Systems and methods for predicting bacteriophage host range
A machine learning system predicts bacteriophage susceptibility by analyzing protein-protein interactions, addressing the inefficiencies of traditional methods and enabling faster identification of effective bacteriophages for bacterial infection treatment.
Patent Information
- Application Number
- PCT/IB2023/054805
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-09
- Publication Date
- 2024-11-14
AI Technical Summary
Current methods for identifying the host range of bacteriophages are laborious, time-consuming, and resource-intensive, limiting the efficiency of phage therapy in treating bacterial infections due to reliance on empirical laboratory techniques, which are not suitable for rapid clinical applications.
A machine learning-based system that predicts the susceptibility of bacterial isolates to bacteriophage infection using protein-protein interaction data, comprising an input module for generating interaction scores, a training module for building predictive models, and a prediction module for determining host range and antibacterial activity.
This approach significantly reduces the time and resources required to determine host range, enabling faster identification of effective bacteriophages for clinical and veterinary applications by automating the prediction of bacteriophage-bacteria interactions.
Smart Images

Figure IB2023054805_14112024_PF_FP_ABST
Abstract
Description
SYSTEMS AND METHODS FOR PREDICTING BACTERIOPHAGE HOST RANGEFIELD
[0001] The present disclosure relates to the field of antibacterial agents and machine learning systems.BACKGROUND
[0002] The use of antibiotics has been one of the main methods used by modern medicine to cure life-threatening bacterial infections. However, the ability of bacteria to mutate in a short time allows the development of antibiotic-resistant bacterial strains. In recent years, the proliferation of antibiotic-resistant bacteria has become one of the most challenging and serious problems for public health. The increase of treatments with these drugs and the inadequate use of them to treat non-bacterial diseases are the main cause of this emergency.
[0003] Faced with the potential threat of the rapid emergence and spread of multidrugresistant microorganisms, alternative treatments against bacterial infections are being developed. Bacteriophage viruses are a natural alternative to combat bacterial infections, as they are capable of infecting and killing bacteria throughout their life cycle. Unlike antibiotics, bacteriophages have cohabited and evolved with bacteria, controlling the bacterial population in a continuous genetic exchange. The possibility of harnessing bacteriophages as antibacterial agents was investigated after their initial isolation in the early 19th century, however, their study and use was abandoned after the discovery of antibiotics.
[0004] The use of bacteriophages against pathogenic bacteria, or phage therapy, has become a tool of growing interest due to its adaptive advantages, related to the fact that bacteriophages are able to invade only and exclusively bacteria in a specific range of infection, so that each bacteriophage can recognize a small and specific range of bacterial strains. The spectrum of bacterial strains that a given bacteriophage can infect is called "host range".
[0005] Identifying the host range corresponds to the first challenge in the efficient development of phage therapy. With this, it is possible to identify the bacteriophages necessary to kill or inhibit the growth of a given bacterium to allow its clinical and / or veterinary application and ensure the success of therapeutic use. Currently, bacteriophage-bacteria interactions and host range areidentified through empirical laboratory techniques, specifically by culture-based methods. For example, the ability of bacteriophages to lyse or inhibit bacterial growth is examined by culturing the host bacterium with the bacteriophage using liquid cultures or semi-solid bacterial lawns on agar media. Regardless of the method used, these techniques are laborious, time-consuming and resourceintensive tasks, carried out by one-to-one experiments and dependent on manual and visual inspection by the user, leading to greater variability, as it will depend on the experience and knowledge of each researcher. In clinical applications, the time to find the right bacteriophage is a critical factor and delay in obtaining the results can lead to delayed treatment and complications for the subject in need of treatment. Furthermore, only 1% of bacterial hosts have been successfully cultured in the laboratory, limiting the detection and characterization of the bacteriophage-bacteria relationship.
[0006] The use of computational approaches to predict these interactions has the advantage of being faster, automatable and less resource intensive. Therefore, there is a growing demand for systems that allow modeling of genomic data to predict bacteriophage-bacteria interactions without the need for extensive experimentation.
[0007] Although several in-silico methods have been proposed to identify the host range of a bacteriophage based on the prediction of bacteriophage-bacteria interactions using information contained in public databases, it is necessary to incorporate and test these methods on an industrial scale and with data obtained under real production conditions, which also allow the prediction of bacteriophage-bacteria interaction at the strain level, necessary for effective clinical and / or veterinary applications.
