Methods and systems for probabilistic typing of microbial strains

Abstract

The present invention relates to a microbiological typing method comprising providing: a database of genetic profiles of microbial strains; and a phylogenetic tree, assigning the genetic spectrum to a location in the phylogenetic tree, and assigning a frequency of variation of the spectrum in the tree. The microbial typing comprises the following steps: measuring genetic spectrums of the microorganisms, determining the variation probability of each genetic spectrum in a database relative to the measured genetic spectrums, and calculating the variation probability according to the variation frequency; and determining that the microorganism belongs to a classification unit of the tree if at least the probability of the classification unit is above a threshold. According to the invention, the phylogenetic tree is constructed from the core genome of the microbial species and a predefined set of genetic markers is selected from the auxiliary genome of the microbial species.

Description

[0001] Invention Field This invention relates to the field of microbial genome typing, specifically the characterization of bacterial strains at a taxonomic level below the species level based on their genome. This invention is particularly applicable to the epidemiological surveillance of bacterial strains, the monitoring of bacterial contamination, and the exploration of the root causes of bacterial contamination and / or infection, whether from industrial, environmental, veterinary, clinical, or other sources.

[0002] Existing technology Genome typing of microorganisms (such as bacteria) combines all the techniques used to compare the genomes of bacterial strains to characterize them at a level below the species level. Typing makes it possible to determine the similarity between two bacterial strains or to classify strains into subdivisions at the species level, such as clonal groups or groups defined by phenotypic characteristics of bacteria, such as their serotype or their susceptibility to antibiotics.

[0003] Typing techniques can be divided into two categories. The first is based on the analysis of very small targeted portions of the bacterial genome, such as “16S rRNA” MLST (Multi-Locus Sequence Typing). The second is based on comparisons of genomes obtained through whole-genome sequencing (WGS), such as cg-MLST (Core Genome MLST) typing. While the latter offers the highest level of accuracy, whole-genome sequencing has several drawbacks, including time, cost, and the need for complex bioinformatics processing. Conversely, when the first techniques use a very limited number of markers, they allow for detection using polymerase chain reaction (PCR) amplification platforms, which are much faster and cheaper.

[0004] Regardless of the technology used, existing genomic typing techniques are based on the retrieval and use of conserved portions of the genome to characterize microorganisms. For example, MLST typing is based on housekeeping genes that are highly stable over time; similarly, cg-MLST typing directly targets the entire core genome (or basic genome: these terms are used interchangeably). Using this type of approach, achieving high accuracy requires a very large number of markers. For instance, to detect the species *Pseudomonas aeruginosa* with over 90% accuracy... P. aeruginosa For clones of ), cg-MLST typing requires thousands of markers, making it irrelevant or even impossible to use conventional PCR platforms.

[0005] In summary, users who wish to perform molecular typing, for example for the purpose of monitoring microbial contamination, must choose between accuracy and speed.

[0006] Description of the present invention The purpose of this invention is to propose a molecular typing method using a limited number of high-precision genetic markers.

[0007] Therefore, the present invention relates to an in vitro method for identifying microorganisms contained in a sample, comprising: a. Provide a computer storage device, comprising: - A database of genetic profiles of microbial strains belonging to the microbial species described above, the profiles consisting of the presence or absence of predefined sets of genetic markers; - The phylogenetic tree of the microbial species, the allocation of the genetic spectrum of the database to the position in the phylogenetic tree, and the variation frequency of the genetic spectrum of the microbial strains in the database at a predefined level of the phylogenetic tree; b. Measure the genetic spectrum of the microorganism; c. For at least one level of the set of predefined levels of the phylogenetic tree, computer-mediated: - Determine the probability of variation for each genetic spectrum in the database belonging to the said hierarchy relative to the measured genetic spectrum, and calculate the probability of variation based on the variation frequency of the said hierarchy; - If a subclass of the hierarchy meets a first identification performance criterion, then the membership of the microorganism in the subclass is determined; if at least one mutation probability of the subclass is greater than a first predefined threshold, then the first criterion is met. In the method: - Construct the phylogenetic tree from the core genome of the said microbial species; and - The predefined set of genetic markers is selected from the auxiliary genome of the microbial species.

[0008] This invention is based on the principle that when the variation frequencies in the auxiliary genome are structured using a phylogenetic tree constructed on the core genome, these frequencies effectively explain the diversity of microbial species. Applying this principle, this invention aims to construct a precise and flexible genomic "sensor" for this information, making it simple, rapid, and reliable for user use. This sensor measures the spectrum of a strain under test and estimates the probability of its belonging to a taxonomic unit based on the variation frequencies of the spectrum of taxonomic units in the tree.

[0009] Therefore, this invention allows for classification accuracy across hundreds of subclasses of a species. For example, for Listeria monocytogenes with 23 and 13 markers respectively (… Listeria monocytogenes (and Salmonella, yielding allocations of greater than 80% in an 18-level tree with a 95% confidence interval.) This very small number of genetic targets allows for the use of amplification platforms, such as PCR platforms, like the GENE-UP sold by the applicant. ®And FilmArray platform, or isothermal amplification platform, or DNA chip.

[0010] According to one implementation scheme, if the first performance criterion for identification at that level is not met, then step c) is executed for the next level in the system generation tree.

[0011] According to one implementation, the number of markers in the genetic profile is less than 50, and preferably between 16 and 30.

[0012] According to an implementation plan: - The phylogenetic tree is constructed by applying genetic distance-based clustering; - The genetic distance decreases from the root to the leaf of the tree to define the hierarchy in the tree containing differences between 150 and 350 in the core genome.

[0013] According to one implementation, a predefined set of markers is selected to minimize the entropy or impurity criterion of the group at the hierarchical level of the phylogenetic tree. Specifically, genetic profile markers are selected to minimize the entropy or impurity criterion at the hierarchical level containing differences between 150 and 350 in the core genome.

