Methods for epidemiological identification and surveillance of bacterial outbreaks

Abstract

The present invention relates to a method for detecting and monitoring bacterial outbreaks, which includes predicting that strains from a collection and a database belong to a bacterial outbreak if their genomic distance is below a first predetermined threshold, do not belong to a bacterial outbreak if their genomic distance is above a second predetermined threshold that is strictly above the first threshold, or may belong to a bacterial outbreak if their genetic distance is intermediate. The first threshold is equal to or greater than a third threshold, so that the prediction that two strains with a genomic distance below the third threshold belong to a bacterial outbreak has the greatest specificity. The second threshold is equal to or less than a fourth threshold, so that the prediction that two strains with a genomic distance above the fourth threshold do not belong to a bacterial outbreak has the greatest sensitivity.

Description

[Technical field]

[0001] The present invention relates to the field of bacterial epidemiology, in particular the detection and monitoring of bacterial outbreaks depending on partial or complete sequencing of the genome of bacterial strains, in particular their DNA and / or RNA. [Background technology]

[0002] The detection of infectious bacterial outbreaks traditionally consists of determining whether several strains collected from subjects (e.g., from patients and by extension animals) originate from a recent transmission of the same strain between subjects, e.g., transmission of a strain from a "source" subject to several subjects, or transmission of a strain from subject to subject. Based on classical microbiological instrumentation, detection is usually performed in two steps. First, suspect a bacterial outbreak. This suspicion arises when the sampled strains belong to the same bacterial species and share common phenotypic characteristics, e.g., the same or similar resistance records of pathogenic bacteria. b. In case of suspicion, epidemiological investigations are carried out with the aim of proving or disproving that these strains are indeed derived from subject-to-subject transmission. This type of investigation consists specifically in investigating whether the sampled subjects have had recent contact, shared the very same location (for example, the very same operating room or the very same hospital room), or been cared for by the very same caregiver. This type of investigation is usually lengthy, laborious, and manpower-intensive. Moreover, investigations can cause significant disruption to the operations of the institution or company suspected of being the subject of the epidemic, since preventive measures, for example, putting a room or department under quarantine or closing an operating room, are usually introduced before the investigation is completed.

[0003] In this context, the advent of sequencing, especially of the WGS (Whole Genome Sequencing) type, represents a remarkable development in bacterial epidemiology, since complete bacterial genomes contain a much higher level of information than those provided by classical microbiological techniques. Not only are the criteria for deciding to embark on epidemiological studies more precise, but in addition, the use of genomics can also greatly simplify and normalize downstream processes. For example, if two strains of Staphylococcus aureus found in samples within the same hospital department several days apart are strictly identical from a genomic standpoint, it can be determined without any additional information that the two strains do indeed form part of one and the same bacterial outbreak.

[0004] Although sequencing has shown to be a remarkable development, as it is, it does not allow to determine the lineage of two strains, regardless of their type. In fact, certain bacterial species have moldable genomes that evolve very quickly over the course of a few days, even more so when antibiotic treatment is used, so that strict discrimination between genomes cannot be used as the only criterion. To take this molding force into account, a method for detecting bacterial outbreaks consists of evaluating whether strains belong to exactly the same outbreak if the difference in their genomes, calculated for example as a function of the number of single nucleotide polymorphisms, is lower than a given threshold, as described in the December 2018 paper "Beyond the SNP threshold: identifying outbreak clusters using inferred transmission" by J. Simson et al., but this approach is rather inaccurate due to many sources of uncertainty, such as, for example, the conditions in which bacteria evolve, or the variability of mutation rates depending on the type. The authors of this paper therefore also propose to take into account the time sequence of the collection of samples, including a priori knowledge information about the strains and the mechanisms of mutation and transmission of the strains.

[0005] In addition to making epidemiological prediction models more complex, the use of a single threshold inevitably leads to a difficult compromise between sensitivity and specificity of prediction. On the one hand, if the prediction of the assignment to a bacterial outbreak is too sensitive but too non-specific, it will too often trigger epidemiological investigations that lead to the disproval of the epidemic nature of the event, which entails high costs in terms of resources, work and budget. On the other hand, if the prediction of the assignment to an outbreak is insensitive, the bacterial outbreak will go undetected, with serious consequences in terms of health, e.g. the health of the patient or consumer. Summary of the Invention

[0006] The aim of the present invention is to propose a method for identifying and monitoring bacterial outbreaks, based on the comparison of bacterial genomes, which, while offering freedom in terms of sensitivity and specificity, explicitly takes into account sources of uncertainty in the prediction of the attribution of strains to bacterial outbreaks.

