Highlights in BioScience ISSN:2682-4043 DOI:10.36462/H.BioSci.202304 Review Article Open Access 1 Department of Microbiology, Central University of Tamil Nadu, Tamil Nadu, India. * To whom correspondence should be addressed: rkaushik@cutn.ac.in Editor: Alsamman M. Alsamman, International Center for Agricultural Research in the Dry Areas (ICARDA), Giza, Egypt. Reviewer(s): Suleiman Aminu, Department of Biochemistry, Ahmadu Bello University, Zaria, Nigeria. AbdulAziz Ascandari, Chemical and Biochemical Sciences-Green Process Engineering, University Mohammed VI Polytechnic, BenGuerir, Morocco. Received: October 10, 2023 Accepted: December 24, 2023 Published: December 30, 2023 Citation: Mary AS, Patil MM, Kundu G, Rajaram K. Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach. 2023 Dec 30;6:bs202304 Copyright: © 2023 Mary et al.. This is an open access article distributed under the terms of the Cre- ative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: All relevant data are within the paper and supplementary materials. Funding: The authors have no support or funding to report. Competing interests: The authors declare that they have no competing interests. Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach Aarcha Shanmugha Mary 1 >< , Manali Manik Patil 1 >< , Godhuli Kundu 1 >< , Kaushik Rajaram *,1 ><� Abstract The emergence of antimicrobial resistance (AMR) has overwhelmed the contemporary curatives and have turned into one of the major challenges in the biomedical sector. With increasing deaths being associated with AMR every year; early detection of pathogens and development of novel drugs and alternative therapies, have all become ad hoc in diagnosis, prognosis and patient survival. Bacteriophage therapy remains a viable strategy to counteract AMR, yet unduly restrained by phage resistance. Phage infection is a natural phenomenon and can be widely manipulated in vitro using advanced techniques including the CRISPR/Cas systems which renders phage therapy an upper hand in comparison to conventional drugs. Phage identification, host range detection, determination of phage-receptor binding efficiency, adsorption rate, phage genome analysis are crucial stages in phage selection and phage cocktail preparation and moreover pivotal in flourishing phage therapy. The ascent of translational research and omics has allowed the development of quick, reliable and precise strategies for phage-based diagnosis and treatment techniques. However, in vitro evaluation of AMR and phage factors as well as storing, processing and analyzing large laboratory data outputs are expensive, time-consuming and labor-intensive. Machine learning (ML) is a utilitarian strategy to organize, store, analyze data sets and more importantly allows prediction of certain features by recognizing patterns in the data sets. With the huge number of research been carried out around the globe and enormous data sets being published and stored in databases, ML can utilize the available data to perform and guide in developing alternative therapeutics. Several ML based tools have been developed to predict resistance in host, phage grouping for cocktail preparation, resistance and lysogenic genes detection, phage genomic evaluation and to understand phage-host interactions. ML also allows the in silico analysis of large samples (drug/phage) and reduces sample size for in vitro evaluation thereby reducing overall costs, time and labor. The present review summarizes the available ML algorithms and corresponding databases used in AMR and phage research. It also emphasizes the status quo of antimicrobial and phage resistance in the healthcare sector and analyses the role of ML in analyzing biological databases in order to predict possible phage/drug-host interaction patterns, phage susceptibility, suitability of phage strains for therapy and recommends the most efficient drug combinations and treatment strategies. Keywords: Antimicrobial, AMR, Bacteriophage, Resistance, Machine learning, Database, Phage, Bacteria, Pathogen, Infection Background An alarming hike in morbidity and mortality due to bacterial infections have been reported worldwide in the past decade [1]. The Centre for Disease Control and Prevention and the World Health Organisation have arbitrated in this global health threat, and have identified and listed the priority pathogens entitled with the acronym ESKAPE, and have implemented strategies to combat pan drug-resistant (PDR), multidrug-resistant (MDR) and extensively drug-resistant (XDR) bacteria [2; 3]. The Clinical and Laboratory Standards Institute (CLSI) has disclosed the ineffectiveness and failure of antibiotics against ESKAPE pathogens overtime [4]. Highlights in BioScience Page 1 of 18 December 2023|Volume 6 https://doi.org/10.36462/H.BioSci.202304 https://creativecommons.org/licenses/by/4.0/ https://creativecommons.org/licenses/by/4.0/ mailto:aarchaachu3939@gmail.com mailto:manalipatil85801@gmail.com mailto:gksonai03@gmail.com mailto:rkaushik@cutn.ac.in http://orcid.org/0000-0002-1830-4083 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach Development of alternative drugs is on the horizon and in 2019, Mulani et al., catalogued the emerging approaches like antimicrobial peptides, antibiotic combination therapies, photo- dynamic light therapy, bacteriophage therapy, nanoparticles and phytochemicals to combat antimicrobial resistance (AMR) [4]. Additionally, state-of-the-art treatment strategies like CRISPR/ Cas9 and bacterial vaccines have also been surfaced and are under exploratory research and development [5]. Bacteriophage are ubiquitous and are present alongside the host and are pivotal in regulating bacterial populations; thereby maintaining microbial balance. Phage therapy has been largely neglected during the antibiotic era; however, with the emergence of drug-resistant bacteria it has regained its lost glory. On one hand, bacteriophage remains a logical and sustainable remedy for AMR, whereas bacterial evolution and fitness in presence of phage allows the persistent cells/mutants to grow and prop- agate [6]. Bacteriophage resistance can also be associated to bacterial abortive mechanisms, receptor mutations, Restriction- Modification (RM) systems and CRISPR-Cas systems [7]. While bacteriophage resistance is a backlash for phage based treat- ment regimen, certain clinical implications incurred due to phage resistance are of therapeutic importance. In a clinical setting, phage resistance does not only result in treatment failure but can also aid cross-resistance across other phage and antibiotics [7]. In 2018, Wright et al., demonstrated the modular network of cross-resistance in spontaneous mutants of P. aeruginosa against 27 distinct phage; and emphasized the role of cross-resistance network based prediction of mutation frequency and phage com- binations for successful therapy [8]. Screening and identifying putative phage for effective host inhibition is yet another major challenge where conventional methods can turn out to be unreli- able, time-consuming and labour-intensive. Melo et al., in 2022, evinced the use of Flow Cytometry for fast, reliable and high- throughput phage screening [9]. Genomic analysis associated to receptor synthesis of P. aeruginosa phage (K8) resistant-mutants by Pan et al., in 2016, revealed the absence of O-antigen (serve as receptor) in mutants [10]. Similarly, innumerable research has documented the genes associated to the receptors modification in host [11; 12; 13], receptor binding proteins (RBP) [14; 15], antimicrobial genes (ARGs) [16; 17], cross-resistance network [8], phage-host interaction [18; 19], endotoxins [20; 21], lysins [22; 23; 24; 25], pharmacokinetics [26; 27], whole genome data [28; 29; 30], gene expression data [31; 32] and structural infor- mation [33; 34] all of which are paramount information in estab- lishing phage therapy. The advancements in research strategies, subsequent outputs and innumerable data has been largely con- trolled by computational intervention and developments in bioin- formatics [35]. Even when bioinformatics provides large range of data storage, indexing, analysis and data retrieval; recent develop- ments in machine learning including deep learning and artificial neural network (ANN) have allowed the unprecedented predic- tion by analysing massive data sets [36]. Biological databases or repositories are key prerequisites in machine learning and are classified into primary, secondary and composite, and typically reserve molecular sequences (GenBank, DDBJ, PIR), structures (PDB), metabolic pathways (KEGG, MetaCyc), enzyme structure and interactions (BRENDA), molecules (PubChem), microarray gene expression data (GEO), taxonomic data (Catalogue of life), disease data (OMIM), model organisms (RGD, Flybase) and bibliographic data (PubMed) [37]. Machine learning algorithms learns from these data sets by analysing and identifying patterns; and this is particularly useful in prediction, classification, fea- ture identification/selection and clustering [38]. ML algorithms have remarkable applications in biomedical research including disease diagnosis, drug discovery and development, personal- ized medicine, medical image analysis, electronic health record analysis and supreme role in analysing raw data in genomics and proteomics [39]. Several concerns arise in using machine learning and deep learning for biological data analysis, including ethical integrity and analysis of noise data along with selected features [40]. This review, elaborates the use of Machine learn- ing in devising efficient therapeutic strategies to combat AMR as well as its applications in exploratory and basic research. It also briefs the potential of phage therapy and corresponding databases and ML algorithms to aid in the development of phage based curatives and strategies to overcome phage resistance by pre- dicting bacterial susceptibility, phage host range and infectivity. This review also summarizes currently used ML algorithms and databases for research and medical purposes. Phage therapy The antibiotic pipeline has shown seldom growth with very few novel compounds being commercialized in the recent years [41]. It is safe to say that the ’post-antibiotic era’ is approaching, and the development of effective alternative therapies is essential. The discovery of the mighty penicillin and the subsequent devel- opment of other broad-spectrum antibiotics marked the beginning of the golden age of antibiotics, and as a result, the then-used alternatives like phage and other therapeutics have been side- lined since the 1940s [42]. Lytic phages are potential bactericidal agents and have been