[0014] According to one implementation, the first identification performance criterion also includes the difference between the mutation probability of the subclass and the mutation probabilities of other subclasses in the hierarchy; if the difference is greater than a second predefined threshold, the first identification performance criterion is met. Specifically, the second identification performance criterion also includes the difference between the mutation probability of the group and the mutation probabilities of other groups; if the difference is greater than a fourth predefined threshold, the second identification performance criterion is met.

[0015] According to one embodiment, the database contains pairs of first and second thresholds, and for each threshold, includes at least one identification performance metric, which comprises at least one indicator related to the specificity and / or sensitivity of the identification method. Specifically, the identification performance metric includes the degree of accuracy of assignment and / or raw accuracy and / or balanced accuracy. Advantageously, the performance metrics of the first and second threshold pairs are displayed on a screen, wherein a user selects a displayed identification performance metric value, and wherein the method appends the first and second thresholds to their values ​​corresponding to the selected performance metric values.

[0016] According to one embodiment, for at least some of the genetic profiles, the database contains memberships of the genetic profiles in different microbiomes that are not present in the phylogenetic tree, and the method includes computer-mediated: - Determine the probability of variation of the genetic spectrum belonging to the group relative to the measured genetic spectrum, and calculate the probability of variation based on the variation frequency at a predefined level in the phylogenetic tree; - If, for the subclass of the level, one of the groups of microorganisms meets the second identification performance criterion, then the membership of the group is determined; if at least one of the group's mutation probabilities is greater than a third predefined threshold, then the second criterion is met.

[0017] Specifically, the second identification performance criterion also includes the difference between the mutation probability of the stated group and the mutation probability of other groups; if the difference is greater than a fourth predefined threshold, the second identification performance criterion is met. It is worth noting that the microbiome is a serotype or a standard sequence.

[0018] According to one implementation, the method also includes exploring the root cause when food contamination occurs.

[0019] According to one implementation scheme, the genetic profile of microorganisms is measured by amplifying the target genetic sequence without performing complete sequencing, particularly by polymerase chain reaction.

[0020] The subject of this invention is also a system for in vitro typing of microorganisms contained in a sample, comprising: a. A computer storage device, comprising: - A database of genetic profiles of microbial strains belonging to the microbial species described above, the profiles consisting of the presence or absence of a predefined set of genetic markers; - The phylogenetic tree of the microbial species, the allocation of the genetic spectrum of the database to the position in the phylogenetic tree, and the variation frequency of the genetic spectrum of the microbial strains in the database at a predefined level of the phylogenetic tree; b. A device for collecting and storing measurement results of the genetic spectrum of the microorganism; c. A computer-mediated apparatus for targeting at least one of the predefined levels of the phylogenetic tree: - Determine the probability of variation for each genetic spectrum in the database belonging to the said hierarchy relative to the measured genetic spectrum, and calculate the probability of variation based on the variation frequency of the said hierarchy; - If the subclass meets the first identification performance criterion, the membership of the microorganism in the subclass at the level is determined, and if at least one mutation probability of the subclass is greater than a first predefined threshold, the first criterion is met.

[0021] Such a system is configured to execute methods of the type described above.

[0022] The subject of this invention is also a computer program product recorded on a computer-readable support, which includes instructions for performing step c of a method of the type described above.

[0023] The subject of this invention is also a computer program product recorded on a computer-readable support, which includes instructions for constructing a phylogenetic tree based on the variation frequencies of a set of markers of the core genome and auxiliary genome at the levels of the phylogenetic tree.

[0024] The subject of this invention is also a composition for in vitro typing of microorganisms contained in a sample by amplifying target nucleotide sequences, particularly by polymerase chain reaction amplification, said composition comprising a group of oligonucleotides for amplifying at least one group of genetic markers obtained according to the above method.

[0025] The subject of this invention is also a method for preparing biological samples, which contain or may contain bacterial strains for typing the strains by polymerase chain reaction, the method comprising contacting the sample with the aforementioned composition. In particular, the method involves lysing the sample obtained from the contact to release genetic material from the bacterial strain.

[0026] Brief description of the attached figures The invention will be better understood by reading the following description, which is given by way of example only and in conjunction with the accompanying drawings, wherein: - Figure 1 This is an overview of the method according to the present invention; - Figure 2 This is a flowchart detailing the learning phase of system fractalization according to the present invention; - Figure 3 This is a diagram illustrating the gradually decreasing genomic distances used to construct the phylogenetic tree; - Figure 4 This is a schematic diagram of a binary tree system; - Figure 5 This describes a scheme for calculating the variation frequency of the genetic spectrum for taxonomic units in a phylogenetic tree according to the present invention; - Figure 6 This is a flowchart illustrating the production of the PCR kit for typing according to the present invention; - Figure 7 This is a flowchart detailing the system occurrence type prediction stage according to the present invention; - Figure 8 This is a flowchart illustrating the calculation of the thresholds used during the prediction phase; - Figure 9 This is a flowchart illustrating the learning phase of serotyping according to the present invention; - Figure 10This is a flowchart illustrating the serotype prediction stage according to the present invention; - Figure 11 This is a flowchart illustrating the calculation of the thresholds used during the serotype prediction phase; - Figure 12 This is a graph showing the decrease in Shannon entropy during the selection of genetic markers that constitute the genetic spectrum of Listeria typing; - Figure 13 This is a set of diagrams showing the variation frequencies of the genetic spectrum in the phylogenetic tree of Listeria. - Figure 14 This is a graph showing the change in the correct serotype allocation of Listeria monocytogenes as a function of balance accuracy. - Figure 15 These are a set of graphs showing how the correct phylogenetic allocation of Listeria changes with average accuracy. - Figure 16 This is a graph showing the decrease in Shannon entropy during the selection of genetic markers that constitute the genetic spectrum used for Salmonella typing; - Figure 17 This is a set of diagrams showing the variation frequencies of the genetic spectrum in the phylogenetic tree of Salmonella; - Figure 18 This is a graph showing how the correct serotype allocation of Salmonella changes with the balance accuracy. - Figure 19 These are a set of graphs showing the degree of correct phylogenetic partitioning of Salmonella as a function of average accuracy; and - Figures 20 to 22 A computer architecture for performing classification according to the present invention is shown. Invention Details A. Definition The following definitions apply in this document: - Biological sample: refers to any sample that contains or may contain one or more microorganisms; - The core genome (or "basic genome", denoted as "Gb") of a microbial species refers to the portion of the genome (genes or genome) common to all microbial strains belonging to one or more microbial species under consideration. For the purposes of this invention, the core genome can evolve if a group of strains producing the core genome evolves. Preferably, the core genome is constructed from at least 100 different microbial strains, more preferably from at least 1000 strains. It should be noted that the genomic portions of the microbial strains belonging to the core genome can differ from each other. For example, a particular gene can belong to the core genome and have several different alleles. Hereinafter, when applied to microbial strains, the term "core genome" refers to the portion of the genome of the strain that belongs to the core genome of the species.