[0008] Therefore, there is still a need to develop systems and methods for the prediction of the host range of a bacteriophage, such as a machine learning based approach that allows the identification of sensitivity at bacterial strain level and which can be used in real production conditions in several applications.SUMMARY
[0009] The present disclosure relates to the field of machine learning systems for the prediction of the host range of a bacteriophage.
[0010] In some aspects, the present disclosure provides a system and method based on machine learning for the prediction of susceptibility of a bacterial isolate to infection by a bacteriophage, using information obtained from protein-protein interaction data of a bacteriophage and bacterial isolates. This system may aim to reduce the number of tests performed to determine the host range of a bacteriophage and to accelerate the study of bacteriophage-bacterial isolate interactions. The system considers the information of protein-protein relationships between bacterial isolates and bacteriophage to predict susceptibility of a bacterial isolate to infection by a bacteriophage.
[0011] Additional aspects and advantages of the present disclosure will become readily apparent to those skilled in this art from the following detailed description, wherein only illustrative embodiments of the present disclosure are shown and described. As will be realized, the present disclosure is capable of other and different embodiments, and its several details are capable of modifications in various obvious respects, all without departing from the disclosure. Accordingly, the drawings and description are to be regarded as illustrative in nature, and not as restrictive.INCORPORATION BY REFERENCE
[0012] All publications, patents, and patent applications mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference. To the extent publications and patents or patent applications incorporated by reference contradict the disclosure contained in the specification, the specification is intended to supersede and / or take precedence over any such contradictory material.BRIEF DESCRIPTION OF THE DRAWINGS
[0013] The novel features of the invention are set forth with particularity in the appended claims. A better understanding of the features and advantages of the present invention will be obtained by reference to the following detailed description that sets forth illustrative embodiments, in which the principles of the invention are utilized, and the accompanying drawings (also “figure” and “FIG.” herein), of which:
[0014] Figure 1 illustrates a diagram of an exemplary embodiment of the system according to the present disclosure.
[0015] Figure 2 illustrates a diagram of various stages and elements of an example of an input module of a system according to the present disclosure.
[0016] Figure 3 illustrates a diagram of various stages and elements of an example of a training module of a system according to the present disclosure.
[0017] Figure 4 illustrates a diagram of various stages and elements of an example of a prediction module of a system according to the present disclosure.DETAILED DESCRIPTION OF THE INVENTION
[0018] While various embodiments of the invention have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only. Numerous variations, changes, and substitutions may occur to those skilled in the art without departing from the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed.
[0019] The present disclosure provides a system and method based on machine learning models and protein-protein interaction data between bacteriophages and bacterial isolates, for predicting susceptibility of a bacterial isolate to infection by bacteriophages. The described system allows to optimally determine the host range of a bacteriophage and to predict in silico its antibacterial activity, considering several features, among which can be considered, but not limited to, protein-protein interactions between bacteriophages and bacterial isolates. Therefore, the described system aims to provide a solution that decreases the time to obtain the host range of a bacteriophage with bacterial isolates, by harnessing information of protein-protein interactions between a bacteriophage and bacterial isolates.
[0020] In some embodiments, the system comprises the following modules:1. input module, which receives genomic data from bacterial isolates and from one bacteriophage and determines their protein-protein interactions to obtain a matrix of interaction scores between bacterial isolates and the bacteriophage;2. training module, which receives the interaction score matrix obtained by the input module and data from in vitro experiments to evaluate the individual bacteriophage- bacterial isolate lytic susceptibility to build predictive model of the interactions ofbacterial isolates and the bacteriophage, and select the best models for predicting interactions based on prediction metrics;3. prediction module, which uses the predictive model selected by the training module to obtain prediction results of the interaction between a bacterial isolate and the bacteriophage.
[0021] Figure 1 illustrates a diagram of an embodiment of the system of the present disclosure.
[0022] In some embodiments, the system considers features related to in vitro experimental data to assess individual bacteriophage-bacterial isolate lytic susceptibility. In a preferred embodiment, the in vitro experimental data refer to results of challenge between bacterial isolates and bacteriophages obtained from experiments in liquid or semi-solid culture media on agar media.
[0023] Figure 2 illustrates a diagram of an embodiment of the input module of a system of the present disclosure. In some embodiments, the input module comprises the following steps: a) receiving genomes of bacterial isolates and a bacteriophage; b) annotating the genomes of bacterial isolates and bacteriophage; c) identifying and annotate domains in the amino acid sequences of bacterial isolates and the bacteriophage; d) constructing a bacterial pangenome (clusters of orthologous genes in the analyzed genomes) based on the genomes of bacterial isolates; e) cross-referencing the domain-domain interactions of bacterial isolates and bacteriophage and assign domain-domain interaction scores for each bacterial isolate; f) assigning a protein-protein interaction score based on the domain-domain interaction scores between bacterial isolates and the bacteriophage; g) assigning an interaction score to each pangenome cluster; and h) obtaining a matrix of interaction scores between bacterial pangenome clusters and the genome of bacterial isolates.