successfully used in clinical practices. The phage-bacterial interactions are commonly observed to be ob- ligate lytic (virulent), pseudolysogenic, lysogenic, and chronic [43]. Lytic phages are bacterial viruses that replicate immedi- ately after entering the host cell and release the progenies; on the other hand, lysogenic phages are viruses that integrate the phage genome with the bacterial genome for generations and can resume the lytic life cycle [44]. Pseudolysogeny occurs when the host cells are deprived of nutrients, and the phage genome neither enters the lytic nor lysogenic cycle but stays inactive [45]. Chronic life cycle is also known as the carrier state where the progeny is released through the host cell membrane without rupturing or damaging the cell, resulting in a long-term infec- tion [44]. While lytic phages are the only bacterial viruses that could be translated for therapy, lysogenic conversion of phage in bacteria has allowed the acquisition of undesirable genes. For Highlights in BioScience Page 2 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach instance, Corynebacterium diphtheria and enterohaemorrhagic Escherichia coli acquired toxin-producing genes - diphtheria toxin (siphovirus β-phage) and Shiga toxin (lambdoid phage) respectively from the integrated phage genome [46; 47]. Besides the type of phage life cycle, the optimal bactericidal efficacy of phage is also dependent on factors like adsorption rate, latency period, multiplicity of infection (MOI), and burst size. However, bacterial features including the host outer membrane (LPS, cap- sule, peptidoglycan), receptors, presence or absence of flagella or pili can alter the fate of phage infectivity. Xuan et al., through their experiment in 2022, proved that quorum sensing upregu- lated the synthesis of lipopolysaccharides (typically for biofilm production) which are receptors for phage adsorption, thereby promoting phage infection [42]. Phage-encoded enzymes like depolymerases are capable of degrading the glycan protective layer in bacteria to establish infection [48]. At a clinical setting, the use of phage cocktails and lysins are also potential strategies to ensure bacterial growth suppression. In 2019, Aslam et al., reported the use of bacteriophage cocktails for treatment against P. aeruginosa and Burkholderia dolosa infection in lung transplant recipients [49]. Similarly, independent studies have evaluated the use of CT-PA (P. aerugi- nosa cocktail), AB-SA01 and NOV012 (Staphylococcus aureus cocktail) for treatment against chronic rhinosinusitis [50; 51; 52]. Uyttebroek et al., in 2021, also summarized the use of bacterio- phage against recalcitrant chronic rhinosinusitis due to S. aureus and P. aeruginosa colonization and biofilm formation [53]. A recent study in 20 patients positive for Mycobacterium infec- tion treated with intravenous phage administration, 11 patients showed a favorable response, and phage neutralizing antibodies were found in 8 patients [54]. Topical phage administration is a reliable strategy to ensure maximum phage stability, with minimal immune responses. PP1- 131 , phage cocktail used against P. aeruginosa burn wound infec- tions, was found to reduce the bacterial load, but at a significantly lower rate in comparison to the standard of care treatment. The drop in the phage performance was linked to the loss of titer during manufacturing where the participants received low phage concentrations than intended [55]. FAGOMA (Spanish Network of Bacteriophages and Transducing Elements) in its regard to bacteriophage therapy advised the use of phage-based therapy only for patients infected with MDR pathogens or in case of antibiotic hypersensitivity and infection in antibiotic reluctant areas such as the prosthetics [56]. Even with a vast number of in vivo and clinical studies, com- mercialization of phage-based therapy has been under scrutiny due to the heterogeneity in the outcome and also due to the lack of unreported adverse effects. From the stage of selection of phages (single phage/cocktails), mode of administration to the treatment duration, no standards have been established till date. Onsea et al., in 2021, devised a multidisciplinary strategy to overcome aforementioned hurdles in establishing phage therapy as a standard of care treatment [57]. Consecutive reports have been updated on the status of phage therapy in Germany from the 1930s till date along with the present challenges in phage produc- tion and commercialization on a large scale [58]. Phage-based therapeutic product development should ensure high quality, ef- ficacy, and safety for clinical usage and, moreover, should be GMP-certified [59]. Another widely accepted aspect of phage application in clin- ical practice is the adjunct use of phage with antibiotic. Liu et al., in 2020, evaluated the phage-antibiotic interactions (antag- onism, synergy, additive) by analyzing the stoichiometry and among different classes of antibiotics. Through specific real-time readout synograph’, they concluded that the mechanism of phage- antibiotic synergy is antibiotic class dependent, and the syner- gistic activity can be suppressed by bacterial growth conditions [60]. The resensitization of antibiotic-resistant strains during phage-antibiotic combination therapy is attributed to the trade-off costs during bacterial fitness. Wang et al., in 2021, demonstrated the use of colistin-phage (Phab24) combinations against Acineto- bacter baumannii, and reported that phage Phab24 was capable of eliminating both colistin-sensitive and resistant strains along with increased sensitivity of phage-resistant mutants to colistin [61]. Emerging proofs are indicative of successful phage therapy, whereas resistance is a significant factor in establishing the same. Antimicrobial and phage resistance The resurrection of phage therapy and augmentation of an- tibiotics have shown potential in developing efficient bacterici- dal therapeutics; whereas, phage and antibiotic/drug resistance prevails as a major concern. Bacteria attain antimicrobial re- sistance by employing one or more mechanisms that may in- volve either drug alteration/inactivation, overcoming intracellular drug accumulation, modification of the drug binding sites, ef- flux pumps, or by forming biofilms, Figure 1 [62]. ESKAPE pathogens also acquire novel resistance mechanisms which are not part of their natural intrinsic defense methods. Enterococcus faecium produces penicillin-binding-protein 5 (PBP5) that pro- vides protection against β-lactam drugs (penicillin, ampicillin, and cephalosporins). E. faecium also resists the combined doses of aminoglycosides and β-lactam/glycopeptides by chromoso- mal AAC (6’)-I enzyme [63]. The clonal complex 17 (CC17) strains of E. faecium are assigned to be responsible for hospital- acquired E. faecium infection and contain virulence and resis- tance genes conferring protection against a series of antibiotics, including ampicillin and quinolones [64]. Similarly, S. aureus have attained resistance against penicillinase-resistant drugs like methicillin, cloxacillin, and oxacillin (MRSA strains) but remain susceptible to glycopeptides. Resistance towards β-lactam drugs is conferred by plasmid-encoded blaz gene as well as through PBPs (mecA & mecC) [65]. Last resort drugs like colistin and tigecycline are used for ESBL (extended-spectrum β-lactamases) and carbapenemase-producing Klebsiella pneumoniae infections; whereas, colistin resistance is also reported (mcr gene) giving rise to untreatable pan-drug infections [66]. Highlights in BioScience Page 3 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach In regard to Acinetobacter baumannii and P. aeruginosa, the intrinsic defense is quite intriguing with impermeable outer mem- brane and enhanced efflux pumps. β-lactamases and PBP confer protection against β-lactam and carbapenem drugs in most cases [67]. In P. aeruginosa, OprD loss is associated with imipenem resistance [68]. Antibiogram is one of the most commonly used cost-effective in vitro resistance detection strategies. Other phenotypic meth- ods include MIC detection (micro-dilution method), breakpoint agar method, biochemical tests, immunographic tests (CARBA-5, RESIST-4), electrochemical test (BYG test), motility test, stain- ing, etc. MicroScan walk assay, BD Phoenix, and Vittek 2 are examples of semi-automated MIC determination approaches [69]. Bacterial strain identification and differentiation are carried out using MALDI-TOF analysis and strain-specific tests like MGP (methyl-α-D-glucopyranoside) test (E. faecium), slide agglutina- tion test ( S. aureus), modified Hodge test, and carbapenemase inhibition method ( P. aeruginosa & A. baumannii), etc. Geno- typic characterization is ad hoc and mostly targets the detection of genes including superoxide dismutase (sodA), vanA, vanB (gly- copeptides resistance genes), mecA, mecC (methicillin resistance genes), blaKPC, blaNDM, blaIMP, blaVIM, blaOXA-48 (carbapenem resistance), mcr (colistin resistance), aphA6 , armA (amikacin resistance) through LAMP, microarray, real-time PCR, and whole genome sequencing (WGS) [69]. Bacteriophage resistance is a complex yet a natural evolution- ary process in bacteria. Nevertheless, the implications of phage resistance have more arena of challenges than antimicrobial resis- tance. Phage resistance is associated with trade-off /fitness costs that can alter the genotypic and phenotypic characteristics of the bacterial strain. Studies have reported reduced virulence, en- hanced cross-resistance, altered antibiogram, changes in biofilm formation, permeability variations, and several other modifica- tions during phage exposure [70]. Hesse et al., in 2020, evaluated 57 phage-resistant mutants ( K. pneumoniae) and reported that each mutant had distinct genes involving in mutation but with similar function associated with assembly or synthesis of surface receptors which ultimately affected phage adsorption in mutants [71]. Phage-resistant UPEC (Uropathogenic E. coli) strains ex- hibited alterations in LPS (lipopolysaccharide) and were con- firmed via WGS [72]. Garb et al., studied the defense-associated sirtuin (DSR) proteins in 2022, and found that SIR2 (N-terminal sirtuin) domain acts as NAD+ depletors and helps bacteria to abort phage adsorption and propagation [73]. One of the major findings by Laure et al., in 2022, on trade-off costs in Salmonella typhimurium was the increased β-lactamase activity of mutants obtained after phage and phage +ampicillin treatment [74]. Li et al., in 2022, investigated multiple mutant strains of P. aerugi- nosa and revealed major mutations in pilT and pilB (type IV