[0007] To this end, the present invention relates to a method for detecting and monitoring bacterial outbreaks, associated with bacterial species in a geographical area, - obtaining digital genomes of strains sampled within a geographical area and belonging to a bacterial species; - calculating the genomic distance of the digital genome obtained using the digital genomes of a database called "epidemiological" that contains at least one digital genome of a strain belonging to a bacterial species; - the sampled strains and the strains in the database belong to a bacterial outbreak if their genomic distance is below a first predetermined threshold; the sampled strains and the strains in the database do not belong to the bacterial outbreak if their genomic distance exceeds a second predefined threshold that is strictly higher than the first threshold, or the sampled strains and the database strains probably belong to the bacterial outbreak if their genomic distance is between the first and second thresholds; and predicting According to the method, the first threshold is equal to or greater than the third threshold, and therefore the prediction that two strains having a genomic distance below the third threshold belong to a bacterial outbreak has the greatest specificity; - The second threshold is less than or equal to the fourth threshold, therefore the prediction that two strains with a genomic distance above the fourth threshold do not belong to a bacterial outbreak has the greatest sensitivity.

[0008] In other words, two different thresholds are used to control the sensitivity and specificity of the method, the lower threshold is used to control the specificity of the prediction that a strain belongs to a bacterial outbreak (hereinafter "attribute specificity"), and the upper threshold is used to control the sensitivity of this prediction (hereinafter "attribute sensitivity"). The zone between these two thresholds is therefore specifically provided to take into account the uncertainty inherent in predictions based on genomic distances. In particular, the third and fourth thresholds, applied beforehand to maximize the attribute specificity and sensitivity, define a zone where it is difficult to know whether strains belong to one and the same outbreak or not due to incomplete or poorly distributed data to learn these thresholds, not knowing which mutation mechanisms are heterogeneous within a bacterial species, methodological imprecision for the selection of genomic comparison methods, or other errors in the characterization of the epidemiological center resulting from epidemiological surveys. This uncertainty zone provides the user with flexibility in epidemic management. Specifically, in contrast to a prediction of attribution to bacterial outbreaks, which triggers epidemiological investigations and preventive measures to suppress bacterial outbreaks if the strains are in the intermediate zone, the user may set up a preliminary investigation, for example by cross-checking with the file of the patient from whom the sample was obtained, or by analyzing their antibiotic resistance sources, their virulence-contributing genes, or their phylogenetic position in the species biodiversity, to determine whether a full epidemiological investigation should be undertaken. Also, the zone between the third and fourth thresholds may be too wide in certain cases, so predictions based on these thresholds are not optimal. The first and second thresholds, which define the zone strictly contained between the third and fourth thresholds, allow analytical optimization of the prediction of attribution or non-attribution to bacterial outbreaks.

[0009] According to one embodiment, the first and second thresholds are: - constructing a learning database of digital genomes of strains belonging to a bacterial species, said database comprising: Strain pairs that have been previously determined to belong to the exact same bacterial outbreak and tagged as "related strain pairs"; strain pairs that are pre-determined not to belong to the exact same bacterial outbreak and are tagged as "unrelated strain pairs"; - selecting a binary predictor configured to predict that the two strains are related or not related by comparing their genomic distance against a fifth threshold; for each fifth threshold value in the predetermined set of fifth threshold values, a confusion matrix of said predictors as a function of a training database; a first quality metric of the predictor as a function of the confusion matrix, the first quality metric being different from the sensitivity and specificity of the predictor; Calculating a second quality index different from the first index as a function of the confusion matrix, the second quality index being different from the first index, the sensitivity and specificity of the predictors; - finding a first value of a fifth threshold that optimizes the first metric and a second value of the fifth threshold that optimizes the second metric; - setting the first threshold equal to the minimum of the first and second values ​​of the fifth threshold and setting the second threshold equal to the maximum of the first and second values ​​of the fifth threshold; is equal to the two genomic distances calculated by

[0010] In other words, a prediction based on the maximum specificity and specificity of an attribute does not necessarily constitute an optimal prediction with respect to the available epidemiological data stored in the training database.By calculating first and second thresholds that optimize a binary prediction quality, in fact, an optimization of the management of an epidemic event is obtained, while preserving a sufficiently wide intermediate zone to continue to alert the user of a possible bacterial outbreak.