[0024] In some embodiments, the input module receives genomes of bacterial isolates and a bacteriophage from a database or from assembled sequencing data.
[0025] In some embodiments, the database used is selected from external free-to-use databases and internal databases. In another embodiment, the external databases are selected from NCBI GenBank, European Nucleotide Archive (ENA), PATRIC (Pathosystems Resource Integration Center), IMG, JGI Genome Portal, or any other suitable databases.
[0026] In some embodiments, in step b) of the input module, the assembled genomes of bacterial isolates and bacteriophage are annotated using appropriate bioinformatics tools, such as Bakta, Prokka, Pharokka, RAST, BASys, IMG-ER, Prokaryotic Genome Annotation Pipeline (PGAP), Artemis, DRAM (Distilled and Refined Annotation of Metabolism), GeneMark, Prodigal, Glimmer, AUGUSTUS, MetaGeneAnnotator, Phanotate, GeneMarkS, GlimmerHMM, MetaGeneAnnotator, Phage_Finder, ProphET, PHANOTATE-PRO, or any other appropriate bioinformatics tools.
[0027] In some embodiments, to identify protein-protein interactions, the input module uses protein domain databases that provide a complete and accurate classification of protein families and domains which are known as Pfam domains. In another embodiment, the databases are selected from Pfam, Interpro, Cath / Gene3D, Hamap, Panther, Tigrfam, Smart, or any other suitable databases.
[0028] In some embodiments, in step c) of the input module, the protein domains present in the amino acid sequences of bacterial isolates and bacteriophages are identified using appropriate bioinformatics tools, such as HMMER, InterProScan, PfamScan, Phyre2, SMART (Simple Modular Architecture Research Tool), CDD (Conserved Domain Database), PANTHER (Protein ANalysis THrough Evolutionary Relationships) or any other appropriate bioinformatics tools.
[0029] In some embodiments, in step d) of the input module, pangenomes of bacterial isolates are constructed using appropriate bioinformatics tools, such as Panaroo, Roary, PanX, PGAP (Pangenome Analysis Pipeline), GET_HOMOLOGUES, BPGA (Bacterial Pan Genome Analysis), PanCGH, chewBBACA, or any other appropriate bioinformatics tools.
[0030] In some embodiments, in step e) of the input module, the interactions between each identified protein domain of bacterial isolates and bacteriophages are cross- referenced using freely available external databases and internal databases possessing experimentally confirmed and / or predicted protein domain-domaininteraction information. In a preferred embodiment of the invention, such databases provide a quality score associated with each domain-domain interaction that measures the level of certainty of the interaction.
[0031] In some embodiments, in step f) of the input module, the score of the proteinprotein interactions between the bacterial isolates and the bacteriophage is obtained from a proper combination of the scores of their domain-domain interactions. In a preferred embodiment, a proper combination of the scores may comprise the sum of the scores. In such embodiments, in the case that no interactions are identified between the domains of a protein of a bacterial isolate with the domains of a protein of the bacteriophage, the score associated with such protein-protein interaction is 0.
[0032] In some embodiments, in step g) of the input module, an interaction score is assigned to the gene clusters of the bacterial isolates pangenomes. In a preferred embodiment, the interaction score of each cluster is obtained from a proper combination of the protein-protein interaction scores between bacterial isolates and the bacteriophage of each gene belonging to the cluster. In a preferred embodiment, a proper combination of the scores may comprise the sum of the scores.
[0033] Figure 3 illustrates a diagram of an embodiment of the training module of a system of the present disclosure. In some embodiments of the present disclosure, the training module comprises the following steps: a) receiving the interaction score matrix obtained in the input module; b) obtaining data from results of in vitro experiments to assess individual bacteriophage-bacterial isolate lytic susceptibility and associating the bacterial isolates genomes to the results of the in vitro experiments; c) building predictive model of bacterial isolate-bacteriophage interactions using machine learning models; and, d) selecting the best predictive model based on prediction metrics.
[0034] In some embodiments, the in vitro experiments refer to results of challenge between bacterial isolates and bacteriophages obtained from experiments in liquid or semi-solid culture media on agar media.