pili) genes and chromosomal deletions of approximately 294 kb in- volving galU (UTP-glucose-1-phosphate uridylyltransferase) and hmgA (homogentisate 1,2-dioxygenase) [75]. Enhanced DNA exonuclease activity through overexpression of mpr gene in My- cobacterium smegmatis was proved to confer resistance against phage [76]. Owen et al., in 2021, identified BstA-phage-defense proteins encoded by prophage that inhibit exogenous lytic phage infections [77]. Similarly, Charity et al., in 2022, correlated mTmII prophage genome integration in Salmonella typhimurium to cause increased fitness and drug resistance [78]. Aforemen- tioned studies are rather a glimpse at the ongoing research to unravel phage-host interactions and resistance mechanisms. A compelling need to develop sophisticated strategies to control and allow the development of prediction models prevails. The data produced through these studies could be put in use to manifest a much clearer picture for the development of phage and other antimicrobial therapeutics. Machine learning Machine learning (ML) is a scale-up strategy in refining and developing Artificial Intelligence (AI). Contradictory to sym- bolic AI; ML focuses on learning from datasets and develops an algorithm that could be novel with a distinct understanding of certain features and their respective weights [79]. In biomedical research, computational methodologies have failed in analyzing enormous datasets and ML comes in handy where a labeled/un- labeled dataset is used to train, validate, and test algorithms. Interpretation using ML becomes more systematic and allows classification, clustering, and, more importantly, prediction [80]. Supervised, semi-supervised, unsupervised, and reinforcement learning are the major learning methods used in ML. In layman’s terms, ML algorithms are fed with raw data, data that are labeled (input/output, cause/effect), or with unlabeled data; either way, ML identifies hidden patterns and allows classification/regression (supervised) grouping/clustering (unsupervised) [81]. Supervised learning uses a labeled dataset with identified features and tar- gets, and unsupervised datasets use unlabeled (raw) data where the algorithm determines the best features and target [82]. Re- gression and classification are two common tasks in supervised learning. On the other hand, association, clustering, and anomaly detection are tasks in unsupervised ML. Reinforcement learn- ing is a trial-and-error-based learning strategy and is regarded as the best attempt at modeling human-like learning experience [83]. Support vector machine (SVM), linear regression, logistic regression, naïve Bayes classifier (NB), ANN, k-nearest neighbor (kNN), random forest, and decision trees are ML algorithms that are used to understand the relationship among the features and outcome/target [84]. While linear regression (univariate/multi- variate) uses a linear line to describe the relationship, logistic regression predicts a sigmoidal relationship between features and the probability of an outcome. Decision trees classify the data based on features, beginning from a root node and parti- tioning into decision and terminal nodes. Random forest is an ensemble producing several decision trees [83]. Deep learning is a subset of ML, where hidden neural layers perform learning and decision-making more efficiently without other intervention [85]. Stochastic models are more approachable in the case of Highlights in BioScience Page 4 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach Figure 1. Antibiotic and Phage resistance mechanisms in bacteria. A) Intrinsic phage defence involves production of thick outer membrane and receptor modifica- tions. After the phage genome enters the bacterial cell restriction modification (RM) system or CRISPR Cas system degrades or inactivates the genome and avoids further transcription, translation and phage assembly. B) Antibiotic or drug resistance mechanisms involves enzymatically inactivating the drugs, drug modification and target site alteration. Intrinsic mechanisms involve biofilm production, reduced drug uptake by altering membrane permeability and enhanced efflux pumps. randomness and ambiguity in data. In 2021, Goodswen et al., reviewed the uses of ML in biology, which included disease diagnosis, predicting outbreaks/vaccine candidate/drug targets, drug resistance identification, and microbial interactions [86]. With the assistance of some reliable methodological strategies, it will be more realistic to model the global pan-genome. The Expectation-Maximization (EM) algorithm is the multitude ap- proach for determining maximum a posteriori estimates (MAP) or maximum likelihood estimates (MLE) for undocumented mod- els in statistical analysis [87]. Reboredo et al., in 2021, have updated the trends in the usage of ML in drug discovery; the authors also emphasized on the economic advantages of ML in detecting active compounds and thereby reducing the cost and effort in processing large sample sizes in pre-clinical and clinical studies [88]. Convolutional Neural Network (CNN) in ML is re- portedly efficient in analyzing images (MRI, CT, PET scans) with significant application in nuclear medicine [89]. With unlimited application in the biomedical sector along with novel develop- ments, ML has become a reliable tool. Among several other applications, ML in AMR and development of phage therapy is a growing area of interest. Machine Learning and Antimicrobial Resistance Computation of data ensures greater accuracy in analysis, which is the foremost requirement for data analysis in the biomed- ical sector. Several infectious diseases are frequently being treated with antimicrobial drugs; however, antimicrobial resis- tance in bacteria has become a generic reason for treatment fail- ure. Developments in machine learning (ML) have allowed the detection and prediction-based applications in antimicrobial re- sistance [90]. Distinct tools for determining the antimicrobial resistance genes and virulence genes based on ML are currently available (Figure 2). A specific approach of ML in AMR is by analyzing the suit- able synergistic drugs for the development of possible combi- natorial therapies. INDIGO (Inferring drug interactions using chemogenomics and orthology) uses the best-fitted synergistic drugs for the treatment of intra-abdominal infection using Gen- tamycin -Ampicillin developed using model predictions [91]. Another refined model that is a bit different in application is the MAGENTA (metabolism and genomics-based tailoring of antibiotic regimens), which facilitates the treatment of biofilm using combinatorial Rifampicin dosage [92]. Determination of antimicrobial resistance using ML is now being used in various Highlights in BioScience Page 5 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach Figure 2. Machine learning algorithm utilized antibiotic and/or virulence gene databases to predict antimicrobial resistance, alternative therapies and its correspond- ing side effects. The Machine learning algorithms can be used to develop personalized medicine and also can predict the synergistic effects of certain combinatorial therapies. sectors, including the first line of intensive care. ML is not only limited to a targeted approach but also narrows down the frequent use of multiple antibiotics under the same diseased conditions [93]. For medical professionals, ML has now become the key predictor for antimicrobial resistance, where ML-based tools are taken into consideration for allowing the screening of the best- suited antibiotics for the pathogens. Recursive partitioning is applicable for the prediction-based analysis of ESBL production from Klebsiella sp. and E. coli [94]. Modified recursive par- titioning models are found to be advantageous for developing the logistic regression models [95]. XGBoost, one of the openly accessible platforms for ML-based algorithms, has guided the prediction-based analysis of antibiotic resistance for some of the gram-negative bacterial species [96]. The predicted perfor- mance by XGBoost is better than other risk assessment tools; however, the prediction model is a bit selective and applicable to limited antibiotics. The ML system can also be futuristic for determining the extent of antibiotic resistance during the culture collection. ML would capture the empowerment for the upcom- ing advancement of antimicrobials determination. ML enables the development of personalized medicinal approaches to cut down frequent uses of antibiotics. The Random Forest Classifier, one of the most supervised ML algorithms, is allocated to accel- erate the screening of MDR bacteria among the patients in ICU [97]. The associated empirical data of the ML-based algorithm can also be implemented for local antimicrobial susceptibility assessment. Some of the ICU antimicrobial susceptibility-based ML algorithms are Multilayer Perceptron and J48 [98]. As discussed earlier, the principles of ML-based systems are i.) being trained from the existing datasets (supervised) and ii.) analyzing or giving output based on unlabeled data (unsuper- vised). Information regarding the underlying drug interactions has to be made available for ML to learn and develop algorithms for predicting the best combinatorial treatment approach. Primar- ily, certain attributes like changes in gene expression, response to the drugs, etc., need to be characterized before computerized for decision-making [99]. Based on the available datasets of drug information like chemical structure and function, the algo- rithms are trained to determine the combinatorial therapy. Among multiple algorithms, CoSynE (Combination Synergy Estimation) governs the direct structural analysis for an individual drug can- didate to analyze the overall combinatorial therapy. Basically, to have knowledge of every aspect of the combinatorial compounds, there should be prior data for the drug and the target as well [100]. With the CoSynE approach, the level of application can be ex- tended towards diversified combinatorial drug designing against Highlights in BioScience Page 6 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach drug-resistant E. coli and also against malarial parasite Plasmod- ium [99]. Through consistent records, the ML algorithms were able to predict drugs to counteract antimicrobial resistance in Mycobacterium and Salmonella [101]. The predictability is well- curated to reframe the ML so that antibiotic resistance can lay the foundation for the model-based prediction. The extent of genetic variations and drug resistance can be supervised using Machine Learning algorithms. The drug re- sistance predictability is