[0011] According to one embodiment, the first index is selected by considering the imbalance between the number of related strain pairs and the number of related strain pairs in the learning database. Specifically, the first index is the Matthews correlation coefficient or F1 score. Usually, data on bacterial outbreaks, i.e. the number of strains considered to be related, is significantly less than the number of strains considered to be unrelated. By using a quality index that explicitly considers this imbalance, a better optimization of prediction is obtained. Moreover, the threshold value corresponding to the Matthews coefficient or F1 score favors specificity, but does not only consider specificity.

[0012] According to one embodiment, the second index is the Youden index. This index, which explicitly considers specificity and sensitivity, allows the prediction of non-attributes, whose learning is usually performed on important data, to be naturally optimized. The imbalance of the database has the effect that the Youden index is more influenced by sensitivity, and the specificity is close to 1 in the complete interval between the third and fourth thresholds.

[0013] In one embodiment, the predictor is: - a true positive corresponds to a pair of related strains with a genomic distance below the fifth threshold, - a false negative corresponds to a pair of related strains having a genomic distance above a fifth threshold, - a false positive corresponds to an unrelated strain pair having a genomic distance below the fifth threshold, - true negatives correspond to unrelated strain pairs with a genomic distance above the fifth threshold, It is selected as follows.

[0014] According to one embodiment, the epidemiological database comprises a training database, in other words the training database is supplemented as the method is applied, allowing the various thresholds to be refined as the database increases in size.

[0015] According to one embodiment, the genomic distance is a normalized distance. More specifically, the genomic distance between two strains is: - selecting primarily at a set of loci, the loci being common to the digital genome of said strain; - counting the number of allelic differences between the two digital genomes of said strains at common loci; - dividing the number of differences by the number of common loci; It is calculated by:

[0016] Normalizing by the number of common loci reduces the effect of sequencing errors, specifically the fact that it does not distinguish between loci in strains.

[0017] According to a preferred embodiment, the first and second values ​​of the fifth threshold, if greater than 0.1, - the second threshold is set equal to 0.1, - The first threshold is max(D g \D g <0.2), where max(D g \D g <0.2) is the maximum genomic distance among a pair of related strains that is strictly below 0.2.

[0018] Specifically, the inventors found that values ​​above 0.1, which are usually obtained due to incomplete or insufficient diversity of the training database, result in failure of training. The inventors further showed that in the context of a suitable training database, the first and second thresholds are equal to or less than 0.1. One of the two thresholds is therefore fixed at this upper limit. Furthermore, the inventors found that two strains of the same subtype have a genomic distance of less than 0.2 in the vast majority of cases. Thus, max(d r \d rBy setting another threshold equal to 0.01 < 0.2) for two strains with a genomic distance greater than the latter, it is predicted that these strains do not belong to the same bacterial subtype and therefore to the same outbreak, which constitutes an important indicator for suspecting an epidemic. Thus, even though the data is still insufficient to accurately calculate the first and second thresholds, the user has a way to do what he wants by default.

[0019] According to one embodiment, the distance between the digital genomes is calculated as a function of a database of markers, in particular the genetic or SNP databases wgMLST, cgMLST, MLST.

[0020] According to one embodiment, if a sampled strain is predicted to belong to a bacterial outbreak, it is tagged in the epidemiological database as "associated" with strains in the bacterial outbreak and "unassociated" with other strains.

[0021] According to one embodiment, if the sampled strain is predicted to likely belong to a bacterial outbreak, additional characterization of the sampled strain is performed to determine if the sampled strain actually belongs to the outbreak, and if so, the sampled strain is tagged in an epidemiological database as "related" to strains in the bacterial outbreak and "unrelated" to other strains.

[0022] According to one embodiment, the first and second thresholds are recalculated periodically and / or immediately after N new strains are added to the epidemiological database, where N is an integer equal to or greater than 1.