[0035] In some embodiments, in the training module of the present system, predictive models are constructed using information from protein-protein interaction scores of each isolate, together with the results of in vitro experiments to assessindividual bacteriophage-bacterial isolate lytic susceptibility. In a preferred embodiment, the in vitro experiments refer to results of challenge between bacterial isolates and bacteriophages obtained from experiments in liquid or semi-solid culture media on agar media.
[0036] In some embodiments, data from results of in vitro experiments to assess individual bacteriophage-bacterial isolate lytic susceptibility are translated into binary variables, where 0 equals resistance, or no lysis, or lysis less than a threshold, and 1 equals sensitivity or partial lysis greater than the threshold, or total lysis. In a preferred embodiment, the threshold may be 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40% of lysis. As will be obvious to those skilled in the art, this threshold value may be replaced by any other numerical value between 0% and 40% of lysis.
[0037] In another embodiments, the data from results of in vitro experiments to assess individual bacteriophage-bacterial isolate lytic susceptibility are translated into continuous variables, which identify the percentage of lysis of the challenge between the bacterial isolates and the bacteriophage, where 0% equals resistance or no lysis and 100% corresponds to sensitivity or total lysis.
[0038] In some embodiments, in the training module, predictive models are built using appropriate machine learning models, such as Random Forest (RF), Gradient Boost (GB), Support Vector Machines (SVM), k-Nearest Neighbors (k-NN), Neural Networks (NNs), Bayesian Regression (BR), Logistic Regression (LR), Decision Trees (DTs), and other appropriate machine learning related models.
[0039] In some embodiments, the training module evaluates the performance of the models using prediction metrics, such as accuracy (percentage of agreement between the model prediction and the results of in vitro experiments), precision (percentage of samples with predicted sensitivity to the bacteriophage that match the result of in vitro experiments), sensitivity (percentage of samples with predicted sensitivity to the bacteriophage that the model is able to identify), f- score (harmonic mean between precision and sensitivity) and specificity (percentage of resistant samples that the model is able to identify), between the predictions of each machine learning model and the results obtained in the in vitro experiments and selects the appropriate machine learning models for the prediction of the interactions between a bacteriophage and bacterial isolates.
[0040] In some embodiments, the training module selects the most appropriate models based on the best results obtained in the evaluated prediction metrics. In one embodiment of the invention, the training module selects the most appropriate models based on the average of the accuracy, f-score and specificity metrics.
[0041] Figure 4 illustrates a diagram of an embodiment of the prediction module of a system of the present disclosure. In some embodiments of the present disclosure, the prediction module comprises the following steps: a) receiving the genome of a bacterial isolate; b) annotating the genome of the bacterial isolate; c) identifying and annotating protein domains in the amino acid sequences of the bacterial isolate; d) cross-referencing the domain-domain interactions of the bacterial isolate with the bacteriophage and assign a domain-domain interaction score; e) obtaining a protein-protein interaction score based on the scores of domaindomain interactions between the bacterial isolate and the bacteriophage; f) associating, based on sequence identity, each gene of the isolate to a cluster of the pangenome created in the training module; g) assigning an interaction score to each pangenome cluster based on the scores associated with each bacterial isolate gene; h) obtaining a prediction score for the interaction between the bacterial isolate and the bacteriophage using the predictive model selected in the training module.
[0042] In some embodiments, the machine learning models used in the prediction module are used individually. In another embodiments, the machine learning models are used as ensemble for obtaining the prediction result.
[0043] In some embodiments, in the prediction module, the prediction results are obtained from the combination of the result of several models. In another embodiment of the invention, the prediction result is obtained by considering the individual results of each model and choosing the most repeated result among the different models, considering the probability of the prediction in each model.
[0044] In some embodiments, the system prediction results are presented as binary variables, where 0 equals resistance, or no lysis, or lysis less than a threshold, and 1 equals sensitivity or partial lysis greater than the threshold, or total lysis.In a preferred embodiment, the threshold may be 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40% of lysis. As will be obvious to those skilled in the art, this threshold value may be replaced by any other numerical value between 0% and 40% of lysis.
[0045] In another embodiment of the invention, the prediction results of the system are presented as continuous variables, where 0% equals resistance or no lysis and 100% corresponds to sensitivity or total lysis.
[0046] Optionally, in some embodiments of the present disclosure, there is a feedback module comprising the following steps: a) visually presenting the interaction predictions between bacterial isolates and bacteriophages obtained; b) selecting bacteriophages of interest; c) directing in vitro experiments using the selected bacteriophages to evaluate the individual bacteriophage-bacterial isolate lytic susceptibility; d) receiving the feedback information to readjust parameters.