done so far in accordance with IC50 indicators. The various searches are based upon the retrieval of different sequence databases. Tuberculosis, caused by M. tuber- culosis, is one of the infectious diseases which could be eradi- cated using antibiotics, but due to the dosage’s dysfunctionality and random uses of antibiotics, it concludes with a highlighted remark towards the development of MDR Tuberculosis [102]. Administration of fluoroquinolones is currently considered, but the usage could lead to after effects due to reported toxicity. ML predicts novel drugs, mechanistic performance of the drugs, mi- croarray data construction and analysis, side effect prediction, etc. Based upon the existing known dataset, ML-based algorithms can display the futuristic blind prediction. The performance based on the suggestive prediction can confer the right indication towards clarifying the Single Nucleotide Variance followed by the mutation within the genome of Mycobacterium tuberculosis. Sequenced based algorithms and structure-based simulations em- ploy the right directive approach to analyze the genomic variation, so that suitability for selecting the target gene can be obtained. Deep learning is efficient in predictable analysis to distinctively diagnose the inherent antimicrobial peptide (AMP) to combat antimicrobial resistance. Deep Learning was able to evaluate AMPs as antimicrobial agents and predicted that AMP can sur- pass MDR in carbapenemase-producing E. coli. Deep learning is also capable of simulating the newer versions of AMPs from the existing peptides [103]. With access to different databases like ARGs, a number of Deep Learning models have been recruited so far to deliberately screen the best-hit searches for antimicrobial genes. A summary of the ML tools for antimicrobial resistance and its associated factors is depicted in Table 1. Machine Learning in phage therapy Machine learning has arisen as a promising strategy in phage therapy prediction, intending to discern the most effective bacte- riophage that can selectively target distinct bacterial strains. Cur- rently, various techniques exist to measure bacteriophage-host interactions experimentally, such as PhageFISH-CLEM [113], microfluidic digital multiplex PCR [114], flow cytometry [115], agar overlay assay [116], RNA-sequencing [117], spot test, and efficiency of plating [118]. While these methods are highly ac- curate, they are costly, labor-intensive, time-consuming, and can turn out to be inconclusive. To overcome these limitations, re- searchers have developed high-precision computational methods for predicting phage-host interactions. The initial phase in phage therapy is the selection of suitable phage, characterized by three indicators that include 1. Presence of temperate markers 2. Pres- ence of anti-microbial genes and 3. Presence of Virulence Gene. ML basically uses the whole phage genome and corresponding proteome analysis, and from the sequence similarity and anal- ysis of the conserved domains, determines whether the phage undergoes lytic or lysogenic cycle. A recent approach has been developed to determine the suitability of phage for therapeutic purposes based on an online single-step predictor tool. This pre- dictor tool utilizes the protein features such as integrase, Cro/CI repressor protein, anti-repressor proteins, immunity repressors, etc., along with ABRicat tools for determining the antibacterial and virulence genes. Apart from ML, other computational tools are also in use to identify host range and host-phage interactions. The alignment-free and alignment-based techniques are widely used tools, where the sequence homology and sequence similarity among host and phage are read by computational tools to predict its host range [119]. BLAST is a classic example of an alignment- based method. While the alignment-based method (e.g., Phirbo) is the most reliable strategy with the maximum prediction accu- racy, certain factors like the prediction of multiple related hosts, spurious alignment, and artifacts lead to comparable false results. Alignment-free tools use similarity in the sequence composition of codons, oligonucleotide frequencies, etc. This tool comes in handy where the host and phage lack sequence homology, and thus alignment-based tools are not suitable. ML develops algorithms based on dataset creation (phage genomes and its corresponding protein sequences), feature gen- eration (custom scripts), and validation. Computational evalu- ation precedes other methods by allowing the comparison in a large dataset, especially advantageous for non-cultivable bacteria and the availability of limited host strains. HostPhinder exam- ines phage genome sequences to predict the bacterial host of phage [120], VirHostMatcher measures CRISPR sequences and alignment-free similarity to predict virus-host interaction [121], WIsH outperforms VirHostMatcher at various taxonomic levels and predicts the host range of bacteriophages through genome sequences [122], Machine learning takes into consideration the measurable properties referred to as features’. Nucleotide se- quence is a characteristic feature and is used by ML algorithms like Prokaryotic Virus Host Predictor (PHP) and Host Taxon Pre- dictor (HTP) for phage-host interaction prediction. The absolute relative oligonucleotide frequencies and a Gaussian model are used to predict the host in HTP and PHP, respectively. Based on the virus-host associations, ILMF-VH and LMFH-VH integrate the virus and host similarity network [123], [124]. Leite’s method uses a One-Class learning method to predict the host-viral interac- tion at the bacterial strain level [125], SpacePHARER (CRISPR Spacer PhageHost Pair Finder) predicts bacterial and viral in- teractions at the protein level by comparing spacers and phage [126]. On the other hand, VirSorter uses phage-host interaction signals to predict the phage-host interaction [127], and Pred- PHI (Predicting Phage- Host Interactions) predicts the prokar- yote-phage interaction by sequence data [128]. PhageTB lever- Highlights in BioScience Page 7 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach Table 1. Methods, Tools & Algorithms, Model Organisms, Genes, Databases, Applications, and References Method Tools & Algorithms used Model organ- ism Genes Database Applications Reference Pan-genome con- struction, Random forest Scoary, Prodigal, CD-HIT, Glim- mer3, Naive-Bayes classifier E. coli AMR genes (from core and accessory gene clusters) PATRIC, CARD Analyses the accessory part of the genome to predict ARGs [104] WGS, Antibiogram K-mer, Random forest, XGBoost S. aureus SCC mec genes PATRIC Predicts Antimicrobial Pheno- type Resistance [105] Colony screening, RNA sequencing, DNA sequenc- ing, SNP calling, Pangenome analy- sis and indel call- ing, Multilocus sequence typing MSA2VCF, Illumina HiSeq 2500, MAFFT, SAMtools, BamTools, BCFtools P. aeruginosa PA14 gene UCBPPPA14, MLST database ML identified biomarkers assess AMR profiles through a molecular test system [106] Phylogenomics, and genome se- quencing mafft, raxml-ng, T-REX, TreeTime E. coli malT gene, manZ BiMat software Studies the phage-bacterial co- evolution dynamics (molecular and ecological mechanisms) [107] Oxford Nano-pore Technologies, Sin- gle Nucleotide Real Time (SMRT) REBASE, DECIPHER, Clustal Omega Cutibacterium acnes Mob genes, Res gene REBASE, NCBI Genome database, PFAM database, CRISPR-CAS Finder Restriction-methylation and host protective mechanisms in C. acnes strains [108] Twitching motil- ity assays, Swim- ming motility as- say. Drosophila melanogaster virulence assays. Secreted enzyme as- say. Antimicrobial resistance assay. Shearing assays. MICROB Express kit, Agilent 2100 bioanalyzer, BEDTools software v2.16.2. P. aeruginosa Morons CLC Genomics Workbench software v5.1 Bacterial and phage symbiotic interaction, Bacterial adaptation in various selective pressures [109] Mathematical model using an arbitrium-like com- munication system in a serial passage set up. Matlab R2017b Bacillus GitHub Small peptide mediated sig- nal communication (phage- phage),phage life cycle predic- tion. [110] Antimicrobial sus- ceptibility testing, Statistical analysis of multidrug resis- tance, Association set mining SENTRY, Apriori S. aureus AMR genes Analysis of Multidrug Resis- tance in Staphylococcus aureus [111] Serotyping, An- tibiotic suscep- tibility testing, Set-covering ma- chine, CMY-2 locus analysis Python libraries, SISTR, IQTree, Prodigal v2.6.3, Kover v2.0.0, BWA-MEM Salmonella enterica AMR genes Plasmidfinder database, DIAMOND v0.8.36 AMR genomic characterization of Non-typhoidal Salmonella serovars to train prediction models for AMR phenotypes. [112] Highlights in BioScience Page 8 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach ages accurately identifies hosts for bacteriophages using genomic sequences [129]. These user-friendly tools enable the study of phage inter- actions, identifying potential phages that can specifically tar- get pathogenic bacteria, leaving beneficial bacteria intact. Ma- chine learning-based approaches for phage therapy prediction have shown great promise in identifying effective bacteriophages for targeted treatment of bacterial infections. These methods offer faster and more cost-effective alternatives to traditional experimental techniques. However, challenges in data avail- ability, model selection, evaluation metrics, interpretability, and user-friendliness must be addressed for their successful imple- mentation in clinical settings [130]. Additionally, tools like PHACTS (Phage Classification Tool Set) and BACPHLIP (BAC- terioPHage LIfestyle Predictor) are widely used for determining phage lifestyle by annotating proteomes and classifying lifestyles based on conserved protein domains [131], [132]. BacteriophageHostPrediction is one other ML which con- siders more features, including genomic sequences, protein se- quences, protein secondary structures, and physiochemical prop- erties. PHERI identifies bacterial host from phage sequence through annotated protein sequence clusters. PhageLeads are ML that focuses on predicting the lifestyle of phage (lytic or temperate) using protein features of temperate markers. The quick prediction times of PhageLeads make it a valuable tool for efficiently identifying phages suitable for combating antibi- otic resistance, and its ability to detect resistance and virulence genes further enhances its utility [133]. Incorporating protein biological feature spaces may further enhance the functional sim- ilarities predictions [134]. Several databases like CARD [135], ShortBred AR [136], MEGARes [137], and VFDB [138] aid in