[0028] - Auxiliary genome of a microbial species (denoted as "Ga"): represents a portion of the genome (genes or genome) of a group of bacterial strains that is not part of the core genome of said group. Therefore, these portions of the genome are not present in the genome of at least one bacterial strain within the group of bacterial strains under consideration. For the purposes of this invention, the auxiliary genome can evolve if the group of strains that produced it evolves. Preferably, the auxiliary genome is constructed from at least 100 different microbial strains, more preferably from at least 1000 strains.

[0029] - Genetic markers: Genes or nucleotide sequences that represent a specific location (also called a "locus") in the genome. They can be equivalent to nucleotide sequences of a single nucleotide, such as two alleles that differ by a single nucleotide, or single nucleotide polymorphisms (or SNPs for "single nucleotide polymorphisms") in a mutational context. - Microorganisms: refers to all microscopic organisms, especially including bacteria, yeast and fungi.

[0030] - Phylogenetic tree of microbial species: A tree representing the evolutionary relationships among a group of microbial strains, determined based on the similarity and differences in their core genomes. For the purposes of this invention, the phylogenetic tree can be enriched if the group of strains that produced it evolves. Preferably, the phylogenetic tree is constructed from at least 100 different microbial strains, more preferably from at least 1000 strains. Typically, the root of the phylogenetic tree is represented as the first level of the tree, and the leaves of the tree are represented as the last level. Thus, levels "above" another level correspond to levels closer to the root.

[0031] B. Embodiments of the present invention The following discloses a probabilistic typing method according to the invention, which is applied to monitor contamination of a bacterial strain of a species within an organization, particularly in a place used for the production of food for human consumption, and is also applied to the detection of its serotype.

[0032] refer to Figure 1 The classification according to the present invention includes the following main stages: - Phase 10: Establishing a database consisting of complete genomes of strains belonging to bacterial species; - Phase 20, referred to as “learning” or “training”, aims to select appropriate genetic profiles and the mutation frequencies of said genetic profiles along the tree based on the phylogenetic tree generated in the database; - “Prediction” phase 30 performs typing of unknown bacterial strains based on the tree and the frequencies generated in phase 20; and - Optionally, a "epidemiological" phase 40 is added, during which corrective and / or preventive measures are taken within the tissue based on the typing results of the unknown strain.

[0033] Preferably, the database of complete genomes represents the genetic diversity of species as comprehensively as possible. Advantageously, if the initial version is incomplete, it is continuously supplemented by adding complete genomes, with this diversity as the target. In this case, typing according to the invention is performed when the database contains at least 100 species genomes, preferably at least 1000 genomes, or even more preferably several thousand genomes.

[0034] refer to Figure 2 , 3 In step 4, training phase 20 begins by generating the core genome in step 202 and the auxiliary genomes in step 204 from genomes in database 200. In the first variant, the genomes include annotations classifying genes as belonging to one or more of these genomes. In the second variant, these are generated provisionally from the database, where genes with allele frequencies above a threshold (e.g., 95%) belong to the core genome, otherwise to the auxiliary genome. In the third variant, the core and auxiliary genomes are generated from the database by comparison with a reference genome of the species stored in the database, where genes with allele frequencies relative to the reference genome above a threshold, e.g., 95%, belong to the core genome, otherwise to the auxiliary genome. In the fourth variant, a database has already been established for the species, as have versions of its core and auxiliary genomes.

[0035] Continuing with the typing according to the invention in 206, a phylogenetic tree is generated based on the core genome. Several techniques for constructing such a tree are possible, for example, as described in the document " Phylogenetics Algorithms and Application (Hu et al., Ambient Communications and Computer Systems, 2019) A Review: Phylogeny Construction Methods (Shaktawat et al., International Journal of Emerging Science and Engineering, 2019) or " Essential Bioinformatics As described in "(JinXiong, Cambridge University, 2011)".

[0036] In a variant of this invention, the tree is constructed using cg-MLST (core genome multi-locus sequencing genotyping) encoding of the core genome. Software such as BioNumerics (bioMérieux, Marcy l'Etoile, France), SeqSphere+ (Ridom, Münster, Germany), and BIGS db (accessible at https: / / github.com / kjolley / BIGSdb and in the documentation of Schürch, A. et al.) are used. Whole genome sequencing options for bacterial strain typing and epidemiologic analysis based on single nucleotide polymorphism versus gene-by-gene–based approaches (Described in Clin. Microbiol. Infect., 2018) or chewBBACA (accessible at https: / / github.com / B-UMMI / chewBBACA and in the article by SilM et al., " chewBBACA: a complete suite for gene-by-gene scheme creation and strain identification. (As described in Microb.Genom., 2018) These are translated into cg-MLST spectra. The advantage of cg-MLST spectra is that they are fixed-length sequences consisting of hundreds of components that match each other, allowing for direct use of tree-building methods.