[0047] This feedback module has the following objectives:Increase human analysis capabilities in order to provide users with the maximum information to select the most suitable bacteriophages.Request new in vitro experiments to the user for those predictions that are feasible but for which the system has a low reliability.Allow the user to change the parameters of the methods to correct the predictions if necessary based on what is observed in the in vitro experiments.
[0048] In some embodiments, the present disclosure provides the use of the system for predicting the susceptibility of a bacterial isolate to infection by a bacteriophage.
[0049] In said embodiment of the invention, the use of the system comprises the following steps: a) receiving the genome of a bacterial isolate; b) annotating the genome of the bacterial isolate; c) identifying and annotating protein domains in the amino acid sequences of the bacterial isolate; d) cross-referencing the domain-domain interactions of the bacterial isolate with the bacteriophage and assign a domain-domain interaction score;e) obtaining a protein-protein interaction score based on the scores of domaindomain interactions between the bacterial isolate and the bacteriophage; f) associating, based on sequence identity, each gene of the isolate to a cluster of the pangenome created in the training module; g) assigning an interaction score to each pangenome cluster based on the scores associated with each bacterial isolate gene; and, h) obtaining a prediction score for the interaction between the bacterial isolate and the bacteriophage using the predictive model selected in the training module.
[0050] In another embodiment, the present disclosure provides a method based on machine learning models and protein-protein interaction data between bacteriophages and bacterial isolates, for predicting susceptibility of a bacterial isolate to infection by bacteriophages. The described method allows to optimally determine the host range of a bacteriophage and to predict in silico its antibacterial activity, considering several parameters, among which can be considered, but not limited to, protein-protein interactions between bacteriophages and bacterial isolates.
[0051] In some embodiments, the method comprises the following stages: a) obtaining genomic data from bacterial isolates and from one bacteriophage and determining protein-protein interactions between bacterial isolates and the bacteriophage; b) obtaining protein-protein interactions data and data from in vitro experiments to assess individual bacteriophage-bacterial isolate lytic susceptibility; building predictive model of the interactions of bacterial isolates and the bacteriophage and selecting the best models for predicting interactions based on prediction metrics; c) using the predictive model selected in the previous step to obtain a prediction result of the interaction between a bacterial isolate and the bacteriophage.
[0052] In some embodiments, in stage a) of the method, the in vitro experiments refer to results of challenge between bacterial isolates and bacteriophages obtained from experiments in liquid or semi-solid culture media on agar media.
[0053] Optionally, in some embodiments of the present disclosure, there is a feedback stage comprising the following steps:a) visually presenting the interaction predictions between bacterial isolated and bacteriophages obtained in the prediction module; b) selecting bacteriophages of interest; c) directing in vitro experiments using the selected bacteriophages to evaluate the individual bacteriophage-bacterial isolate lytic susceptibility; and d) receiving the feedback information to readjust parameters.
[0054] While preferred embodiments of the present invention have been shown and described herein, it will be obvious to those skilled in the art that such embodiments are provided by way of example only. It is not intended that the invention be limited by the specific examples provided within the specification. While the invention has been described with reference to the aforementioned specification, the descriptions and illustrations of the embodiments herein are not meant to be construed in a limiting sense. Numerous variations, changes, and substitutions will now occur to those skilled in the art without departing from the invention. Furthermore, it shall be understood that all aspects of the invention are not limited to the specific depictions, configurations or relative proportions set forth herein which depend upon a variety of conditions and variables. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in practicing the invention. It is therefore contemplated that the invention shall also cover any such alternatives, modifications, variations or equivalents. It is intended that the following claims define the scope of the invention and that methods and structures within the scope of these claims and their equivalents be covered thereby.
[0055] EXAMPLES
[0056] Example 1. Host range prediction of a Salmonella bacteriophage with Salmonella isolates.
[0057] The predictive ability of the method was evaluated by analyzing the interactions of a Salmonella bacteriophage with Salmonella bacterial isolates.
[0058] Genome data of 480 Salmonella isolates and one bacteriophage and results of the challenge between bacterial isolates and the bacteriophage in a liquid culture medium were used. For this experiment, the bacterial isolates were cultured with a suspension of the bacteriophage in TSB medium bringing the mixture to a final volume of 200 pL, the mixtures were added to a 96-well plate and incubated at37°C using BioTek's EPOCH 2 spectrophotometer equipment; 12 hours after the challenge occurred, the absorbance of each sample was measured.