detecting these genes. By overcoming these challenges, machine learning-based phage therapy prediction holds significant poten- tial in combating antibiotic resistance and improving treatment outcomes for bacterial infections. After the development of ML, the usability is determined by several factors such as operating system restriction, automation, and reproducibility. PHP, Host- Phinder are web-based prediction tools that are also available, which do not require a specific OS. Table 2 summarizes the ML algorithms developed for phage research. Databases in antimicrobial resistance and phage re- search In silico approaches for omics studies have become more widely available, which has in turn made it feasible to precisely recognize and catalog determinants in the case of antibiotic re- sistance and its associated genes. AMR archives, and software applications tools have been constructed for WGS-AST based on the available databases, which includes the Comprehensive Antibiotic Resistance Database (CARD), ResFinder and its com- panion database PointFinder, ARG-ANNOT, and many more (Table 3) [139]. The CARD offers an informatics paradigm for the notation and assessment of resistomes through integrating the Antibiotic Resistance Ontology (ARO) [140]. CARD is a database on the molecular basis of antimicrobial resistance that is ontology-focused. CARD has the potential to serve as both reference material and software instruments and resources for directing AMR investigation, particularly for ARG details and other findings from genomic information and metagenomic facts. This is made possible through the combination of an extensive modulated concept of ARO, with ARG [141]. The Resistance Gene Identifier (RGI), anticipates AMR from genome-wide facts and data using the bioinformatics prediction approaches and coordinated in CARD, and is a sophisticated strategy [142]. AM- RFinderPlus, a tool of NCBI, analyzes protein annotations and/or gathered nucleotide sequence to find AMR genes, resistance- linked point mutations, and particular subclasses of genes. The Pathogen Detection operations make use of AMRFinderPlus, and these data are shown in NCBI’s Isolate Browser [143]. AMRFind- erPlus makes use of the carefully selected Hidden Markov Mod- els and Reference Gene Database from the NCBI. The NCBI’s Pathogen Detection Project incorporates the output of AMRFind- erPlus to quickly group and locate associated pathogenic genetic patterns residing in food, environment, and people with illnesses. MicroBIGG-E (genomic data) together with the AMRFinderPlus; findings are provided in a more thorough manner, with extra data such as strain names and source of isolates [144]. In AMRFind- erPlus the outcomes are provided for download by the users in two interfaces graphics [139]. Each unique isolate has a synopsis of its antimicrobial resistance, stress-related responses, and vir- ulence gene sequence generated in the Isolates Browser, which may also be retrieved for more research. Challenges and Opportunities The use of machine learning in phage therapy faces several limitations and challenges. The future of health care system will be influenced by the intervention of AI and ML from organiza- tion to personalized precision care. The input of ML will also positively impact the diagnosis systems. One major obstacle is the need for diverse and unbiased datasets to train predic- tive models. Current datasets often suffer from data imbalance, leading to biased predictions and limited generalization to other bacterial species. Enriching a dataset with a broader range of phage-bacterium pairs involving different species and strains is essential for improving model accuracy and applicability [147]. Another limitation lies in the reliance on outdated databases to extract informative features that affect model relevance and per- formance. Incorporating up-to-date and comprehensive databases is crucial for generating more reliable predictions. Addition- ally, using deep learning models while achieving high accuracy presents challenges in model interpretability. These models are often considered "black boxes," creating challenges for users to comprehend the underlying decision-making process [152]. Balancing accuracy and interpretability is crucial for user trust and the real-world application of the models. The computational module for predicting phage lifestyle faces the limitation of uncer- Highlights in BioScience Page 9 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach Table 2. Machine Learning based tools to aid bacteriophage research. Machine learning predictors Purpose of ML Tools used Major data sources - Algorithms Predicting protein Efficiency Applications References Gradient Boosting Classifier (GBC) Predict phage virion protein (PVPs) using phage protein sequences. protlearn AdaBoost Classifier (ABC), Virion proteins 80 & 83 % accuracy for training and independent dataset. Discovery of holins and other phage-driven proteins, endolysins, and exopolysaccharides Identification of phage virion proteins (PVPs) [145] PhageLeads Determines the presence of temperate markers, antimicrobial resistance virulence genes ABRicate tool,BACPHLIP Plasmid Finder Integrases, Cro/CI repressor proteins, immunity repressors accuracy of 96.5% Predicts the lifestyle of phageDetects lysogenic protein and temperate markers [146] PHACTS Predicts the interaction of phage-bacterium HostPhinder,K-Nearest Neighbors (K-NN), Random Forests (RF)], Support Vector Machines (SVM) , and Artificial Neural Networks(ANN) GenBank, phagesDB.org, GeneMarkS, DOMINE, GeneMarkS, Pfam HMM, HMMER API Receptor-binding proteins 86% to 90% accuracy Predicts phage-host interaction [147] PhageTB, BLASTHost, BLASTPhage , CRISPRPred . Predicts phage-host interactions VirHostMatcher-Net XGBoost, Multi-layer Perceptron Accuracy of 67.9- 93.5% Identifying hosts, assessing phage-host interactions, aids candidate phage therapy [129] AcrNET Anti-CRISPR analysis RaptorX,ESM-1b,POSSUM, PaCRISPR, AcRanker,DeepAcr, Hidden Markov Model (HMM) , MEME anti-CRISPRdb, PaCRISPR, Acrs database, UniProt,UniParc, AlphaFold Acr protein,anti-CRISPR proteins. - Determining Anti CRISPR from the large-scale protein database [148] PredPHI Identification of phage-host interactions from sequence data K-Means clustering PhagesDB ,GenBank, 81.00% Development of personalized treatment for bacterial infections. [149] Cryo-electron tomography (cryo-ET) Phage based therapy against K. pneumoniae strains. ChimeraX,HMMER,HHPred,Phyre2 Capsid and tail fibers of phage. Receptor-binding proteins, protein gp118,protein gp119 IMOD,RoseTTAFold Functional insight into Kp24 adaptation to variable surfaces of capsulated bacterial pathogens [150] Pred-BVP-Unb Identification of BVPs within a huge volume of proteins MATLAB tool,ADBoost, KNN, Universal Protein Resource Bacteriophage virion proteins 92.54% & 83.06% accuracy on benchmark and independent datasets Designing antibacterial drugs Expedite discovery of BVP [151] Highlights in BioScience Page 10 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach Table 3. List of accessible databases for antimicrobial / phage resistant genes, receptor sequences, genomic sequences and SNPs. Databases Type of Input Sequence Format of the input sequence Purpose Developed/ Maintained by Supporting Tools/ DB Link of the website Comprehensive Antibiotic Resistance Database (CARD) Nucleotide, Amino acid FASTA SNPs, validated and curated reference sequencing details and molecular foundation of antimicrobial resistance NCBI RGI https://card.mcmaster.ca /home AMRFinder Nucleotide, Amino acid FASTA, GFF Identifies AMR genes & resistance-associated point mutations NCBI - https://www.ncbi.nlm.ni h.gov/pathogens/antimic robial- resistance/AMRFinder/ Pathosystems Resource Integration Centre (PATRIC) Nucleotide, Amino acid FASTA Large database of infectious bacterial genomic information BRCs by NIAID - http://www.patricbrc.org MEGARes Nucleotide FASTQ A metagenomics dataset consisting of AMR genes that are identified, characterized and evaluated. - AMR++ (Bioinformatics pipeline), CARD, ARG-ANNOT, LAHEY https://megares.meglab. org/ ResFinder Nucleotide FASTA, FASTQ Imports assembled contigs, data sequences or completed genomes and detects AMR genes. - web-based portal and Python script https://cge.cbs.dtu.dk/ser vices/ResFinder/ Food and Environment associated Anti-Microbial Resistance Database (FEAMR DB) - - Nonclinical test results as well as dietary and surroundings related AMR data worldwide Antimicrobial Research Lab, Department of Biotechnology, University of Mumbai CARD, NDARO, MegaRes, NCBI https://feamrudbt- amrlab.mu.ac.in/ National Databases of Antibiotic Resistant Organisms (NDARO) - - A co-operative, centrally controlled, cross-agency center where researchers may obtain AMR data to enable real-time pathogen monitoring. NCBI AMRFinder - Functional Antibiotic Resistant Metagenomic Element Database (FARMEDB) Nucleotide, Amino acid FASTA It explores functional metagenomics antibiotic resistant genetic variables, serves as an opportunity for investigating AR in the majority of bacteria that are difficult to culture in lab. It serves as a repository for globally available antibiotic resistance linked genome sequences, predicted amino acid sequences, regulatory factors, jumping genes, and predicted peptides associated with antibiotic resistant genes. University of Washington - http://staff.washington.e du/jwallace/farme/index. html LREfinder Nucleotide FASTA, FASTQ Offers information on the genes and alterations that cause enterococci to become tolerant to linezolid. Centre for Genomic Epidemiology - https://cge.cbs.dtu.dk/ser vices/LRE-finder/ Galileo AMR Nucleotide FASTA Offers quick and precise labelling of genes associated with AMR for any DNA fragment of gram-negative bacteria. Arc Bio MARA, RAC https://galileoamr.arcbio .com/mara/ ARG-miner - - Gathersandaccessallof thedata fromvariousARGresources. Itoffers proofofARGs,specificallyplasmids,viruses,orprophages, thatcanbe carriedbyMGEs. - ARDB, ARG- ANNOT, MEGARes, CARD, NDARO, ResFinder, UniProt, PATRIC https://bench.cs.vt.edu/a rgminer/#/home ShortBRED Amino acid FASTA Facilitates the highly selective characterization of target protein families and AMR genes in shotgun metagomics sequence reading information. The Huttenhower Lab, Department of Biostatistics, Havard T.H. Chan School of Public Health ARDB, CARD http://huttenhower.sph.h arvard.edu/shortbred Comprehensive β- lactamase Molecular Annotation Resource (CBMAR) Nucleotide, Amino acid FASTA The approach groups beta-lactamases into classes and then subsequently subgroups them based on factors such