[0023] According to one embodiment, if the strain is predicted to belong to a bacterial outbreak, preventative measures are taken to prevent the outbreak.

[0024] The invention will be better understood on reading the description that follows, given purely by way of example, and by reference to the accompanying drawings in which like references indicate like elements, in which: [Brief description of the drawings]

[0025] [Figure 1] 2 is a flow chart of an embodiment of a method according to the present invention; [Diagram 2] 1 is a table of correspondence relationships between strains stored in a training database. [Diagram 3] Confusion matrix of binary predictors predicting the related or unrelated status of two strains. [Figure 4] FIG. 4: Distribution of the number of related and unrelated strain pairs as a function of their genomic distance and the threshold Ti used to calculate the confusion matrix of FIG. 3. [Diagram 5] FIG. 1 shows different thresholds across genomic distances used by the method according to the invention. [Figure 6] FIG. 1 is a diagram of a computing and sequencing system for carrying out the method according to the present invention. [Figure 7] A and B are distribution plots of the number of unrelated strain pairs (upper distribution) and the number of related strain pairs (lower distribution) for the bacterial species Clostridium difficile, and B is a zoomed-in view of A from 0 to 0.1. [Figure 8] 7A and B show genomic distance for various optimal values ​​of quality indices including sensitivity, specificity, precision, accuracy (i.e., (TP+TN) / (N+P)), F1 score, Youden index, and Matthews correlation coefficient for Clostridium difficile species, and B is a zoomed-in view of 0 to 0.1 of FIG. 7B. [Figure 9] A and B are distribution plots of the number of unrelated strain pairs (upper distribution) and the number of related strain pairs (lower distribution) for the bacterial species Staphylococcus aureus, and B is a zoomed-in view of A from 0 to 0.1. [Figure 10]A and B show genomic distances for various optimal values ​​of quality indices including sensitivity, specificity, precision, accuracy, F1 score, Youden index, and Matthews correlation coefficient for Staphylococcus aureus species, and B is a zoomed-in view of B from 0 to 0.1. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0026] In the following, unless strictly specified, "below" or "less than" means "equal to or less than", and "above" or "greater than" means "equal to or greater than".

[0027] An embodiment of the present invention will now be described in connection with the detection and monitoring of microbiological foci of infection of specific bacterial species in hospitals.

[0028] With reference to Figure 1, the method comprises a first step 10 of learning at least two thresholds, denoted S1 and S2, based on which genomic comparisons have been performed for determining whether a strain belongs to a bacterial outbreak or not, and a second step 20 of executing the method according to the invention, parameterized with the thresholds learned in step 10. More specifically, the method comprises: g Based on comparison of genomic distances shown as (BSi, BSj).

[0029] Step 10 is 12, - Digital genomes of the different strains BS1, BS2, BS3,...,BSN belonging to the species; - a correspondence table, shown in Fig. 2, linking each strain of the database to all the other strains, in which each link between two strains in the database may assume a "related" state (black squares) if the two strains have been previously determined to belong to one and the same bacterial outbreak, and a "not related" state (white squares) if the two strains have been previously determined not to belong to one and the same bacterial outbreak, the link state between two strains being determined, for example, during a prior epidemiological investigation. Furthermore, the link of a strain to itself is constant in the "related" state. As can be seen in Fig. 2, several centers of infection of the species in question may be taken into account to determine the "related" and "not related" states of strains in the training database. As will be explained below, the training database may further include strains that have been determined to be "related" but have not been diagnosed as belonging to any bacterial outbreak. The said table further includes the genomic distance D between each pair of strains BSi and BSj of the training database. g (BSi, BSj), - a table listing all of the identified epicenters of infection with their associated strains; - antibiotic resistance sources (set of genetic markers that contribute to bacterial sensitivity or antibiotic resistance) and virulence contributing genes (set of genetic markers that contribute to bacterial virulence) of strains BS1, BS2, BS3, ..., BSN; We start by generating a training database of the species in question, including