[0037] This tree is advantageously generated by applying a series of groupings using a single-linkage clustering method, and is therefore based on, for example, the genetic distance between core genomes calculated using normalized Hamming distance. Subclasses of taxonomic units in the tree then group core genomes within the same subgroup that differ from each other by a predefined threshold. This can be achieved, for example, using software such as HierCC (accessible at https: / / github.com / zheminzhou / pHierCC and in the paper by Zhou et al.), HierCC: a multi-level clustering scheme for population assignments based on core genome MLST The tree was constructed using data from Pathogenwatch (accessible at https: / / pathogen.watch / ) and Bioinformatics, 2021, and stored in database 208.

[0038] In a particularly preferred variant of the invention, the number of levels and clustering distances in the tree are selected to produce at least one level in the tree such that the diversity of the core genome at that level corresponds to or approximates the clonal diversity observed in the species, and such that the taxonomic units at that level allow for the differentiation of distinct clonal complexes among bacterial strains corresponding to the complete genomes in the database. For example, for Salmonella and Listeria, this level has a difference between 150 and 350 in the core genome of the strains, preferably between 200 and 300. In particular, the clustering distance decreases from root to leaf, for example, linearly.

[0039] Learning phase 20 continues with the selection of a genetic spectrum (i.e., a set of genetic markers from the auxiliary genome of a species) in phase 210, during which the absence and presence of said genetic markers are measured in the strains to be tested during prediction phase 30. The selection of the spectrum involves choosing a level in the selection tree, and preferably one of the levels representing clonal diversity, and then selecting a set of markers with the lowest entropy or impurity level at that level. The total number N of markers selected for constructing the genetic spectrum... M Advantageously, the selection should be based on the specific PCR platform used to perform the typing. For example, for the applicant's GENE-UP platform, the number N... M It is equivalent to 23 markers for Salmonella and 16 markers for Listeria monocytogenes.

[0040] In practice, a greedy algorithm is used to select the set of markers to minimize the entropy criterion (e.g., Rényi entropy, Hartley entropy, Shannon entropy, conflict entropy, or minimum entropy) or the Gini impurity criterion.

[0041] For illustrative purposes, and for reference Figure 4 In the phylogenetic tree, a set of labels is selected to minimize the result of Q. N =20taxaT1、T2,…,T 20 The Shannon entropy criterion for the hierarchy N consists of a first marker M1 selected from the auxiliary genome that minimizes the entropy criterion: in Where NG is the total number of core genomes used to construct the phylogenetic tree, and The label exists in the i-th classification unit T i The number of times in the auxiliary genome. After selecting marker M1, the procedure is repeated to select the second marker M2 in the auxiliary genome removed from marker M1. Therefore, the j-th marker M... j It is the mark that makes NG is the total number of core genomes used to construct the phylogenetic tree, and It is a marker Existing in the i-th classification unit T i The number of times in the auxiliary genome.

[0042] As a variant, stepwise label selection is performed by minimizing the conditional entropy of level N according to the following formula: in: - in Where NG is the total number of core genomes used to construct the phylogenetic tree, and The label exists in the i-th classification unit T i The number of times in the auxiliary genome.

[0043] - and in It is the group Among the possible values, the marker group The l-th value exists in the i-th classification unit T i The number of times in, and It is included in the classification unit T i The number of core genomes in the genome.

[0044] The inventors observed that the typing according to the invention requires no more than 50 markers, preferably less than 30 markers, and more specifically, a number between 16 and 30 markers, to obtain high typing performance.

[0045] Learning phase 20 continues in 212, extracting auxiliary genomes from the genetic spectrum formed by selected markers, which are stored in database 214. Therefore, the genetic spectrum of the auxiliary genomes is a vector P. The j-th component P(j) of vector P corresponds to the label. The value when the tag When it does not exist, the value is equal to 0; when the flag is not present, the value is equal to 0. When it exists, the value is equal to 1.

[0046] After forming the spectral database 214, the variation frequency of the spectrum along the tree is calculated in 216. (Reference) Figure 5 For those classified into subclasses T1, T2, ..., T Q Each taxonomic unit T in the tree has a set of genetic profiles associated with taxonomic unit T. It is divided into subclasses T1, T2, ..., T Q A subset of the associated spectrum , … For each marker M in the genetic spectrum j The mutation frequency of this marker in taxonomic unit T The calculation is equal to marker M. j Classification units T1, T2, ..., T that take values ​​of 0 and 1 Q The number divided by Q is the frequency according to the following relationship: in Let represent the cardinality function of set E. Therefore, for each subdivided class unit T, the frequency calculation produces... frequency vector , making After calculation, the frequency vector is stored in database 218.

[0047] Learning phase 20 ends in 220 with the calculation of the hyperparameter mesh used during prediction phase 30, which will be described in detail below. After prediction 30, the parameters are stored in database 222.

[0048] In summary, learning phase 20 provides a phylogenetic tree in which each complete genome extracted from its auxiliary genome and its genetic spectrum are located in the tree, and each taxonomic unit (excluding leaves) in the tree is associated with the variation frequency of each genetic marker in the spectrum.

[0049] refer to Figure 6 The present invention includes the design and manufacture 50 of a composition or "kit" for performing amplification of a selected marker, comprising: - In 500, select the amplification primers used to amplify the selected genetic marker: - The primers are generated in 502, and if necessary, a detection probe is generated to detect the presence of at least one of the markers in its variants, and preferably the maximum number present in its variants; - In 504, produce a buffer solution containing selected primers, detection probes (if needed), dNTPs (deoxyribonucleic acid triphosphates) to provide the energy and nucleotide bases required for amplicon generation, amplase, and salts (e.g., Mg). 2+ (Ions or NaCl) are added to the solution to ensure the enzyme functions properly at the appropriate concentration. A dye can be added to the solution.