as gene location, phylogenetic relationships, active site, parent fingerprints, mutational characteristics, antibiotic resistance characteristics, blocker vulnerability, and nucleotide diversity. - LAHEY, PDB, UniProt, GeneBank, LacED, ARBD, http://proteininformatics .org/mkumar/lactamased b/ DeepARG Nucleotide, Amino acid FASTA, FASTQ It makes highly confident predictions about ARGs from quick reads and complete gene length sequences based on metagenomic study of environmental sources. - ARDB, CARD, UniProt https://bench.cs.vt.edu/d eeparg INTEGRALL Nucleotide FASTA gathers and arranges the integrons' data - - https://integrall.bio.ua. ppt/? Phage Receptor Database (PhReD) - FASTQ, SCF It gathers bacterial receptors that are necessary for phage-host identification and interacting associations. Bio-Conversion Databank Foundation - - The Actinobacteriophage Database - - It sequences, identities, defines and characterizes Mycobacteriophages. Department of Biologicals Sciences at the University of Pitsburg - https://phagesdb.org/ Beta-lactamase database (BLDB) Lahey Clinic database Nucleotide - It collects architectural and biochemical characteristics, along with sequencing data, for every known BL. It describes the genetic factors that give resistance to betalactam substances. Part of Bacterial Antimicrobial Resistance Reference Gene Database, NCBI PDB http://bldb.eu http://www.lahey.org/St udies/ Antibacterial Biocide and Metal Resistance Genes Database (BacMet) Nucleotide, Amino acid FASTA It aims to target genetic factor that provide resistance to metal- based substances and biocidal compounds. - - http://bacmet.biomedici ne.gu.se/ Highlights in BioScience Page 11 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach tain predictions owing to the variability from random sampling during classification. Continuously expanding the phage lifestyle database to include more known phage lifestyles will enhance prediction precision and sensitivity [153]. Furthermore, achiev- ing high-precision rates for complex classification schemes, such as host strains and phage families, remains challenging [132]. Addressing these challenges requires refining the methodology and exploring novel classification methods. It is necessary to update the tool with new features and experimentally validated data to improve its accuracy and reliability [154]. The insuffi- cient data on ssDNA and RNA phage is a significant limitation attributed to experimental constraints and resulted in the lack of respective genomic data in databases. Extensive research on phage clades and its diversity can bring much revolution in the collection of genetic information and databases [155; 156]. Fur- thermore, predicting phage-host relationships in complex micro- bial environments remains challenging because current methods often demand large amounts of homogeneous data. Develop- ing more data-efficient techniques, such as similarity networks and machine learning, is crucial for accurate predictions [152]. Despite advances in computational methods, identification of phage-host interactions remains a central challenge for effective phage therapy. The application of deep learning techniques has shown promise, but their lack of interpretability hinders user un- derstanding. Improving model interpretability while maintaining predictive performance is necessary for practical implementation. Integrating sequence similarity information and exploring novel features for phage-host interaction pairs could enhance model robustness and accuracy [129]. In conclusion, addressing the limitations and challenges of using machine learning in phage therapy, including dataset bias, outdated databases, model inter- pretability, and the need for continuous updates, will be pivotal in developing personalized therapies against antibiotic-resistant bacterial infections and in advancing the field of phage therapy. Acknowledgements I extend my sincere gratitude to the Central University of Tamil Nadu for its profound support for the successful completion of the manuscript. Reference 1. Mancuso G, Midiri A, Gerace E, Biondo C. Bacte- rial Antibiotic Resistance: The Most Critical Pathogens. Pathogens. 2021 Oct;10(10):1310. 2. Basak S, Singh P, Rajurkar M. Multidrug Resistant and Extensively Drug Resistant Bacteria: A Study. J Pathog. 2016;2016:4065603. 3. Magiorakos AP, et al. Multidrug-resistant, extensively drug- resistant and pandrug-resistant bacteria: an international expert proposal for interim standard definitions for acquired resistance. Clin Microbiol Infect. 2012 Mar;18(3):268-81. 4. Mulani MS, Kamble EE, Kumkar SN, Tawre MS, Pardesi KR. Emerging Strategies to Combat ESKAPE Pathogens in the Era of Antimicrobial Resistance: A Review. Front Microbiol. 2019 Apr;10:539. 5. Zohra T, et al. Cracking the Challenge of Antimicrobial Drug Resistance with CRISPR/Cas9, Nanotechnology and Other Strategies in ESKAPE Pathogens. Microorganisms. 2021 Apr;9(5):954. 6. Mangalea MR, Duerkop BA. Fitness Trade-Offs Re- sulting from Bacteriophage Resistance Potentiate Syn- ergistic Antibacterial Strategies. Infect Immun. 2020 Jun;88(7):e00926-19. 7. Egido JE, Costa AR, Aparicio-Maldonado C, Haas PJ, Brouns SJJ. Mechanisms and clinical importance of bacteriophage resistance. FEMS Microbiol Rev. 2021 Sep;46(1):fuab048. 8. Wright RCT, Friman VP, Smith MCM, Brockhurst MA. Cross-resistance is modular in bacteriaphage interactions. PLoS Biol. 2018 Oct;16(10):e2006057. 9. Melo LDR, Monteiro R, Pires DP, Azeredo J. Phage-Host Interaction Analysis by Flow Cytometry Allows for Rapid and Efficient Screening of Phages. Antibiotics (Basel). 2022 Jan;11(2):164. 10. Pan X, Cui X, Zhang F, He Y, Li L, Yang H. Genetic Evi- dence for O-Specific Antigen as Receptor of Pseudomonas aeruginosa Phage K8 and Its Genomic Analysis. Front Microbiol. 2016 Mar;7:252. 11. Ravin V, Räisänen L, Alatossava T. A Conserved C- Terminal Region in Gp71 of the Small Isometric-Head Phage LL-H and ORF474 of the Prolate-Head Phage JCL1032 Is Implicated in Specificity of Adsorption of Phage to Its Host, Lactobacillus delbrueckii. J Bacteriol. 2002 May;184(9):2455-9. 12. Kim M, Ryu S. Spontaneous and transient defence against bacteriophage by phase-variable glucosylation of O-antigen in Salmonella enterica serovar Typhimurium. Molecular Microbiology. 2012;86(2):411-25. 13. Sumrall ET, et al. Phage resistance at the cost of virulence: Listeria monocytogenes serovar 4b requires galactosylated teichoic acids for InlB-mediated invasion. PLoS Pathog. 2019 Oct;15(10):e1008032. 14. Dunne M, et al. Reprogramming Bacteriophage Host Range through Structure-Guided Design of Chimeric Receptor Binding Proteins. Cell Reports. 2019 Oct;29(5):1336-50.e4. 15. Dunne M, Prokhorov NS, Loessner MJ, Leiman PG. Repro- gramming bacteriophage host range: Design principles and strategies for engineering receptor binding proteins. Curr Opin Biotechnol. 2021 Apr;68:272-81. Highlights in BioScience Page 12 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach 16. Pfeifer E, Bonnin RA, Rocha EPC. Phage-Plasmids Spread Antibiotic Resistance Genes through Infection and Lyso- genic Conversion. mBio;13(5):e01851-22. 17. Monteiro R, Pires DP, Costa AR, Azeredo J. Phage Ther- apy: Going Temperate? Trends in Microbiology. 2019 Apr;27(4):368-78. 18. Pires DP, Melo LDR, Azeredo J. Understanding the Com- plex Phage-Host Interactions in Biofilm Communities. An- nual Review of Virology. 2021;8(1):73-94. 19. Smet JD, Hendrix H, den Bossche AV. Analyzing PhageHost ProteinProtein Interactions Using Strep-tagő II Purifications. In: Clokie MRJ, Kropinski A, Lavigne R, editors. Bacteriophages: Methods and Protocols, Vol- ume IV. Methods in Molecular Biology. New York, NY: Springer; 2019. p. 117-36. 20. Luong T, Salabarria AC, Edwards RA, Roach DR. Stan- dardized bacteriophage purification for personalized phage therapy. Nat Protoc. 2020 Sep;15(9):9. 21. Palma M. Aspects of Phage-Based Vaccines for Pro- tein and Epitope Immunization. Vaccines (Basel). 2023 Feb;11(2):436. 22. usiak Szelachowska M, Weber-Dbrowska B, Górski A. Bac- teriophages and Lysins in Biofilm Control. Virol Sin. 2020 Mar;35(2):125-33. 23. Sharma U, Vipra A, Channabasappa S. Phage-derived lysins as potential agents for eradicating biofilms and per- sisters. Drug Discovery Today. 2018 Apr;23(4):848-56. 24. Cahill J, Young R. Phage Lysis: Multiple Genes for Multi- ple Barriers. Adv Virus Res. 2019;103:33-70. 25. Ferry T, et al. Past and Future of Phage Therapy and Phage- Derived Proteins in Patients with Bone and Joint Infection. Viruses. 2021 Dec;13(12):2414. 26. Nang SC, et al. Pharmacokinetics/pharmacodynamics of phage therapy: a major hurdle to clinical translation. Clini- cal Microbiology and Infection. 2023 Jun;29(6):702-9. 27. Danis-Wlodarczyk K, Dbrowska K, Abedon ST. Phage Therapy: The Pharmacology of Antibacterial Viruses. Cur- rent Issues in Molecular Biology. 2021 Jan;40(1):1. 28. Zhang YZ, Liu Y, Bai Z, Fujimoto K, Uematsu S, Imoto S. Zero-shot-capable identification of phage-host relationships with whole-genome sequence representation by contrastive learning. Brief Bioinform. 2023 Jul:bbad239. 29. Brown P, et al. Whole Genome Sequence Analysis of Phage-Resistant Listeria monocytogenes Serotype 1/2a Strains from Turkey Processing Plants. Pathogens. 2021 Feb;10(2):199. 30. Zhu H, et al. Whole Genome Sequence Analysis of Lacti- plantibacillus plantarum Bacteriophage P2. Pol J Microbiol. 2022 Sep;71(3):421-8. 31. Meaden S, et al. Phage gene expression and host responses lead to infection-dependent costs of CRISPR immunity. ISME J. 2021 Feb;15(2):534-44. 32. Wong YC, et al. Phage N15-Based Vectors for Gene Cloning and Expression in Bacteria and Mammalian Cells. ACS Synth Biol. 2023 Apr;12(4):909-21. 33. Straus SK, Bo HE. Filamentous Bacteriophage Proteins and Assembly. In: Harris JR, Bhella D, editors. Virus Protein and Nucleoprotein Complexes. Subcellular Biochemistry. Singapore: Springer; 2018. p. 261-79. 34. Chen W, et al. Structural changes in bacteriophage T7 upon receptor-induced genome ejection. Proc Natl Acad Sci U S A. 2021 Sep;118(37):e2102003118. 35. Manisekhar SR, Siddesh GM, Manvi SS. Introduction to Bioinformatics. In: Srinivasa KG, Siddesh GM, Man- isekhar SR, editors. Statistical Modelling and Machine Learning Principles for Bioinformatics Techniques, Tools, and Applications. Algorithms for Intelligent Systems. Sin- gapore: Springer Singapore; 2020. p. 3-9. 36. Tang B, Pan Z, Yin K, Khateeb A. Recent Advances of Deep Learning in Bioinformatics and Computational Biol- ogy. Front Genet. 2019 Mar;10:214. 37. Villalba GC, Matte U. Fantastic databases and where to find them: Web applications for researchers in a rush. Genet Mol Biol;44(2):e20200203. 