Methods for epidemiological identification and surveillance of bacterial outbreaks

Abstract

The present invention relates to a method for detecting and monitoring bacterial outbreaks, comprising predicting that if their genomic distance is below a first predetermined threshold, the collected strains and the strains from the database belong to a bacterial outbreak; if their genomic distance is above a second predetermined threshold, the second threshold is strictly higher than the first threshold, the collected strains and the strains from the database do not belong to a bacterial outbreak; if their genomic distance is between the first and second predetermined thresholds, the collected strains and the strains from the database may belong to a bacterial outbreak. The first threshold is greater than or equal to 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 less than or equal to 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 to the detection and monitoring of bacterial outbreaks based on partial or complete sequencing of the strain genome, in particular the strain DNA and / or RNA. Prior art

[0002] Detection of infectious bacterial outbreaks usually consists of determining whether several strains obtained from a subject (such as a patient or an inducing animal) are caused by recent transmission of the same strain between subjects, such as from a "source" subject to several subjects, or from subject to subject. Based on classical microbiological tools, detection is usually completed in two steps:

[0003] a. A bacterial outbreak is first suspected when the isolates collected belong to the same bacterial species and have common phenotypic characteristics (e.g., the same or similar antimicrobial spectrum in terms of pathogenic bacteria);

[0004] b. And if suspected, an epidemiological investigation is carried out, aimed at proving that these strains are indeed or not caused by transmission between subjects. This type of investigation consists specifically of studying whether the subjects of the collection have had recent contact, shared the same place (such as the same operating room or the same room in a hospital), were cared for by the same caregiver, etc. This type of investigation is generally long and arduous, and involves many people. In addition, the investigation may cause considerable interference to the operation of the institution or company suspected of being an epidemiological object, because preventive measures are usually implemented before the end of the investigation, such as setting up a room or department as an isolation area, or closing the operating room.

[0005] In this context, the emergence of sequencing, especially WGS (whole genome sequencing) type sequencing, represents a significant advance in bacterial epidemiology, since whole bacterial genomes contain a far higher level of information than delivered by classical microbiological techniques. Not only are the criteria for deciding to conduct epidemiological studies more precise, but the application of genomics can also greatly simplify and standardize the latter. For example, if two strains of Staphylococcus aureus found a few days apart in samples from the same hospital department are strictly identical from a genomic point of view, it can be determined that these two strains are indeed part of the same bacterial outbreak without the need for additional information.

[0006] Although sequencing has proven to be a significant advance, it is still impossible to determine the strain of two strains, regardless of species. In fact, some bacterial species have plastic genomes that evolve rapidly within a few days, and if treated with antibiotics, they evolve even faster, so that strict consistency between genomes cannot be used as the only criterion. Taking this plasticity into account, the method for detecting bacterial outbreaks consists of evaluating whether the strains belong to the same outbreak if their genomic differences (e.g., calculated according to the number of single nucleotide polymorphisms) are below a predetermined threshold, as described in the article "Beyond the SNPthreshold:identifying outbreak clusters using inferred transmission", J. Simson et al., December 2018, but this method is quite imprecise due to many sources of uncertainty, such as in the context of bacterial evolution or mutation rate variability that varies with species. The author of the article therefore suggests that the chronological order of the sample collection containing the strain and the prior knowledge about the mutation and propagation mechanism of the strain should also be considered.

[0007] In addition to making epidemiological prediction models more complex, the use of a single threshold inevitably leads to an unfavorable compromise between prediction sensitivity and specificity. On the one hand, if the prediction of the bacterial outbreak assignment is too sensitive but too non-specific, epidemiological investigations leading to refutation of the epidemiological characteristics of the event will be triggered too frequently, which involves considerable costs in terms of resources, operations and budget. On the other hand, if the prediction sensitivity of the outbreak assignment is low, the bacterial outbreak cannot be detected, which has serious consequences in terms of health, such as patient or consumer health. SUMMARY OF THE INVENTION

[0009] The aim of the present invention is to provide a method for identification and monitoring of bacterial outbreaks based on comparative bacterial genomes, which offers freedom in sensitivity and specificity while explicitly taking into account sources of uncertainty in the prediction of strain assignment to bacterial outbreaks.

[0010] To this end, the present invention relates to a method for detecting and monitoring a bacterial outbreak associated with a bacterial species within a certain geographical area, comprising:

[0011] - obtaining digital genomes of strains collected in said geographical area and belonging to said bacterial species;

[0012] - calculation of the genomic distance of the digital genome obtained with the genomes of a database, called "epidemiology", containing at least one digital genome of a strain belonging to said bacterial species;

[0013] -predict:

[0014] o If their genomic distance is below a first predetermined threshold, the collected strain and the database strain belong to a bacterial outbreak; or

[0015] o If their genomic distance is above a second predetermined threshold, which is strictly higher than the first threshold, the collected strain and the strain in the database do not belong to a bacterial outbreak;

[0016] o the collected strain and the library strain of the data may belong to a bacterial outbreak if their genomic distance is between the first and second predetermined thresholds;

[0017] According to the method:

[0018] - the first threshold is greater than or equal to 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; and

[0019] - The second threshold is less than or equal to the fourth threshold, so that the prediction that two strains having a genomic distance higher than the fourth threshold do not belong to the bacterial outbreak has the greatest sensitivity.