[0030] The genome of the strain is - obtaining a sample from a patient carrying the strain; - preparing an isolate of the strain, for example by spreading and culturing a sample on an agar medium to grow colonies of the strain; - taking a portion of the colony and preparing a volume to be examined for sequencing (e.g. lysis to release the bacterial DNA, amplification of the released DNA if necessary, and preparation of libraries for sequencing techniques that require it); - sequencing (or WGS sequencing), preferably of the complete DNA, using a "next generation sequencing" type technology, such as, for example, using the "MiSeq" sequencing platform available from the company Illumina Inc., San Diego, California, to generate digital sequences commonly referred to as "reads"; - optionally assembling the reads to generate an assembled sequence, known by the term "contig"; - characterization of the genome in the form of contigs or reads, commonly referred to as a "wgMLST profile", by wgMLST ("whole genome multilocus sequencing format") technology; As known per se, this characterization consists of locating the loci in the genome from a set of predefined loci and, for each identified locus, determining the alleles that correspond to this locus. The wgMLST technique is described, for example, in the article "MLST revisited: the gene-by-gene approach to bacterial genomics" by Martin CJ Maiden, published in Nature Reviews Microbiology in 2013.

[0031] The learning continues with the calculation of thresholds S1 and S2 depending on the learning database. More specifically, this calculation is - Genomic distance D g A first predictor of attribution or non-attribution of two strains to bacterial outbreaks based on a single threshold T that divides the space of genomic distances over (BSi, BSj) into just two intervals, f T TIFF0007644082000001.tif14170 - Genomic distance D g A second predictor of attribution or non-attribution of the two strains to bacterial outbreaks, g, based on two thresholds S1 and S2 that divide the space of genomic distances across (BSi, BSj) into three intervals. S1,S2 It consists of converting it to TIFF0007644082000002.tif22170.

[0032] In a preferred variant, the first predictor f T teeth, It is specified as TIFF0007644082000003.tif17170, The second predictor is It is specified as TIFF0007644082000004.tif23170.

[0033] Preferably, the genomic distance D g (BSi, BSj) is the normalized distance, so it is between 0 and 1, a. Identifying the loci that the two strains BSi and BSj share in common in their wgMLST profiles; b. if at least one allelic difference is found, for each common locus, determine whether an allelic difference exists between the two strains, and in this case, increment an allelic difference counter Compt by 1; c.D from the following formula g Calculate (BSi, BSj) and where N lc is the number of common loci. TIFF0007644082000005.tif13170

[0034] The calculation of the thresholds S1 and S2 is performed in 14, for example, -4 For each value Ti in the set of values ​​{T1,T2,...,TM} of threshold T from 0 to 1 with increments of T We start by computing the confusion matrix MC(Ti) for the threshold Ti. The computation of the confusion matrix (Ti) shown in FIG. 3 is shown in FIG. 4 for the threshold Ti: -D g true positives, denoted by “TPis”, which is equal to the total number of related strain pairs in the database such that (BSi, BSj) ≤ Ti; -Dg false negatives, denoted as “FNi”, which is equal to the total number of related strain pairs in the database such that (BSi, BSj)>Ti; -D g false positives, denoted as “FPi”, which is equal to the total number of unrelated strain pairs in the database such that (BSi, BSj)≦Ti; -D g true negatives, denoted by “TNi”, which is equal to the total number of unrelated strain pairs in the database such that (BSi, BSj) ≦ Ti; This consists of counting the

[0035] Once the set of confusion matrices {MC(T1),MC(T2,...,MC(TM))} is calculated, the method proceeds to calculate different thresholds, as shown in FIG. - predictor f T , i.e., maximizing the specificity of the prediction that the two strains are related, i.e. TIFF0007644082000006.tif14170, where N is the number of unrelated strain pairs; and a threshold S3. - predictor f T , i.e., the sensitivity of the prediction that the two strains are unrelated is maximized. TIFF0007644082000007.tif14170, where P is the number of related strain pairs, and a threshold S4; - predictor f T The first quality index is optimized, which, unlike sensitivity and specificity, explicitly takes into account the imbalance between the numbers P and N and is preferably the Matthews correlation coefficient (MCC), i.e. A threshold S1 is TIFF0007644082000008.tif17170, - predictor f T A second quality index is optimized, which is different from sensitivity and specificity and is preferably the Youden index, i.e. A threshold S2 is TIFF0007644082000009.tif14170, Continue using