[0050] - Produce a lysis buffer containing beads (e.g., magnetic or ceramic beads) and optionally present dyes.

[0051] The generated kit was then used during prediction phase 30. (Reference) Figure 7This stage begins with the collection of biological samples 300 from an tissue (e.g., a food production facility). During the subsequent step 302, methods known in the art, such as culturing on plates, are used. ® Type immunological or molecular biological assays, such as those using GENE-UP ® The platform allows for sample preparation to detect a given pathogen. After the pathogen is detected, a confirmation step can be performed.

[0052] Simultaneously, if the colony is identified as belonging to the stated species, prediction phase 30 continues in 302, using isolated colonies or enriched liquid medium for characterization by PCR amplification. Specifically, the sample undergoes a series of preparatory steps, such as lysis to release DNA from the bacteria constituting it, and optionally, extraction of DNA from the lysate using beads (e.g., magnetic or ceramic beads). After releasing the bacterial strain's DNA, in 304, it is added to a PCR kit designed to at least target the genetic markers selected during learning step 20. Finally, PCR is performed in 306 to measure the strain's genetic profile, for example using the applicant's GENE-UP. ® platform.

[0053] Then, the measured spectrum is analyzed in 308, denoted as P. mes This is done to locate bacterial strains within the phylogenetic tree. Starting from level N of the tree, for example, the level directly above the leaf level, analysis 308 includes the following steps: a. For each classification unit of level N, it is represented as: (The i-th classification unit of the Nth level): a1. For those assigned to classification units In database 214, it is represented as The measured spectrum is calculated from the stored spectra according to the following relationship. It is the stored spectrum The probability of mutation : Therefore, it will be noted that if the measured spectrum With spectrum If they are the same, then the probability is equal to 1 (no variation), and if the spectrum... Deviation spectrum Then the probability decreases rapidly when the frequency... When the temperature is low, all situations are even worse.

[0054] a2. Select the classification unit Calculated probability maximum value .

[0055] b. In 310, the measured spectrum Classification units assigned to level N One of them makes: in and It is a predetermined non-zero positive threshold; c. If no classification unit at level N meets the criteria in step b, move to a higher level N-1 in the tree and perform steps a and b on that new level.

[0056] Spectrum After being assigned a level in the system tree and a classification unit for that level, prediction phase 30 delivers a report to the user, for example, as a display on the screen and / or a file stored on the computer and / or a message to the user (e.g., an email). If two classification units have the same result at the end of step b, both classification units are included in the report, or an indeterminate classification is recorded.

[0057] The position of a new bacterial strain tested in the phylogenetic tree allows it to be typed relative to bacterial strains with complete genomes corresponding to those in database 200. In addition to this typed classification with reference to previously observed strains and those included in various databases, the typed classification according to the invention allows for relative typed classification of two bacterial strains if they are assigned to the same taxonomic unit. Furthermore, the invention makes it possible to define the lineage of a strain relative to strains in a database or relative to another tested strain. Specifically, the relative positions of two strains in the tree allow for characterizing their relative evolution.

[0058] The path from leaf to root through the phylogenetic tree has been described: tracing upwards through the tree as long as the assignment relation has not been verified. As a variation, tracing the phylogenetic tree from root to leaf: tracing downwards through the tree as long as the assignment relation has been verified.

[0059] threshold and The value of is a hyperparameter preferably determined during step 220 of the learning phase 30 using cross-validation. (See reference) Figure 8 First, determine the threshold. and The values ​​are gridded, and for each pair of thresholds in the grid, the correct allocation degree and average accuracy are determined by: - Based on the cross-validation strategy 804, such as the strategy called "ten-fold cross-validation", the complete genome database 200 is divided into a training database 800 and a test database 802. - Perform learning phase 20 on training database 802; - Perform prediction phase 30 for each genetic spectrum from test library 802, and calculate the average accuracy ACC and specified correctness TA of the genetic spectrum from database 802 in phylogenetic tree in 806; In 808, the maximum value of average accuracy and assignment correctness is selected in the folding of the cross-validation strategy. Therefore, these two values ​​correspond to the thresholds respectively. and The average accuracy and degree of assignment correctness of the system occurrence prediction phase 30 are determined. A threshold pair with optimal accuracy and degree is selected for system occurrence typing according to the invention. As a variation, the user can select their desired accuracy and / or degree of assignment, and the system occurrence typing selects the threshold pair whose accuracy and degree are closest to the user's selected threshold pair.

[0060] In a preferred approach, the hierarchy in the phylogenetic tree and the distance decay law (e.g., affine) for clustering are also hyperparameters, therefore, Figure 8 The described method is supplemented by a grid for these two parameters, and a path-combined threshold for the grid. and The selection of the grid path.

[0061] Preferred variations of the invention are now described, including predicting the membership of bacterial strains within a specific group of bacterial species based on their measured genetic profiles. In the example shown, these groups are serotypes, but this variation applies to any clearly defined subclass at the bacterial species level. It should be noted that such a subclass of a species does not correspond to the first level of the phylogenetic tree and is defined independently of it.

[0062] refer to Figure 9 The above about Figure 2 The described learning phase 20 supplements serotype learning 20A, which includes: a. Collect serotypes from all or some bacterial strains corresponding to the complete genome, and store the serotypes in database 224; b. In 226, ST1, ST2, ... ST were distributed across different serotypes in the NS region. NS The genetic profiles observed in a database of 214 serotyped strains. The genetic profiles associated with serotypes are referred to below as "serotyped." c. In step 228, the variation frequencies of the serotype genetic profile are calculated. It is worth noting that for each marker M... j Its variation frequency in serum type STi It equals the number of times the marker takes the value 1 out of the total number of genetic profiles assigned to serotype STi; d. Store the variation frequencies of serotyped genetic profiles for different serotypes in database 230. .