[0020] In other words, 2 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 about the strain belonging to the bacterial outbreak (hereinafter "attribution specificity") and the higher threshold is used to control the sensitivity of this prediction (hereinafter "attribution sensitivity"). The area between the 2 thresholds is thus specifically provided for taking into account the inherent uncertainty of the prediction based on the genomic distance. Specifically, the third and fourth thresholds, which are applied in advance to maximize the attribution specificity and sensitivity, define areas where it is difficult to know whether the strains belong to the same outbreak, which is due to incomplete or insufficient diversity of data to learn these thresholds, the lack of knowledge of the various mutation mechanisms within the bacterial species, the imprecise method caused by the selection of the genome comparison method, or other infectious focus characterization errors caused by epidemiological investigations. This uncertainty area provides users with flexibility in epidemiological management. In particular, unlike the prediction of belonging to a bacterial outbreak that triggers epidemiological investigations and preventive measures to control the bacterial outbreak, when the strain is in the middle area, the user can set a preliminary investigation, such as repeating the patient files from which the sample was obtained, or by analyzing its drug resistance genome, its virulence group or its phylogenetic position in the species biodiversity to decide whether a thorough epidemiological investigation must be taken. Furthermore, the area between the third and fourth thresholds may in some cases be so large that predictions based on these thresholds are not optimal. The first and second thresholds define a strictly contained area between the third and fourth thresholds, enabling analytical optimization of predictions regarding belonging or not belonging to a bacterial outbreak.

[0021] According to one embodiment, the first and second thresholds are equal to two genomic distances calculated as follows:

[0022] - Constructing a learning database of digital genomes of strains belonging to bacterial species, said database comprising:

[0023] o Pairs of strains that were pre-determined to be part of the same bacterial outbreak and labeled as “related strain pairs”;

[0024] o strain pairs that were predetermined not to be part of the same bacterial outbreak and were labeled as “unrelated strain pairs”;

[0025] - selecting a binary prediction indicator configured to predict whether two strains are related or unrelated by comparing their genomic distance with a fifth threshold;

[0026] - for each value of the fifth threshold belonging to the set of predetermined values ​​of the fifth threshold,

[0027] o Calculate the confusion matrix of the prediction indicators based on the learning database;

[0028] o calculating a first quality index of the predictor based on the confusion matrix, the first index being different from the sensitivity and specificity of the predictor;

[0029] o calculating a second quality index different from the first index based on the confusion matrix, the second index being different from the first index, the sensitivity and specificity of the predictor;

[0030] - finding a first value that optimizes the fifth threshold of the first index and a second value that optimizes the fifth threshold of the second index;

[0031] - setting a first threshold value equal to a minimum value of the first value and the second value of the fifth threshold value, and setting a second threshold value equal to a maximum value of the first value and the second value of the fifth threshold value.

[0032] In other words, the predictions based on maximum specificity and attribution specificity do not necessarily constitute the best predictions with respect to the available epidemiological data stored in the learning database. By calculating the first and second thresholds that optimize the quality of the binary predictions, an optimization of the management of the epidemiological events is in fact obtained, while retaining a sufficiently wide intermediate zone to continue to alert users to possible bacterial outbreaks.

[0033] According to one embodiment, a first index is selected taking into account 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 the F1 score. Generally, the data related to bacterial outbreaks, i.e., the number of strains considered as related, is much smaller than the number of strains considered as unrelated. By using a quality index that explicitly takes into account this imbalance, a better optimization of the prediction is obtained. In addition, the threshold corresponding to the Matthews coefficient or the F1 score favors specificity, rather than considering specificity alone.

[0034] According to one embodiment, the second index is the Youden index. This index explicitly takes into account specificity and sensitivity, allowing for natural optimization of predictions about non-belongings, whose learning is usually done on important data. The imbalance of the database has the following effect: the Youden index is more affected by sensitivity, with specificity close to 1 in the complete interval between the third and fourth thresholds.

[0035] According to one embodiment, the predictive indicator is selected in the following manner:

[0036] -True positives correspond to related strain pairs with genomic distances below the fifth threshold;

[0037] - False negatives correspond to related strain pairs with genomic distances above the fifth threshold;

[0038] - False positives correspond to unrelated strain pairs with a genomic distance below a fifth threshold; and

[0039] -True negatives correspond to unrelated strain pairs with genomic distance above the fifth threshold.

[0040] According to one embodiment, the epidemiological database comprises a learning database. In other words, as the method is implemented, the learning database is supplemented, making it possible to improve various thresholds as the database size increases.