[0036] A step 18 of checking the quality of the thresholds S1 and S2 is then performed. More specifically (the symbol "\" means "such that"): - if the thresholds S1 and S2 are less than or equal to 0.1, they are stored, meaning that the learning database is suitable for their calculations and their subsequent use; - If the thresholds S1 and S2 differ by more than 0.1 or less than 1%, their values ​​are set to 0.1 and M=max(D g (BSi,BSj)\(D g (BSi,BSj)<0.2), where max(Dg(BSi,BSj)\(D g (BSi, BSj)<0.2) is the maximum genomic distance closest to 0.2 between pairs of related strains in the training database in this case, - if one of the thresholds S1 or S2 is greater than 0.1, then this threshold is set to a minimum of 0.1, and if this minimum value differs from the other threshold (e.g., by more than 1%), then max(D g (BSi,BSj)\(D g (BSi,BSj)<0.2), otherwise the threshold is fixed to the maximum of these two values.

[0037] For simplicity, in what follows we will assume that threshold S1 is below threshold S2, such that these thresholds divide the space of genomic distances into three intervals, as illustrated in FIG. 4. - Lower interval]0,S1]. If the genomic distance between two strains is included in this interval, these strains are predicted to be "related". (g S1,S2 =1) - upper interval]S2,1]. If the genomic distance between two strains falls within this interval, these strains are predicted to be "unrelated". (g S1,S2 =-1) - Intermediate interval]S1,S2]. If the genomic distance between two strains is included in this interval, these strains are predicted to be "potentially related". (g S1,S2=0)

[0038] The thresholds S1 and S2 are then stored in a computer memory of a computer system used to perform step 20 described herein, said system further comprising a learning database. Step 20, which is performed in a hospital to detect and monitor the nature of bacterial epidemics, is for example performed systematically immediately after a patient is affected by a bacterial infection, and environmental samples contain pathogenic bacteria or a patient presenting symptoms that are identical or similar to another patient in the hospital. Other criteria can of course be used to trigger this step.

[0039] Step 20 begins with taking a sample containing a pathogenic strain at 22, continuing this sampling if it has not already been performed, and beginning with sequencing the strain at 24 and establishing its wgMLST profile as described in relation to step 12. At 26, the genomic distance D between the sampled strain and each of the strains in the training database is calculated. g (BSi, BSj) are then calculated. Then, at 28, a first epidemiological diagnosis is made. More specifically, - If the sampled strain is not related to any strain in the database, i.e., any strain in the database is within a genomic distance D from the sampled strain g If (BSi, BSj) is above a threshold S2, the sampled strain is determined to not belong to any bacterial outbreak. - If the sampled strain is related to a strain in the database, i.e. these two strains have a genomic distance D below a threshold S1 g If the patient has (BSi, BSj), an alarm is triggered for the user's attention and a deeper epidemiological investigation 30 is initiated, as well as, if applicable, preventive measures to combat the transmission of the sampled strain within the hospital. - If the sampled strain is potentially related to a strain in the database (indicated by "Related?" in Figure 1), i.e., these two strains are within a genomic distance D between thresholds S1 and S2.g If (BSi, BSj), then an auxiliary analysis is performed in 34 to remove the uncertainty regarding the link between these two strains. The antibiotic resistance sources and virulence contributing genes of the sampled strain are preferably determined and then compared with the antibiotic resistance sources and virulence contributing genes of potentially related strains. If the antibiotic resistance sources and virulence contributing genes are in sync, the strains are determined to be related, an alarm is triggered and a deeper investigation 30 is performed. If not, the strains are determined to be unrelated. Finally, if this comparison does not solve the problem, a deeper investigation 30 is performed. Other data may be used in this auxiliary investigation, such as, for example, the time elapsed between taking the sample and getting the strain in the database, the number of different SNPs in the formative genes, etc.

[0040] As is known per se, one of the objectives of the investigation 30 carried out by the epidemiology team of the hospital is to determine whether the various strains sampled in the hospital constitute an epidemic. At the completion of this investigation, the link between the various strains, i.e. "associated" or "unassociated", is definitively established. Furthermore, if an epidemic is detected, the epidemic strain is also tagged according to this epidemic. The genome of the sampled strain, its wgMLST profile, antibiotic resistance sources and virulence contributing genes, its links with other strains in the database, as well as data on bacterial occurrence, are then stored in the learning database so that they can be used later. The thresholds S1 and S2 can therefore be updated periodically or with each new entry in the database in order to improve their values.

