Table 1.
Main Artificial Intelligence technologies used in the included studies.
| Reference | Characteristics | Outcomes | Technology |
|---|---|---|---|
| da Cunha et al., 2021 | Detecting certain metabolic fingerprints through spectroscopy; using ML technology to analyze and further predict the mechanism of action and potency of different antibiotics | Successfully predicted the mechanism of action; accurate estimation of the antibiotic potency | ML + high-throughput Fourier-transform infrared spectroscopy |
| Zoffman et al., 2019 | Searching, identifying, and predicting potency of compounds with a random forest model | Assess phenotypic changes and antibacterial potency; predicted the phenotypic changes in compounds with identical and different mechanism of action | ML-random forest model |
| Stokes et al., 2020 | Using DL and NN to search databases and predict potential antimicrobial compounds, further empirically testing them | Successfully combined AI technologies and clinical investigation; halicin displayed strong antibacterial properties | DL + NN |
| Parvaiz et al., 2021 | Using ML to search for and identify potential candidates possessing beta-lactamase inhibition quality | Identified 74 compounds, out of which one showed great promise and further used ML in order to search for compounds structurally similar, concluding that all of the 28 additionally returned results had antibacterial activity | ML-random forest model |
| Hamid et al., 2019 | Used neural networks in order to distinguish between bacteriocin and con-bacteriocin sequences | The algorithm can successfully predict and classify bacteriocins based on their sequence | RNN |
| Fields et al., 2020 | Used ML to design and test bacteriocin-derived compounds and further assess their antimicrobial activity | The study designed and empirically tested compounds returned by the ML algorithm, with significant results | ML |
| Badura et al., 2021 | Used ANN to generate computational chemistry models and identify and classify compounds | Transformed chemical information into computational models used to further search and identify antimicrobial compounds | ANN |
| Feng et al., 2019 | Used IDQD in order to analyze and search for patterns in certain sequences and predict further patterns in antibacterial peptides | The study used this type of ML to successfully identify antimicrobial agents based on certain features of the antibacterial peptides | ML-IDQD |
| Bhadra et al., 2018 | Used ML to analyze the distribution pattern of amino acids in antibacterial peptides | The model grouped amino acids based on certain properties in different groups and further predicted and identified antimicrobial peptides | ML-random forest model |
| Napgal et al., 2018 | Used ML to search, analyze and predict peptides based on certain features | Analyzed peptides capable of inducting response of the APCs and further used ML to predict such peptides based on their structure | ML |
| Su et al., 2019 | Used NN trained on various datasets to achieve performance in feature selection and structure analysis | Analyzed the features and structure of amino acids and peptides in order to identify novel antimicrobial peptides | NN |
| Fjell et al., 2007 | Used Hidden Markov models to construct an algorithm that enables recognition of individual classes of antimicrobial peptides | Constructed a database that functions as a discovery tool for antimicrobial peptides | NN-Hidden Markov models |
| Cherkasov et al., 2009 | Used NN to search various databases and identify and further design antimicrobial peptides | Screened a large number of peptides and selected the most potent ones for in vitro testing, further concluding that two compounds exhibited strong antimicrobial effects | NN |
| Cruz-Monteagudo et al., 2011 | Used ML to create and define classification rules for antimicrobial peptides | The study aimed to assess both the toxicity and potency of antimicrobial peptides | ML |
| Grafskaia et al., 2018 | Used ML to create an algorithm for the identification of toxin-like that also acted as antimicrobial agents | The combined ML and proteomic technologies showed the potential of such research, even though the study returned a small number of candidate peptides | ML |
| Macesic et al., 2020 | ML was used to assess and quantify both bacterial resistance and susceptibility to certain antibiotics | Used ML to predict phenotypic polymyxin resistance in Klebsiella pneumoniae and to assess antimicrobial susceptibility | ML |
| Mansbach et al., 2020 | The study used ML to construct and design molecules capable of penetrating the membrane of Pseudomonas aeruginosa | The algorithm constructed and considered every possible fragment-based design, obtaining five compounds that were experimentally validated and showed good membrane penetration | ML-Hunting Fox Algorithm |
| Smith et al., 2020 | The study used ML in combination with genetic algorithms to assess intrinsic activity and efficacy of compounds | The study aimed to optimize dosing regimens when using antibiotic combinations, particularly against A. baumannii, the algorithm returning six regimens capable of eradicating the bacteria; even though these were not empirically tested | ML |
| Hu et al., 2007 | The study used ML and conventional methods to find new antimicrobial agents to counter the antimicrobial resistance of Yersinia spp. | The study combined ML and multiple conformational high-throughput docking in order to find YpkA inhibitors; the algorithm returned 7 compounds that were empirically tested and showed antimicrobial activity | ML |
ML: Machine Learning; DL: Deep Learning; NN: Neural Networks; RNN: Recurrent Neural Networks; ANN: Artificial Neural Networks; IDQD: Increment of Diversity with Quadratic Discriminant.