38. Statistical Modelling and Machine Learning Principles for Bioinformatics Techniques, Tools, and Applications. Algo- rithms for Intelligent Systems. Singapore: Springer Singa- pore; 2020. . 39. Furizal F, Maarif A, Rifaldi D. Application of Machine Learning in Healthcare and Medicine: A Review. Journal of Robotics and Control (JRC). 2023 Sep;4(5):5. 40. Aldoseri A, Al-Khalifa KN, Hamouda AM. Re-Thinking Data Strategy and Integration for Artificial Intelligence: Concepts, Opportunities, and Challenges. Applied Sciences. 2023 Jan;13(12):12. 41. Iskandar K, et al. Antibiotic Discovery and Resistance: The Chase and the Race. Antibiotics (Basel). 2022 Jan;11(2):182. 42. Altamirano FLG, Barr JJ. Phage Therapy in the Postan- tibiotic Era. Clinical Microbiology Reviews. 2019 Jan;32(2):10.1128/cmr.00066-18. Highlights in BioScience Page 13 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach 43. Venturini C, Fabijan AP, Lubian AF, Barbirz S, Iredell J. Biological foundations of successful bacteriophage therapy. EMBO Mol Med. 2022 May;14(7):e12435. 44. Naureen Z, et al. Bacteriophages presence in nature and their role in the natural selection of bacterial populations. Acta Biomed. 2020;91(Suppl 13):e2020024. 45. Mäntynen S, Laanto E, Oksanen HM, Poranen MM, Díaz- Muñoz SL. Black box of phagebacterium interactions: ex- ploring alternative phage infection strategies. Open Biology. 2021 Sep;11(9):210188. 46. Holmes RK. Biology and Molecular Epidemiology of Diph- theria Toxin and the tox Gene. The Journal of Infectious Diseases. 2000 Feb;181(Supplement_1):S156-67. 47. Xuan G, Lin H, Tan L, Zhao G, Wang J. Quorum Sens- ing Promotes Phage Infection in Pseudomonas aeruginosa PAO1. mBio;13(1):e03174-21. 48. Majkowska-Skrobek G, et al. Phage-Borne Depolymerases Decrease Klebsiella pneumoniae Resistance to Innate De- fense Mechanisms. Front Microbiol. 2018 Oct;9:2517. 49. Aslam S, et al. Early clinical experience of bacteriophage therapy in three lung transplant recipients. Am J Transplant. 2019 Sep;19(9):2631-9. 50. Fong SA, et al. Activity of Bacteriophages in Removing Biofilms of Pseudomonas aeruginosa Isolates from Chronic Rhinosinusitis Patients. Front Cell Infect Microbiol. 2017 Sep;7:418. 51. Ooi ML, et al. Safety and Tolerability of Bacteriophage Therapy for Chronic Rhinosinusitis Due to Staphylococ- cus aureus. JAMA Otolaryngol Head Neck Surg. 2019 Aug;145(8):723-9. 52. Drilling AJ, et al. Long-Term Safety of Topical Bacterio- phage Application to the Frontal Sinus Region. Front Cell Infect Microbiol. 2017 Feb;7:49. 53. Uyttebroek S, et al. The Potential Role of Bacteriophages in the Treatment of Recalcitrant Chronic Rhinosinusitis. Antibiotics (Basel). 2021 Jun;10(6):675. 54. Dedrick RM, et al. Phage Therapy of Mycobacterium Infections: Compassionate Use of Phages in 20 Patients With Drug-Resistant Mycobacterial Disease. Clin Infect Dis. 2022 Jun;76(1):103-12. 55. Jault P, et al. Efficacy and tolerability of a cocktail of bacte- riophages to treat burn wounds infected by Pseudomonas aeruginosa (PhagoBurn): a randomised, controlled, double- blind phase 1/2 trial. Lancet Infect Dis. 2019 Jan;19(1):35- 45. 56. Vázquez R, et al. Essential Topics for the Regulatory Con- sideration of Phages as Clinically Valuable Therapeutic Agents: A Perspective from Spain. Microorganisms. 2022 Mar;10(4):717. 57. Onsea J, et al. Bacteriophage Therapy for Difficult-to- Treat Infections: The Implementation of a Multidisciplinary Phage Task Force (The PHAGEFORCE Study Protocol). Viruses. 2021 Aug;13(8):1543. 58. Willy C, et al. Phage Therapy in GermanyUpdate 2023. Viruses. 2023 Feb;15(2):588. 59. Bretaudeau L, Tremblais K, Aubrit F, Meichenin M, Arnaud I. Good Manufacturing Practice (GMP) Compliance for Phage Therapy Medicinal Products. Front Microbiol. 2020 Jun;11:1161. 60. Liu CG, et al. Phage-Antibiotic Synergy Is Driven by a Unique Combination of Antibacterial Mechanism of Action and Stoichiometry. mBio. 2020 Aug;11(4):e01462-20. 61. Wang X, Loh B, Altamirano FG, Yu Y, Hua X, Leptihn S. Colistin-phage combinations decrease antibiotic resis- tance in Acinetobacter baumannii via changes in envelope architecture. Emerg Microbes Infect;10(1):2205-19. 62. Santajit S, Indrawattana N. Mechanisms of Antimicro- bial Resistance in ESKAPE Pathogens. Biomed Res Int. 2016;2016:2475067. 63. Hollenbeck BL, Rice LB. Intrinsic and acquired re- sistance mechanisms in enterococcus. Virulence. 2012 Aug;3(5):421-569. 64. Lee T, Pang S, Abraham S, Coombs GW. Antimicrobial- resistant CC17 Enterococcus faecium: The past, the present and the future. J Glob Antimicrob Resist. 2019 Mar;16:36- 47. 65. McGuinness WA, Malachowa N, DeLeo FR. Vancomycin Resistance in Staphylococcus aureus. Yale J Biol Med. 2017 Jun;90(2):269-81. 66. Sianipar O, Asmara W, Dwiprahasto I, Mulyono B. Mor- tality risk of bloodstream infection caused by either Escherichia coli or Klebsiella pneumoniae producing extended-spectrum -lactamase: a prospective cohort study. BMC Res Notes. 2019 Nov;12:719. 67. Lupo A, Haenni M, Madec JY. Antimicrobial Resistance in Acinetobacter spp. and Pseudomonas spp. Microbiol Spectr. 2018 Jun;6(3). 68. Li H, Luo YF, Williams BJ, Blackwell TS, Xie CM. Structure and function of OprD protein in Pseudomonas aeruginosa: From antibiotic resistance to novel therapies. Int J Med Microbiol. 2012 Mar;302(2):10.1016/j.ijmm.2011.10.001. Highlights in BioScience Page 14 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach 69. Muntean MM, et al. Phenotypic and genotypic detection methods for antimicrobial resistance in ESKAPE pathogens (Review). Exp Ther Med. 2022 Jun;24(2):508. 70. Oechslin F. Resistance Development to Bacteriophages Occurring during Bacteriophage Therapy. Viruses. 2018 Jun;10(7):351. 71. Hesse S, et al. Phage Resistance in Multidrug-Resistant Klebsiella pneumoniae ST258 Evolves via Diverse Muta- tions That Culminate in Impaired Adsorption. mBio. 2020 Jan;11(1):e02530-19. 72. Zulk JJ, et al. Phage Resistance Accompanies Reduced Fitness of Uropathogenic Escherichia coli in the Urinary Environment. mSphere. 2022 Aug;7(4):e0034522. 73. Garb J, et al. Multiple phage resistance systems inhibit infection via SIR2-dependent NAD+ depletion. Nat Micro- biol. 2022 Nov;7(11):1849-56. 74. Laure NN, Ahn J. Phage resistance-mediated trade-offs with antibiotic resistance in Salmonella Typhimurium. Mi- crob Pathog. 2022 Oct;171:105732. 75. Li N, et al. Characterization of Phage Resistance and Their Impacts on Bacterial Fitness in Pseudomonas aeruginosa. Microbiol Spectr. 2022 Oct;10(5):e0207222. 76. Seniya SP, Jain V. Decoding phage resistance by mpr and its role in survivability of Mycobacterium smegmatis. Nucleic Acids Res. 2022 Jul;50(12):6938-52. 77. Owen SV, et al. Prophages encode phage-defense sys- tems with cognate self-immunity. Cell Host Microbe. 2021 Nov;29(11):1620-33.e8. 78. Charity OJ, et al. Increased phage resistance through lyso- genic conversion accompanying emergence of monophasic Salmonella Typhimurium ST34 pandemic strain. Microb Genom. 2022 Nov;8(11):mgen000897. 79. Garnelo M, Shanahan M. Reconciling deep learning with symbolic artificial intelligence: representing objects and relations. Current Opinion in Behavioral Sciences. 2019 Oct;29:17-23. 80. Taye MM. Understanding of Machine Learning with Deep Learning: Architectures, Workflow, Applications and Fu- ture Directions. Computers. 2023 May;12(5):Art. no. 5. 81. Verma VK, Verma S. Machine learning applications in healthcare sector: An overview. Materials Today: Proceed- ings. 2022 Jan;57:2144-7. 82. Ih S. Machine Learning: Algorithms, Real-World Appli- cations and Research Directions. SN computer science. 2021;2(3). 83. Choi RY, Coyner AS, Kalpathy-Cramer J, Chiang MF, Campbell JP. Introduction to Machine Learning, Neu- ral Networks, and Deep Learning. Transl Vis Sci Tech- nol;9(2):14. 84. Hassan MM, et al. A comparative assessment of machine learning algorithms with the Least Absolute Shrinkage and Selection Operator for breast cancer detection and predic- tion. Decision Analytics Journal. 2023 Jun;7:100245. 85. Razavi S. Deep learning, explained: Fundamentals, explain- ability, and bridgeability to process-based modelling. Envi- ronmental Modelling & Software. 2021 Oct;144:105159. 86. Goodswen SJ, Barratt JLN, Kennedy PJ, Kaufer A, Calarco L, Ellis JT. Machine learning and applications in microbi- ology. FEMS Microbiol Rev. 2021 Mar;45(5):fuab015. 87. Seo YA, Park JS. Expectation-Maximization Algorithm for the Calibration of Complex Simulator Using a Gaussian Process Emulator. Entropy. 2021 Jan;23(1):Art. no. 1. 88. Carracedo-Reboredo P, et al. A review on machine learning approaches and trends in drug discovery. Comput Struct Biotechnol J. 2021 Aug;19:4538-58. 89. Manimegalai P, Kumar RS, Valsalan P, Dhanagopal R, Raj PTV, Christhudass J. 3D Convolutional Neural Net- work Framework with Deep Learning for Nuclear Medicine. Scanning. 2022;2022:9640177. 90. Serafim MSM, et al. The application of machine learn- ing techniques to innovative antibacterial discovery and development. Expert Opinion on Drug Discovery. 2020 Oct;15(10):1165-80. 91. Chandrasekaran S, et al. Chemogenomics and orthology- based design of antibiotic combination therapies. Molecular Systems Biology. 2016 May;12(5):872. 92. Cokol M, Li C, Chandrasekaran S. Chemogenomic model identifies synergistic drug combinations robust to the pathogen microenvironment. PLOS Computational Biology. 2018 Dec;14(12):e1006677. 93. Zoffmann S, et al. Machine learning-powered antibiotics phenotypic drug discovery. Sci Rep. 2019 Mar;9(1):Art. no. 1. 94. Vock I, Aguilar-Bultet L, Egli A, Tamma PD, Tschudin- Sutter S. Independent, external validation of clinical pre- diction rules for the identification of extended-spectrum -lactamase-producing Enterobacterales, University Hospi- tal Basel, Switzerland, January 2010 to December 2016. Euro Surveill. 2020 Jul;25(26):1900317. 95. Zeileis A, Hothorn T, Hornik K. Model-Based Recursive Partitioning. Journal of Computational and Graphical Statis- tics. 2008;17(2):492-514. Highlights in BioScience Page 15 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach 96. Sakagianni A, et al. Using Machine Learning to Predict Antimicrobial ResistanceA Literature Review. Antibiotics (Basel). 2023 Feb;12(3):452. 97. Noman SM, et al. Machine Learning Techniques for An- timicrobial Resistance Prediction of Pseudomonas Aerugi- nosa from Whole Genome Sequence Data. Comput Intell Neurosci. 2023 Mar;2023:5236168. 98. Feretzakis G, et al. Using Machine Learning Techniques to Aid Empirical Antibiotic Therapy Decisions in the Intensive Care Unit of a General Hospital in Greece. Antibiotics (Basel). 