[0041] FIG. 6 shows a computing and sequencing system 40 for carrying out the method according to the invention. The system 40 comprises a sequencing platform 42 for sequencing the bacterial DNA of a sample 44, thus generating a set of digital sequences or "reads". The platform 42 is connected to a data processing unit 46, for example a personal computer that receives the sequences and optionally applies a program for assembling the reads to generate contigs. The unit 46 is also connected to a remote server 48, for example in the form of a cloud solution, using software as a service (or "Saas"). The unit 46, where the "front-end" software runs, sends to the server 48 the genomes sequenced by the platform 42, in the form of reads or contigs. The server 48, where the information service runs in the form of a "back-end" and is connected to a learning database 50, receives the genomes and carries out the processing steps of the method according to the invention (for example steps 14-18 and 24-32 of FIG. 1), the server storing in its computer memory the instruction sets necessary for carrying this out. The server sends back the processing results to the unit 46 in the form of a report 52. The system 40 further comprises one or more servers 54 connected to the unit 42, in particular servers of computer systems storing patient and epidemiological data, which are used in deeper investigations to characterize epidemiological bacterial outbreaks.

[0042] Figures 7 and 9 show the distribution of the number of related and unrelated strain pairs of Clostridium difficile (Figures 7A and 7B) and Staphylococcus aureus (Figures 9A and 9B), respectively. As can be seen from these figures, there are related strain pairs with large genomic distances (e.g., Clostridium difficile above 0.6) and unrelated strain pairs with small genomic distances (e.g., Staphylococcus aureus below 0.2). Thus, when a single threshold is used, there is a zone where the genomic distance can code for both a "related" or "unrelated" state. This intermediate zone naturally exists and corresponds to strains that, for example, belong to exactly the same subtype but are not determined to belong to exactly the same bacterial outbreak. It is also observed from Figures 8A-B and 10A-B that for choosing thresholds S3 (maximum specificity, indicated by "specificity") and S4 (maximum sensitivity, indicated by "sensitivity") to divide the space of genomic distances into three, the intermediate zone is large and a good number of strains are determined to be potentially related. Note that using thresholds S1 (e.g., maximizing the Matthews coefficient MMC) and S2 (e.g., optimizing the Youden index), which optimize the quality of the prediction, the intermediate area is significantly reduced, while maintaining very good sensitivity overall.

[0043] The application to epidemiology of pathogenic bacteria in hospitals has been described. Naturally, the invention is not limited to this application and can be used in the fields of industrial (e.g. control in the food and beverage industry), environmental and veterinary microbiological control.

[0044] The use of wgMLST profiles to calculate genomic distances has been described. Other profiles, such as cgMLST ("core genome multilocus sequencing format") profiles, MLST, collections of SNPs or genes, can be used.

[0045] The use of Youden's index and Matthews correlation coefficient has been described. Other quality indices can be used, such as F1 score (i.e., 2TP / (2TP+FP+FN)), coefficient χ1, accuracy (i.e., (TP+TN) / (N+P)), precision (i.e., TP / (TP+FP)). It is preferred that at least one of these indices takes into account the imbalance of the database.

[0046] A training database has been described which is also used to compare with the sampled strains. As a variant, a separate database, or "epidemiological database", can be used to process the sampled strains. Such a database is suitable for example for hospitals, institutions, companies, etc., and the training database is then used only to establish the threshold value.

Claims

1. 1. A method for detecting and monitoring a bacterial outbreak associated with a bacterial species within a geographic area, comprising: obtaining digital genomes of strains sampled within said geographic area and belonging to said bacterial species; calculating the genomic distance of said digital genome obtained from the digital genomes of a database called "Epidemiological" that contains at least one digital genome of a strain belonging to said bacterial species; The following, i.e. the sampled strains and the strains in the database belong to the bacterial outbreak if their genomic distance is below a first predefined threshold; said sampled strains and said strains in the database do not belong to said bacterial outbreak if their genomic distance exceeds a predefined second threshold that is strictly higher than the first threshold; or the sampled strain and the strain in the database likely belong to the bacterial outbreak if their genomic distance is between the first threshold and the second threshold; and predicting According to the method, said first threshold being equal to or greater than a third threshold, such that the prediction that two strains having a genomic distance below said third threshold belong to said bacterial outbreak has maximum specificity; The method, wherein the second threshold is less than or equal to a fourth threshold, such that the prediction that two strains having a genomic distance above the fourth threshold do not belong to the bacterial outbreak has the greatest sensitivity.