[0059] A total of 4,511 Salmonella pangenome clusters presented interactions with score >0 with the proteins of this bacteriophage, data that were used to build the machine-learning models. Four machine learning approaches (k-Nearest Neighbors, Random Forests, Support Vector Machines, Artificial Neural Networks) were explored individually and as a model ensemble. The analysis of the models showed that the combination of the k-Nearest Neighbors and Random Forests models, using a binary variable as output (0 and 1), presented better prediction metrics for isolate resistance / sensitivity, obtaining 88% prediction accuracy, 89% precision, 93% sensitivity, 91% f-score and 81% specificity with the validation set, which corresponds to 20% of the data not used in the training module.
[0060] Table 1. Summary of results of machine-learning models generated from bacteriophage-bacteriophage protein interaction data.
[0061] Example 2. Host range prediction of six Salmonella bacteriophages with Salmonella isolates.
[0062] The predictive ability of the method was evaluated by analyzing the interactions of 6 different Salmonella bacteriophages with Salmonella bacterial isolates.
[0063] Genome data of Salmonella isolates and six bacteriophages and results of the challenge between bacterial isolates and bacteriophages in a liquid culture medium were used. For this experiment, the bacterial isolates were cultured with a suspension of each bacteriophage in TSB medium bringing the mixture to a final volume of 200 pF, the mixtures were added in a 96-well plate and incubated at 37°C using BioTek's EPOCH 2 spectrophotometry equipment; after 12 hours of the challenge, the absorbance of each sample was measured.
[0064] Three machine learning approaches were explored to measure the lytic activity as a binary target (0 and 1) (k-Nearest Neighbors, Random Forests and Support Vector Machines), individually and as a model ensemble, using a voting strategy which considers the probability associated with each model. Three machine learning approaches were explored using as output the percentage of lysis of each isolate after the challenge (Random Forest (RF), Gradient Boost (GB) and Support Vector Machines (SVM)).
[0065] The results of the best model in each case are shown in Table 2. Since the type of model with the best prediction varies depending on the bacteriophage, regardless of the identity between their genomic sequences it is necessary to generate a new model to generate predictions for a new bacteriophage.
[0066] Table 2. Machine-learning models with better prediction metrics for bacteriophages.
[0067] Example 3. Host range prediction of an E. coli bacteriophage with bacterial isolates.
[0068] The predictive ability of the method was evaluated by analyzing the interactions of 3 different E. coli bacteriophages with E. coli bacterial isolates.
[0069] Genome data of E. coli isolates and three bacteriophages, and results of the challenge between bacterial isolates and each bacteriophage in a liquid culture medium were used. For this experiment, the bacterial isolates were cultured with a suspension of the bacteriophage in TSB medium bringing the mixture to a final volume of 200 pL, the mixtures were added in a 96-well plate and incubated at 37°C using BioTek's EPOCH 2 spectrophotometry equipment; after 12 hours of the challenge, the absorbance of each sample was measured.
[0070] Three machine learning approaches were explored, using as output binary variables (0 and 1), (k-Nearest Neighbors, Random Forests and Support Vector Machines), individually and as an ensemble. Three machine learning approaches were explored using as output the percentage of lysis of each isolate after the challenge (Random Forest (RF), Gradient Boost (GB) and Support Vector Machines (SVM)).
[0071] The results of the best model in each case are shown in Table 3. The type of model with the best prediction metrics varies depending on the bacteriophage and the type of variable used as output, so it is necessary to generate a new model when it is required to generate predictions for a new bacteriophage.
[0072] Table 3. Machine-learning models with better prediction metrics for bacteriophages.
[0073] Example 4. Automatic prediction of the sensitivity of a bacterial isolate of Salmonella and E. coli to a bacteriophage.
[0074] The model resulting from the training module of each phage was used to predict the phenotype of 4 Salmonella isolates and 4 E. coli isolates against phages 2 Salmonella bacteriophages and 2 E. coli bacteriophages. The comparison between the percentage of lysis observed in vitro and the prediction of sensitivity or resistance of the machine learning model is described in Table 4. In all cases, the prediction of the model agrees with the experimental result.
[0075] Table 4. Percentage of lysis observed in vitro of Salmonella and E. coli isolates against 4 bacteriophages, and prediction of sensitivity (1) or resistance (0) from the machine learning model trained with protein-protein interaction data.
[0076] The results demonstrated that the system was able to automatically predict the sensitivity of different bacterial isolates to bacteriophages having high concordance between predicted lytic score and observed in vitro lytic score and indicates that the system may be used to contribute to the prediction of susceptibility of a bacterial isolate to infection by a bacteriophage.