[0063] if Figure 2 If the elements of the described learning phase 20 remain unchanged, specifically meaning that the type of genetic spectrum used is determined based on a phylogenetic tree rather than serotype subclasses, then the prediction phase 30 can take two forms. In the first form, with... Figure 2 In the second form, steps 308-312 are performed in parallel to supplement the already described stage 30 with a prediction of the serotype of the new strain to be tested. In the third form, only the prediction is performed, replacing steps 308-312.

[0064] refer to Figure 10 The prediction of serotypes of new bacterial strains 30A follows steps 300-306 of measuring the genetic profile of bacterial strains of a species contained in a biological sample, and includes the following steps: a. For each serotype, denoted as ST i : a1. For group ST in database 228 i The genetic profiles of each serotype (represented as) The measured spectrum is calculated based on the following relationship. It is a stored spectrum The probability of mutation : a2. Selecting serum-type ST i Calculated probability maximum value ; b. The measured spectrum Assigned to serum type ST j One of them makes: in and It is a predetermined non-zero positive threshold; c. In 318, generate and deliver reports on the serotypes detected for bacterial strains that have undergone serotyping.

[0065] In variant 30A of serological typing, if the second condition is not met... Then the serotype typing returns the highest probability. And the first two or three serotypes associated with the probability.

[0066] threshold and The value of is a hyperparameter preferably determined during step 220 of the learning phase 30 using cross-validation. (See reference) Figure 11First, determine the threshold. and The values ​​are arranged in a grid, and for each pair of thresholds in the grid, the accuracy of serotype assignment and average precision are determined as follows: - - Based on cross-validation strategy 1104, such as the strategy called "ten-fold cross-validation", the whole genome database 200 and serotype 224 are divided into training database 1100 and test database 1102. - Perform learning phases 20-201 on training database 1102; - Perform prediction phase 30A for each genetic spectrum from test database 1102, and calculate the mean accuracy ACC and assignment correctness TA of the genetic spectrum from database 802 in serotypes in 1106; In step 1108, the maximum value of the average accuracy and the degree of assignment correctness is selected in the folding of the cross-validation strategy. Therefore, these two values ​​correspond to the thresholds respectively. and The average accuracy and assignment correctness of the serotype prediction phase 30A are determined. According to the invention, a threshold pair with optimal accuracy and accuracy is selected for serotype typing. As a variation, the user can select their desired accuracy and / or accuracy of the assignment, and the serotype typing according to the invention selects the threshold pair whose accuracy and accuracy are closest to the user's selected threshold pair.

[0067] In the variant just described, determining the serotype of the strain to be tested involves, as per... Figure 9 and Figure 10 The calculations described. In another variation, for each taxonomic unit in the phylogenetic tree, the probability of each serotype in that taxonomic unit is calculated based on the serotype database 224. For example, taxonomic unit T i Given serotype The probability of a serotype is equal to the number of genomes with that serotype grouped within that taxonomic unit divided by the total number of genomes in that taxonomic unit. These probabilities are stored in a reference table or dictionary, and predicting the serotype of a strain to be tested involves first determining the taxonomic unit to which it belongs in the phylogenetic tree, as described in steps 308-312. After identifying the taxonomic unit, serotyping of the strain involves querying the probabilities in the reference table associated with that taxonomic unit and providing the user with at least the serotype with the highest probability.

[0068] In the typing and serotyping variants just described, after measuring the spectrum, the values ​​are executed. , The calculation and the measured spectrum The allocation. As a variant, the known spectrum is directly associated with one or more taxa in the binary phylogenetic tree, for example, using a reference table. Specifically, the taxa are known, and the allocation calculation has been pre-compiled. It is worth noting that the spectrum... The strain is assigned to its associated taxonomic unit through the calculations described previously. Therefore, if the spectrum of the new strain matches the spectrum... If they are the same, the reference table is called for that spectrum, and the associated classification unit is directly output. Similarly, when a new spectrum... When a measured spectrum differs from all spectra in the reference table, its assignment is calculated as previously described, and then the new spectrum is stored along with one or more taxonomic units identified in the reference table. In this way, the assignment calculations already performed are avoided by simply calling the reference table. As a variation, instead of a reference table, each known spectrum is stored in memory space dedicated to one or more taxonomic units in the phylogenetic tree. In other words, taxonomic units in the tree are assigned "addresses" equal to the known and assigned spectra.

[0069] D. Example D.1. Listeria monocytogenes 37,000 Listeria strains were collected, fully sequenced, and serotyped. The number of levels in the tree is equal to 23, which is the applicant's GENE-UP. ® The total number of targets on the PCR platform was determined, and the distance decay law was selected for clustering to obtain diversity similar to clonal diversity at levels between 10 and 15 in a phylogenetic tree comprising 24 levels. Figure 12 The decrease in Shannon entropy of the genetic spectrum is shown because the markers are selected from the auxiliary genome at level 13 of the phylogenetic tree, and the entropy of the final spectrum differs from the free entropy of that level by only a few percent. Figure 13 The graph shows the frequency of variation in the spectrum at different levels of the tree. Curves in the graph correspond to taxa at that level, and bold curves correspond to the frequencies of tagged labels. From Figure 14 As can be seen, it illustrates serotyping of 12 known Listeria serotypes. For a 100% accuracy rate in serotyping, the average equilibrium accuracy is 90%, demonstrating the effectiveness of this serotyping method according to the invention. It should be noted that a higher accuracy rate can be selected for a satisfactory lower level of serotyping. Notably, for an 80% accuracy rate, the accuracy is close to 95%, with the 95% confidence interval completely contained within the last decimal place. Figure 15 The average accuracy of the levels is shown as the degree of allocation varies with the correct system. Again, note the excellent performance of the invention, where both values ​​are greater than 95% and 70% for all levels above level 15, and close to 100% for levels above level 9.

