict AMPs. Most of them can only predict if a peptide will show any antimicrobial potency, but to the best of our knowledge, there are no tools which can predict antimicrobial potency against particular strains. Here we present a predictive model of linear AMPs being active against particular Gram-negative strains relying on a semi-supervised machine-learning approach with a density-based clustering algorithm. The algorithm can well distinguish peptides active against particular strains from others which may also be active but not against the considered strain. The available AMP prediction tools cannot carry out this task. The prediction tool based on the algorithm suggested herein is available on https://dbaasp.org.

Evaluation performance of substitution matrices, based on contacts between residue terminal groups.

Sequence alignment is a standard method for the estimation of the evolutionary, structural, and functional relationships among amino acid sequences. The quality of alignments depends on the used similarity matrix. Statistical contact potentials (CPs) contain information on contact propensities among residues in native protein structures. Substitution matrices (SMs) based on CPs are applicable for the comparison of distantly related sequences. Here, contact between amino acids was estimated on the basis of the evaluation of the distances between side-chain terminal groups (SCTGs), which are defined as the group of the side-chain heavy atoms with fixed distances between them. In this paper, two new types of CPs and similarity matrices have been constructed: one based on fixed cutoff distance obtained from geometric characteristics of the SCTGs (TGC1), while the other is distance-dependent potential (TGC2). These matrices are compared with other popular SMs. The performance of the matrices was evaluated by comparing sequence with structural alignments. The obtained results show that TGC2 has the best performance among contact-based matrices, but on the whole, contact-based matrices have slightly lower performance than other SMs except fold-level similarity.

Comparison of the efficiency of evolutionary change-based and side chain orientation-based fold recognition potentials.

The present article describes residue level knowledge based potential SORDIS. SORDIS incorporates the information on side-chain orientation in relation to hydrophobic core centres, distance of residue from the globule centre and secondary structure. SORDIS has been tested and compared with widespread evolutionary change-based substitution matrices (BLOSUM, PAM, GONNET, Johnson-Overington, BLAJ, HSDM, and STROMA) in fold recognition experiments within the zone of weak sequence similarity (<16%). The obtained results show that the lower is the amino acid similarity between homologous pairs the higher is the performance of SORDIS in comparison with the potentials, based on the information about the evolutionary changes. Therefore, we propose that the employment of SORDIS in fold recognition can be useful.