Claims
CLAIMSWHAT IS CLAIMED IS:
1. A machine learning model-based system for predicting susceptibility of a bacterial isolate to infection by a bacteriophage, wherein the system comprising: an input module configured to receive genomic data from bacterial isolates and from one bacteriophage and determine their protein-protein interactions to obtain a matrix of interaction scores between bacterial isolates and the bacteriophage; a training module configured to receive the interaction score matrix obtained by the input module and data from in vitro experiments to evaluate the individual bacteriophage-bacterial isolate lytic susceptibility to build predictive model of the interactions of bacterial isolates and the bacteriophage and select the best models for predicting interactions based on prediction metrics; and a prediction module configured to use the predictive model selected by the training module to obtain prediction results of the interaction between a bacterial isolate and the bacteriophage.
2. The system of claim 1, wherein the training module considers parameters related to in vitro experimental data to assess individual bacteriophage-bacterial isolate lytic susceptibility.
3. The system of claim 2, wherein the in vitro experimental data relate to results of challenge between bacterial isolates and bacteriophages obtained from experiments in liquid or semi-solid culture media on agar media4. The system of claim 1, wherein the input module is further configured to perform at least: a) receiving genomes of bacterial isolates and a bacteriophage; b) annotating the genomes of bacterial isolates and bacteriophage; c) identifying and annotate protein domains in the amino acid sequences of bacterial isolates and the bacteriophage; d) constructing a bacterial pangenome based on the genomes of bacterial isolates; e) cross-referencing the domain-domain interactions of bacterial isolates and bacteriophage and assign domain-domain interaction scores for each bacterial isolate; f) assigning a protein-protein interaction score based on the domain-domain interaction scores between bacterial isolates and the bacteriophage; g) assigning an interaction score to each pangenome cluster; andh) obtaining a matrix of interaction scores between bacterial pangenome clusters and the genome of bacterial isolates.
5. The system of claim 4, wherein the input module is further configured to receive genomes of bacterial isolates and a bacteriophage from a database or from assembled and annotated sequencing data.
6. The system of claim 5, wherein the database is selected from external free-to-use databases and internal databases.
7. The system of claim 4, wherein the interactions between each identified protein domain of bacterial isolates and bacteriophages are cross-referenced using freely available external databases and internal databases possessing experimentally confirmed and / or predicted protein domain-domain interaction information.
8. The system of claim 7, wherein the databases provide a quality score associated with each domain-domain interaction that measures the level of certainty of the interaction.
9. The system of claim 4, wherein the score of the protein-protein interactions between the bacterial isolates and the bacteriophage is obtained from a combination of their domain-domain interactions scores.
10. The system of claim 4, wherein the combination of the scores comprises the sum of their domain-domain interactions scores.
11. The system of claim 4, wherein an interaction score is assigned to the gene clusters of the bacterial isolates pangenomes.
12. The system of claim 11, wherein the interaction score of each cluster is obtained from a combination of the protein-protein interaction scores between bacterial isolates and the bacteriophage, of each gene belonging to the cluster.
13. The system of claim 12, wherein the combination of the scores comprises the sum of the protein-protein interaction scores.
14. The system of claim 1 , wherein the training module is further configured to perform at least: a) receiving the interaction score matrix obtained in the input module; b) obtaining data from results of in vitro experiments to assess individual bacteriophage- bacterial isolate lytic susceptibility and associating the bacterial isolates genomes to the results of the in vitro experiments; c) building predictive model of bacterial isolate-bacteriophage interactions using machine learning models; and, d) selecting the best predictive model based on prediction metrics.
15. The system of claim 14, wherein the predictive models are constructed using information from protein-protein interaction scores of each isolate, together with the results of in vitro experiments to assess individual bacteriophage-bacterial isolate lytic susceptibility.
16. The system of claim 14, wherein the data from results of in vitro experiments are translated into binary variables, where 0 equals resistance, or no lysis and 1 equals sensitivity or total lysis.
17. The system of claim 14, wherein the data from results of in vitro experiments are translated into continuous variables, which identify the percentage of lysis of the challenge between the bacterial isolates and the bacteriophage, where 0% equals resistance or no lysis and 100% corresponds to sensitivity or total lysis.
18. The system of claim 14, wherein the predictive models are built using appropriate machine learning models, such as Random Forest (RF), Gradient Boost (GB), Support Vector Machines (SVM), k-Nearest Neighbors (k-NN), Neural Networks (NNs), Bayesian Regression (BR), Logistic Regression (LR), Decision Trees (DTs), and other appropriate machine learning related models.
19. The system of claim 14, wherein the training module evaluates the performance of the models using prediction metrics between the predictions of each machine learning model and the results obtained in the in vitro experiments.