[0070] D2. salmonella 39,000 Salmonella strains were collected, fully sequenced, and serotyped. The tree has 23 levels, which is the applicant's GENE-UP. ® The total number of targets on the PCR platform was determined, and the distance decay law was selected for clustering to achieve a diversity similar to clonal diversity at levels 10 and 15 in a phylogenetic tree comprising 24 levels. Figure 16 The results show that when markers are selected from the auxiliary genome at level 13, the Shannon entropy of the genetic spectrum decreases, and the entropy of the final spectrum is essentially zero. Figure 17 The graph shows the frequency of variation in the spectrum at different levels of the tree. Curves in the graph correspond to taxa at that level, and bold curves correspond to the average frequency at that level. Figure 18 It can be seen from this that Figure 18 Serotyping of over 2600 known serotypes of Salmonella is shown, with an average equilibrium accuracy greater than 90% for 100% correct serotype allocation, demonstrating the efficacy of this serotyping method according to the invention. It should be noted that a higher accuracy can be selected for a satisfactory lower allocation. Notably, for 80% accuracy, the accuracy approaches 100%, with the 95% confidence interval completely contained within the last decimal place. Figure 19 The average accuracy of the hierarchy as a function of the degree to which the correct system assignment occurs is shown. Again, note the excellent performance of the invention; for all hierarchies above level 13, these two values ​​are greater than 95% and 80%, respectively.

[0071] E. Prevention and Corrective Measures When contamination of finished food products is identified during routine inspections, the solution's ability to easily and rapidly identify strains makes its use for investigating root causes conceivable. This includes, in particular, searching for strains in the relevant factory environment through surface samples collected, for example, with swabs, sponges, or cloths. Strains can also be searched in starting materials used to create the relevant finished product to link contaminants to one or more suppliers and / or transport vehicles and / or storage locations. It also makes it possible to identify laboratory strains or contamination through their DNA during the laboratory testing phase, which may interfere with release tests identifying the presence of relevant species. In such cases, it will allow for planning to retest the finished product and avoid its disposal. Finally, the solution can be used during health crises to rapidly characterize the responsible strains and link them to strains identified in food, cosmetics, veterinary, or other implicated products. Therefore, its usefulness covers internal needs within production sites, such as investigative needs during food, health crises, or on-site regulatory audits. Thus, the use of this solution simplifies and enables the implementation of appropriate preventative and / or corrective measures to reduce or even eliminate the risk of recurrence of contamination and consequently, recurrence of infection.

[0072] F. Computer hardware implementation of the present invention Learning phases 20 and 20A and prediction phases 30-30A, except for the steps of preparing samples and measuring the genetic profile of the strains to be tested, are performed by a computer, i.e., by means of hardware circuitry including computer memory (cache, RAM, ROM, etc.) and one or more microprocessors or processors, which may or may not be organized as computing nodes, and are necessary to execute computer instructions stored in memory for performing the phases. For example, as Figure 20-22 As shown, several architectures are possible. In the first architecture variant ( Figure 20 In this architecture, a first organization 2000 (e.g., the applicant) hosts or controls one or more computing servers 2002 associated with one or more databases 2002 for storing genomes, phylogenetic trees, phylogenetic frequencies, observed genetic profiles, serotypes, and serotype frequencies, and learning phases 20, 20A are performed by organization 2000 on its servers 2002. A second organization 2006, such as a microbiology laboratory, hosts a PCR platform 2008 connected to or incorporated into computer 2010 and performs a portion of prediction phase 30, i.e., all steps up to measuring the genetic profile of the bacterial strain to be tested. The measured profile is then sent to the first organization 2000 via a remote connection network to implement the remainder of prediction phases 30, 30A on server 2002. The resulting report is then sent to the second organization 2006 via network 2012, and the second organization 2006 decides whether to take epidemiological measures based on the report. Second architecture variant ( Figure 21The difference from the first architecture variant is that copies of the database 2004 and the software executing prediction phases 30 and 30A are downloaded to a second organization that uses a computer 2010 or a computing server (not shown) to execute all prediction phases 30 and 30A. In the third variant ( Figure 22 In this process, learning phases 20 and 20A, and prediction phases 30 and 30A, are all performed by a single organization in 2006.

[0073] G. Extension of the teachings of the implementation plan One embodiment of the present invention, which uses PCR to measure the genetic profile of bacterial strains, has already been described. As a variation, the profile is measured using a DNA chip that targets the marker in a manner known per se.

[0074] An implementation has been described in which a complete genome database, at least in its initial version, is independent of the bacterial ecology of the tissue performing typing for its own needs. As a variant, this database consists primarily of bacterial strains derived solely from the tissue. However, a minimal first version of the database can be provided to the tissue, and then supplemented with strains from the tissue.

[0075] Implementation schemes for bacterial species have been described. This invention is also applicable to yeasts and fungi.

[0076] A serotyping method has been described. It can characterize other subtypes of a species, such as their resistance and / or sensitivity to microbial agents or groups based on standard sequences.

[0077] The correctness of the assignment and the accuracy (balanced or unbalanced) have been described as criteria for selecting hyperparameters for prediction. Other criteria are possible, such as sensitivity and specificity or confidence intervals other than 95%.