2. The first and second threshold values ​​are: and constructing a training database of digital genomes of strains belonging to said bacterial species, said training database comprising: Strain pairs that were previously determined to belong to the exact same bacterial outbreak and tagged as "related strain pairs"; constructing a training database that includes strain pairs that are pre-determined not to belong to the exact same bacterial outbreak and are tagged as "unrelated strain pairs"; selecting a binary predictor configured to predict that the two strains are related or not related by comparing their genomic distance against a fifth threshold; For each fifth threshold value in the predetermined set of fifth threshold values, a confusion matrix of the predictors as a function of a training database; a first quality metric of the predictor as a function of the confusion matrix, the first quality metric being different from the sensitivity and specificity of the predictor; calculating a second quality index different from the first quality index as a function of the confusion matrix, the second quality index being different from the first quality index of the sensitivity and specificity of the predictor; finding a first value of a fifth threshold that optimizes the first quality metric and a second value of a fifth threshold that optimizes the second quality metric; setting the first threshold equal to the minimum of the first and second values ​​of a fifth threshold and setting the second threshold equal to the maximum of the first and second values ​​of a fifth threshold; The method of claim 1, wherein the distance is equal to two genomic distances calculated by:

3. The method of claim 2 , wherein the first quality index is selected taking into account an imbalance between the number of related strain pairs and the number of related strain pairs in the training database.

4. The method of claim 3 , wherein the first quality index is a Matthews correlation coefficient or an F1 score.

5. The method according to claim 2 , wherein the second quality index is Youden's index.

6. The predictor is: a true positive corresponds to the related strain pair having a genomic distance below the fifth threshold; a false negative corresponds to the related strain pair having a genomic distance above the fifth threshold; a false positive corresponds to the unrelated strain pair having a genomic distance below the fifth threshold; A true negative corresponds to the unrelated strain pair having a genomic distance above the fifth threshold. The method according to any one of claims 2 to 5, wherein the method is selected such that

7. The method of claim 2 , wherein the epidemiological database comprises the training database.

8. The method of claim 1 , wherein the genomic distance is a normalized distance.

9. The genomic distance between two strains is selecting primarily at a set of loci that are common to the digital genomes of the strains; counting the number of allelic differences between the two digital genomes of the strains at the common loci; and dividing the number of differences by the number of said loci in common; The method of claim 8, wherein the calculated value is:

10. If the first and second values ​​of the fifth threshold are greater than 0.1, the second threshold is set equal to 0.1; The first threshold value is max(D ｇ \D ｇ <0.2), where max(D ｇ \D ｇ 9. The method of claim 4 or 5, wherein the maximum genomic distance between related strain pairs is strictly below 0.

2.

11. 11. The method according to any one of claims 1 to 10, wherein the distance between the digital genomes is calculated as a function of a database of markers, in particular the genetic or SNP databases wgMLST, cgMLST, MLST.

12. 12. The method of any one of claims 1 to 11, wherein if a sampled strain is predicted to belong to the bacterial outbreak, it is tagged in the epidemiological database as being "associated" with the strain of the bacterial outbreak and as being "unassociated" with other strains.

13. 13. The method of any one of claims 1 to 12, wherein if the sampled strain is predicted to likely belong to the bacterial outbreak, further characterization of the strain is performed to determine if the sampled strain actually belongs to the outbreak, and if so, the sampled strain is tagged in the epidemiological database as "related" to the strains of the bacterial outbreak and "unrelated" to other strains.

14. 14. The method of any one of claims 1 to 13, wherein the first and second thresholds are recalculated periodically and / or immediately after N new strains are added to the epidemiological database, where N is an integer greater than or equal to 1.

15. 15. The method of any one of claims 1 to 14, wherein if a strain is predicted to belong to said bacterial outbreak, preventative measures are taken to prevent said outbreak.

Images