20. The system of claim 19, wherein the predictions metrics comprises accuracy, precision, sensitivity, f-score and specificity.
21. The system of claim 20, wherein the training module selects the most appropriate models based on the best results obtained in the evaluated prediction metrics.
22. The system of claim 21, wherein the training module selects the most appropriate models based on the average of the accuracy, f-score and specificity metrics.
23. The system of claim 1, wherein the prediction module is further configured to perform at least: a) receiving the genome of a bacterial isolate; b) annotating the genome of the bacterial isolate; c) identifying and annotating protein domains in the amino acid sequences of the bacterial isolate; d) cross-referencing the domain-domain interactions of the bacterial isolate with the bacteriophage and assign a domain-domain interaction score; e) obtaining a protein-protein interaction score based on the scores of domain-domain interactions between the bacterial isolate and the bacteriophage;f) associating, based on sequence identity, each gene of the isolate to a cluster of the pangenome created in the training module; g) assigning an interaction score to each pangenome cluster based on the scores associated with each bacterial isolate gene; h) obtaining a prediction score for the interaction between the bacterial isolate and the bacteriophage using the predictive model selected in the training module.
24. The system of claim 23, wherein the machine learning models are used individually.
25. The system of claim 23, wherein the machine learning models are as ensemble.
26. The system of claim 23, wherein the prediction results are obtained from the combination of the result of several machine learning models.
27. The system of claim 26, wherein the prediction results are obtained by considering the individual results of each model and choosing the most repeated result among the different models, considering the probability of the prediction in each model.
28. The system of claim 26, wherein the prediction results are presented as binary variables, where 0 equals resistance, or no lysis, and 1 equals sensitivity or total lysis.
29. The system of claim 26, wherein the prediction results are presented as continuous variables, where 0% equals resistance or no lysis and 100% corresponds to sensitivity or total lysis.
30. The system of claim 1 , wherein further comprises a feedback module configured to perform at least: a) visually presenting the interaction predictions between bacterial isolated and bacteriophages obtained in the prediction module; b) selecting bacteriophages of interest; c) directing in vitro experiments using the selected bacteriophages to evaluate the individual bacteriophage-bacterial isolate lytic susceptibility; and, d) receiving the feedback information to readjust parameters.
31. Use of the machine learning model-based system for predicting susceptibility of a bacterial isolate to infection by a bacteriophage, the use comprising: a) receiving the genome of a bacterial isolate; b) annotating the genome of the bacterial isolate; c) identifying and annotating protein domains in the amino acid sequences of the bacterial isolate; d) cross-referencing the domain-domain interactions of the bacterial isolate with the bacteriophage and assign a domain-domain interaction score;e) obtaining a protein-protein interaction score based on the scores of domain-domain interactions between the bacterial isolate and the bacteriophage; f) associating, based on sequence identity, each gene of the isolate to a cluster of the pangenome created in the training module; g) assigning an interaction score to each pangenome cluster based on the scores associated with each bacterial isolate gene; and, h) obtaining a prediction score for the interaction between the bacterial isolate and the bacteriophage using the predictive model selected in the training module.
32. A machine learning-based method for predicting susceptibility of a bacterial isolate to infection by a bacteriophage, the method comprising: a) obtaining genomic data from bacterial isolates and from one bacteriophage and determining protein-protein interactions between bacterial isolates and the bacteriophage; b) obtaining protein-protein interactions data and data from in vitro experiments to assess individual bacteriophage-bacterial isolate lytic susceptibility; building predictive models of the interactions of bacterial isolates and the bacteriophage and selecting the best models for predicting interactions based on prediction metrics; c) using the predictive models selected in the previous step to obtain a prediction result of the interaction between a bacterial isolate and the bacteriophage.
33. The method of claim 32, wherein the in vitro experiments refer to results of challenge between bacterial isolates and bacteriophages obtained from experiments in liquid or semi-solid culture media on agar media.
34. The method of claim 32, wherein further comprises a feedback stage comprising: a) visually presenting the interaction predictions between bacterial isolated and bacteriophages obtained in the prediction module; b) selecting bacteriophages of interest; c) directing in vitro experiments using the selected bacteriophages to evaluate the individual bacteriophage-bacterial isolate lytic susceptibility; and, d) receiving the feedback information to readjust parameters.
Citation Information
Patent Citations
Machine learning system for interpreting host phage response
US20230046598A1
Cited By
Method and system for predicting range of hosts targeted by phages, and corresponding computer device or medium
WO2026156913A1