Claims

1. An in vitro method for typing a microorganism contained in a sample, comprising: a. providing a computer storage device containing: - a database of genetic profiles of microorganism strains belonging to a microorganism species of said microorganism, said profiles consisting of the presence or absence of a predefined set of genetic markers; - a phylogenetic tree of the species of said microorganism, an assignment of the genetic profiles in said database to a position in said phylogenetic tree, a frequency of variation of the genetic profiles of microorganism strains in said database for a predefined hierarchy of said phylogenetic tree; b. measuring a genetic profile of said microorganism; c. for at least one hierarchy of the set of said predefined hierarchy of the phylogenetic tree, computer-mediated: - determining a probability of variation of each genetic profile in the database belonging to said hierarchy with respect to the measured genetic profile, said probability of variation being calculated from the frequency of variation of said hierarchy; - determining membership of the microorganism in a sub-class of said hierarchy if the first identification performance criterion is met, said first criterion being met if at least one probability of variation of said sub-class is greater than a first predefined threshold; in said method: - said phylogenetic tree is constructed from a core genome of said microorganism species; and - said predefined set of genetic markers is selected from a supplementary genome of said microorganism species.

2. The method of claim 1, wherein, If said first identification performance criterion of said hierarchy is not met, step c) is performed on a directly upper hierarchy in said phylogenetic tree.

3. The method according to any one of the preceding claims, wherein the number of markers of said genetic profile is less than 50, and preferably between 16 and 30.

4. The method according to any one of the preceding claims, wherein: - said phylogenetic tree is constructed by applying a clustering based on genetic distance; - said genetic distance decreases from the root to the leaves of said tree to define a hierarchy in said tree containing between 150 and 350 differences in said core genome.

5. The method according to any one of the preceding claims, wherein the predefined markers of said set are selected to minimize an entropy or impurity criterion of said set at a hierarchy of said phylogenetic tree.

6. The method according to claims 4 and 5, wherein the markers of said genetic profile are selected to minimize an entropy or impurity criterion at a hierarchy containing between 150 and 350 differences in the core genome.

7. The method according to any one of the preceding claims, wherein said first identification performance criterion further comprises a difference between the probability of variation of said sub-class and the probability of variation of other sub-classes of said hierarchy, said first identification performance criterion being achieved if said difference is greater than a second predefined threshold.

8. The method according to claim 7, wherein said second identification performance criterion further comprises a difference between the probability of variation of said set and the probability of variation of other sets, said second identification performance criterion being met if said difference is greater than a fourth predefined threshold.

9. The method according to any one of the preceding claims, wherein the database comprises pairs of first and second threshold values, and for each of said threshold values, at least one performance indicator of identification is comprised, said performance indicator comprising at least one indicator related to specificity and / or sensitivity of the method of identification.

10. The method according to claim 9, wherein the performance indicator of identification comprises a degree of correctness of assignment and / or a raw accuracy and / or a balanced accuracy.

11. The method according to claim 9 or 10, wherein the performance indicators of the pairs of first and second threshold values are displayed on a screen, wherein a user selects a displayed value of the performance indicator of identification, and wherein the method sets the first and second threshold values to the values corresponding to the selected value of the performance indicator.

12. The method according to any one of the preceding claims, wherein for at least some of the genetic profiles, the database comprises membership of the genetic profiles in different groups of microorganisms, said different groups being absent from the phylogenetic tree, the method comprising computer-mediated: - determining a probability of variation of the genetic profiles belonging to said groups with respect to the measured genetic profile, said probability of variation being calculated from the frequency of variation at a predefined level of the phylogenetic tree; - determining membership of said groups if one of the groups of microorganisms meets a second performance criterion of identification for a subclass of said level, said second criterion being met if at least one of the probabilities of variation of said groups is greater than a third predefined threshold value.

13. The method according to claim 12, wherein the second performance criterion of identification further comprises a difference between the probability of variation of said group and the probabilities of variation of the other groups, said second performance criterion of identification being met if said difference is greater than a fourth predefined threshold value.

14. The method according to claim 12 or 13, wherein the groups of microorganisms are serotypes.

15. The method according to any one of the preceding claims, comprising exploring the root cause when a food contamination occurs.

16. The method according to any one of the preceding claims, wherein the measurement of the genetic profile of the microorganism is performed by amplification of target genetic sequences without complete sequencing, in particular by polymerase chain reaction.

17. A system for in vitro typing of microorganisms contained in a sample, comprising: d. computer storage means comprising: - a database of genetic profiles of strains of microorganisms belonging to a species of microorganism, said profiles consisting of the presence or absence of a predefined set of genetic markers; - a phylogenetic tree of species of said microorganism, an assignment of the genetic profiles in said database to a position in said phylogenetic tree, a frequency of variation of the genetic profiles of strains of microorganisms in said database at a predefined level of said phylogenetic tree; e. means for collecting and storing measurements of genetic profiles of said microorganism; f. computer-mediated means for, for at least one level of a predefined set of levels of said phylogenetic tree: - determining a probability of variation of each genetic profile belonging to said hierarchy in the database with respect to the measured genetic profile, said probability of variation being calculated from the frequency of variation of said hierarchy; - determining the membership of a microorganism in a subcategory of said hierarchy if said subcategory meets a first identification performance criterion, said first criterion being met if at least one probability of variation of said subcategory is greater than a first predefined threshold.

18. The system of claim 16, configured to perform the method of any one of claims 2 to 16.

19. A computer program product recorded on a computer readable support, comprising instructions for performing step c of the method of any one of claims 1 to 16.

20. A computer program product stored on a computer readable support, comprising instructions for constructing a phylogenetic tree from the frequency of variation of a hierarchy of phylogenetic trees from a set of markers of a core genome and of a complementary genome.

21. A composition for in vitro typing of a microorganism contained in a sample by amplification of a target nucleotide sequence, in particular by polymerase chain reaction, said composition comprising a set of oligonucleotides for amplifying at least a set of genetic markers obtained as described in claim 5 or 6.

22. A method of preparing a biological sample comprising or possibly comprising a bacterial strain for typing said strain by polymerase chain reaction, said method comprising contacting said sample with the composition of claim 21.

23. The method of claim 22, comprising lysing the sample obtained from said contacting to release genetic material from the bacteria of said bacterial strain.

Images