2020 Jan;9(2):50. 99. Cantrell JM, Chung CH, Chandrasekaran S. Machine learning to design antimicrobial combination therapies: Promises and pitfalls. Drug Discovery Today. 2022 Jun;27(6):1639-51. 100. Mason DJ, Eastman RT, Lewis RPI, Stott IP, Guha R, Ben- der A. Using Machine Learning to Predict Synergistic Antimalarial Compound Combinations With Novel Struc- tures. Front Pharmacol. 2018 Oct;9:1096. 101. Kim JI, et al. Machine Learning for Antimicrobial Resis- tance Prediction: Current Practice, Limitations, and Clini- cal Perspective. Clin Microbiol Rev;35(3):e00179-21. 102. Flandrois JP, Lina G, Dumitrescu O. MUBII-TB-DB: a database of mutations associated with antibiotic resistance in Mycobacterium tuberculosis. BMC Bioinformatics. 2014 Apr;15:107. 103. Popa SL, et al. Deep Learning and Antibiotic Resistance. Antibiotics. 2022 Nov;11(11):Art. no. 11. 104. Her HL, Wu YW. A pan-genome-based machine learn- ing approach for predicting antimicrobial resistance activ- ities of the Escherichia coli strains. Bioinformatics. 2018 Jul;34(13):i89-95. 105. Wang S, Zhao C, Yin Y, Chen F, Chen H, Wang H. A Practical Approach for Predicting Antimicrobial Pheno- type Resistance in Staphylococcus aureus Through Ma- chine Learning Analysis of Genome Data. Front Microbiol. 2022;13:841289. 106. Khaledi A, et al. Predicting antimicrobial resistance in Pseu- domonas aeruginosa with machine learningenabled molecu- lar diagnostics. EMBO Mol Med. 2020 Mar;12(3):e10264. 107. Gupta A, et al. Leapfrog dynamics in phage-bacteria co- evolution revealed by joint analysis of cross-infection phe- notypes and whole genome sequencing. Ecol Lett. 2022 Apr;25(4):876-88. 108. Knödlseder N, et al. Engineering selectivity of Cutibac- terium acnes phages by epigenetic imprinting. PLOS Pathogens. 2022 Mar;18(3):e1010420. 109. Tsao YF, et al. Phage Morons Play an Important Role in Pseudomonas aeruginosa Phenotypes. Journal of Bacteriol- ogy. 2018 Oct;200(22):10.1128/jb.00189-18. 110. Doekes HM, Mulder GA, Hermsen R. Repeated outbreaks drive the evolution of bacteriophage communication. eLife. 2021 Jan;10:e58410. 111. Cazer CL, et al. Analysis of Multidrug Resistance in Staphylococcus aureus with a Machine Learning- Generated Antibiogram. Antimicrob Agents Chemother. 2021 Mar;65(4):e02132-20. 112. Maguire F, Rehman MA, Carrillo C, Diarra MS, Beiko RG. Identification of Primary Antimicrobial Resistance Drivers in Agricultural Nontyphoidal Salmonella enterica Serovars by Using Machine Learning. mSystems. 2019 Aug;4(4):e00211-9. 113. Jahn MT, et al. Lifestyle of sponge symbiont phages by host prediction and correlative microscopy. ISME J. 2021 Jul;15(7):Art. no. 7. 114. Tadmor AD, Ottesen EA, Leadbetter JR, Phillips R. Prob- ing individual environmental bacteria for viruses by using microfluidic digital PCR. Science. 2011;333(6038):58-62. 115. Melo LD, Monteiro R, Pires DP, Azeredo J. Phage-host interaction analysis by flow cytometry allows for rapid and efficient screening of phages. Antibiotics. 2022;11(2):164. 116. Kauffman KM, et al. Resolving the structure of phagebac- teria interactions in the context of natural diversity. Nature communications. 2022;13(1):372. 117. Leskinen K, Blasdel BG, Lavigne R, Skurnik M. RNA- sequencing reveals the progression of phage-host interac- tions between R1-37 and Yersinia enterocolitica. Viruses. 2016;8(4):111. 118. Mirzaei MK, Nilsson AS. Isolation of phages for phage therapy: a comparison of spot tests and efficiency of plating analyses for determination of host range and efficacy. PLoS one. 2015;10(3):e0118557. 119. Versoza CJ, Pfeifer SP. Computational Prediction of Bacte- riophage Host Ranges. Microorganisms. 2022 Jan;10(1). 120. Villarroel J, et al. HostPhinder: a phage host prediction tool. Viruses. 2016;8(5):116. 121. Wang W, et al. A network-based integrated framework for predicting virusprokaryote interactions. NAR genomics and bioinformatics. 2020;2(2):lqaa044. 122. Galiez C, Siebert M, Enault F, Vincent J, Söding J. WIsH: who is the host? Predicting prokaryotic hosts from metagenomic phage contigs. Bioinformatics. 2017 Jul;33(19):3113-4. Highlights in BioScience Page 16 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach 123. Liu D, Ma Y, Jiang X, He T. Predicting virus-host as- sociation by Kernelized logistic matrix factorization and similarity network fusion. BMC bioinformatics. 2019;20:1- 10. 124. Liu D, Hu X, He T, Jiang X. Virus-host association pre- diction by using Kernelized logistic matrix factorization on heterogeneous networks. 2018:108-13. 125. Leite DMC, et al. Exploration of multiclass and one-class learning methods for prediction of phage-bacteria interac- tion at strain level. 2018:1818-25. 126. Zhang R, Mirdita M, Karin EL, Norroy C, Galiez C, Söd- ing J. SpacePHARER: sensitive identification of phages from CRISPR spacers in prokaryotic hosts. Bioinformatics. 2021;37(19):3364-6. 127. Roux S, Enault F, Hurwitz BL, Sullivan MB. VirSorter: mining viral signal from microbial genomic data. PeerJ. 2015;3:e985. 128. Li M, et al. A deep learning-based method for identi- fication of bacteriophage-host interaction. IEEE/ACM transactions on computational biology and bioinformatics. 2020;18(5):1801-10. 129. Aggarwal S, Dhall A, Patiyal S, Choudhury S, Arora A, Raghava GP. An ensemble method for prediction of phage- based therapy against bacterial infections. Frontiers in Microbiology. 2023;14:1148579. 130. Coclet C, Roux S. Global overview and major challenges of host prediction methods for uncultivated phages. Current Opinion in Virology. 2021;49:117-26. 131. Hockenberry AJ, Wilke CO. BACPHLIP: predicting bacte- riophage lifestyle from conserved protein domains. PeerJ. 2021;9:e11396. 132. McNair K, Bailey BA, Edwards RA. PHACTS, a com- putational approach to classifying the lifestyle of phages. Bioinformatics. 2012;28(5):614-8. 133. Yukgehnaish K, et al. PhageLeads: rapid assessment of phage therapeutic suitability using an ensemble machine learning approach. Viruses. 2022;14(2):342. 134. Sirén K, Millard A, Petersen B, Gilbert MTP, Clokie MR, Sicheritz-Pontén T. Rapid discovery of novel prophages using biological feature engineering and machine learning. NAR genomics and bioinformatics. 2021;3(1):lqaa109. 135. Jia B, et al. CARD 2017: expansion and model-centric cu- ration of the comprehensive antibiotic resistance database. Nucleic acids research. 2016:gkw1004. 136. Kaminski J, Gibson MK, Franzosa EA, Segata N, Dantas G, Huttenhower C. High-specificity targeted functional profiling in microbial communities with ShortBRED. PLoS computational biology. 2015;11(12):e1004557. 137. Doster E, et al. MEGARes 2.0: a database for classifi- cation of antimicrobial drug, biocide and metal resistance determinants in metagenomic sequence data. Nucleic acids research. 2020;48(D1):D561-9. 138. Chen L, Zheng D, Liu B, Yang J, Jin Q. VFDB 2016: hierarchical and refined dataset for big data analysis10 years on. Nucleic acids research. 2016;44(D1):D694-7. 139. Feldgarden M, et al. AMRFinderPlus and the Reference Gene Catalog facilitate examination of the genomic links among antimicrobial resistance, stress response, and viru- lence. Sci Rep. 2021 Jun;11(1):1. 140. Alcock BP, et al. CARD 2023: expanded curation, support for machine learning, and resistome prediction at the Com- prehensive Antibiotic Resistance Database. Nucleic Acids Res. 2023 Jan;51(D1):D690-9. 141. Sánchez-Osuna M, Barbé J, Erill I. Systematic In Sil- ico Assessment of Antimicrobial Resistance Dissemination across the Global Plasmidome. Antibiotics (Basel). 2023 Feb;12(2):281. 142. Alcock BP, et al. CARD 2020: antibiotic resistome surveil- lance with the comprehensive antibiotic resistance database. Nucleic Acids Res. 2020 Jan;48(D1):D517-25. 143. Sherry NL, et al. An ISO-certified genomics workflow for identification and surveillance of antimicrobial resistance. Nat Commun. 2023 Jan;14(1):1. 144. Coordinators NR. Database resources of the National Cen- ter for Biotechnology Information. Nucleic Acids Research. 2016 Jan;44(D1):D7-D19. 145. Barman RK, Chakrabarti AK, Dutta S. Prediction of Phage Virion Proteins Using Machine Learning Methods. Molecules. 2023 Feb;28(5):2238. 146. Yukgehnaish K, et al. PhageLeads: Rapid Assessment of Phage Therapeutic Suitability Using an Ensemble Machine Learning Approach. Viruses. 2022 Feb;14(2):342. 147. Leite DMC, Brochet X, Resch G, Que YA, Neves A, Peña- Reyes C. Computational prediction of inter-species rela- tionships through omics data analysis and machine learning. BMC Bioinformatics. 2018 Nov;19(14):420. 148. Li Y, et al. AcrNET: predicting anti-CRISPR with deep learning. Bioinformatics. 2023 May;39(5):btad259. Highlights in BioScience Page 17 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Mary et al., 2023 Machine Learning Mediated Advanced Phage and Antimicrobial Therapy - A Futuristic Approach 149. Li M, et al. A Deep Learning-Based Method for Identifica- tion of Bacteriophage-Host Interaction. IEEE/ACM Trans Comput Biol Bioinform. 2021;18(5):1801-10. 150. Ouyang R, et al. High-resolution reconstruction of a Jumbo- bacteriophage infecting capsulated bacteria using hyper- branched tail fibers. Nat Commun. 2022 Nov;13(1):7241. 151. Arif M, Ali F, Ahmad S, Kabir M, Ali Z, Hayat M. Pred- BVP-Unb: Fast prediction of bacteriophage Virion proteins using un-biased multi-perspective properties with recursive feature elimination. Genomics. 2020 Mar;112(2):1565-74. 152. Li M, et al. A deep learning-based method for identi- fication of bacteriophage-host interaction. IEEE/ACM transactions on computational biology and bioinformatics. 2020;18(5):1801-10. 153. Andrade-Martínez JS, et al. Computational tools for the analysis of uncultivated phage genomes. Microbiology and Molecular Biology Reviews. 2022;86(2):e00004-21. 154. Edwards RA, McNair K, Faust K, Raes J, Dutilh BE. Com- putational approaches to predict bacteriophagehost relation- ships. FEMS microbiology reviews. 2016;40(2):258-72. 155. Callanan J, Stockdale SR, Shkoporov A, Draper LA, Ross RP, Hill C. Expansion of known ssRNA phage genomes: from tens to over a thousand. Science advances. 2020;6(6):eaay5981. 156. Székely AJ, Breitbart M. Single-stranded DNA phages: from early molecular biology tools to recent revolutions in environmental microbiology. FEMS Microbiology Letters. 2016;363(6):fnw027. Highlights in BioScience Page 18 of 18 December 2023|Volume 6 http://bioscience.highlightsin.org/ Abstract Background Phage therapy Antimicrobial and phage resistance Machine learning Machine Learning and Antimicrobial Resistance Machine Learning in phage therapy Databases in antimicrobial resistance and phage research Challenges and Opportunities Acknowledgements