[0041] According to one embodiment, the genomic distance is a normalized distance. More specifically, the genomic distance between two strains is calculated as follows:

[0042] - selecting a locus common to the digital genomes of the strains from a set consisting mainly of loci;

[0043] - counting the number of allele differences at common loci between the two digital genomes of the strains;

[0044] - Divide the number of differences by the number of common loci.

[0045] After normalization by the number of common loci, the impact of sequencing errors, especially the actual situation of unidentified loci in the strains, was attenuated.

[0046] According to a preferred embodiment, if the first value and the second value of the fifth threshold value are higher than 0.1, then:

[0047] - the second threshold is set equal to 0.1;

[0048] - The first threshold is set equal to max(D g \D g <0.2), where max(D g \D g<0.2) is the largest genomic distance among related strain pairs, which is strictly below 0.2.

[0049] Specifically, the inventors found that values ​​above 0.1, which are usually obtained due to an incomplete or insufficiently diverse learning database, often lead to learning failures. The inventors also noticed that in the context of a suitable learning database, the first and second thresholds are less than or equal to 0.1. One of the two thresholds is thus fixed at this upper limit. In addition, the inventors found that the genome distance between two strains of the same subtype is less than 0.2 in most cases. Therefore, another threshold is set equal to max(d r \d r <0.2), the genome distance of the two strains is higher than the latter, predicting that these strains do not belong to the same subtype and therefore not to the same outbreak, which constitutes an important index of suspected epidemic. Therefore, although the data is still not precise enough to calculate the first and second thresholds, the user can control the method by default.

[0050] According to one embodiment, the distances between the digital genomes are calculated based on a database of markers of genes or SNPs, in particular the databases wgMLST, cgMLST, MLST.

[0051] According to one embodiment, when a collected strain is predicted to belong to a bacterial outbreak, it is marked in the epidemiological database as "associated" with the strain of the bacterial outbreak and "unassociated" with other strains.

[0052] According to one embodiment, when a collected strain is predicted to possibly belong to a bacterial outbreak, additional characterization of the strain is performed to determine whether it actually belongs to the outbreak, and if so, the collected strain is marked in the epidemiological database as "associated" with strains of the bacterial outbreak and "unassociated" with other strains.

[0053] According to one embodiment, the first and second thresholds are recalculated periodically and / or once N new strains are added to the epidemiological database, where N is an integer greater than or equal to 1.

[0054] According to one embodiment, when a strain is predicted to be part of a bacterial outbreak, preventive measures are implemented to stop the outbreak.

[0055] BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The invention may be better understood by reading the following description given by way of example only and with reference to the accompanying drawings, in which like reference numerals indicate like elements and in which:

[0057] - Figure 1 is a flow chart of an embodiment of the method of the present invention;

[0058] - Figure 2Table showing the correspondence between strains stored in the learning database;

[0059] - Figure 3 is the confusion matrix of binary predictors, which predict the related or unrelated status of two strains;

[0060] - Figure 4 Displays the distribution of the number of related strain pairs and the number of unrelated strain pairs according to their genomic distance and the distance used to calculate Figure 3 Threshold Ti of the confusion matrix in;

[0061] - Figure 5 Describing different thresholds of genomic distance used in the methods of the present invention;

[0062] - Figure 6 Computing and sequencing systems for practicing the methods of the invention are shown;

[0063] - Fig. 7A and 7B is the distribution of the number of unrelated strain pairs (upper distribution) and the number of related strain pairs (lower distribution) of the bacterial species Clostridium difficile, Figure 7B is a magnified value between 0 and 0.1 Fig. 7A ;

[0064] - Fig. 8A and 8B The genomic distances for different optimal values ​​of quality indices, including sensitivity, specificity, precision, accuracy (i.e., (TP+TN) / (N+P)), F1 score, Youden index, and Matthews correlation coefficient, are described for the bacterial species Clostridium difficile. Figure 8B is a magnified value between 0 and 0.1 Figure 7B ;

[0065] - Fig. 9A and 9B is the distribution of the number of unrelated strain pairs (upper distribution) and the number of related strain pairs (lower distribution) of the bacterial species Staphylococcus aureus, Fig. 9B is a magnified value between 0 and 0.1 Fig. 9A ;

[0066] - Fig. 10A and 10B Shows the genome distances for Staphylococcus aureus for different optimal values ​​of quality indices, including sensitivity, specificity, precision, accuracy, F1 score, Youden index, and Matthews correlation coefficient. Fig. 10B is a magnified value between 0 and 0.1 Fig. 10B . DETAILED DESCRIPTION OF THE INVENTION

[0068] Unless strictly specified, hereinafter, "lower than" or "less than" means "less than or equal to" and "higher than" or "greater than" means "greater than or equal to".

[0069] Embodiments of the present invention relating to the detection and monitoring of microbial foci of specific bacterial species in hospitals are now described.

[0070] refer to Figure 1 The method comprises a first step 10 of learning at least two thresholds named S1 and S2, based on which a genome comparison is performed to determine whether the strain belongs to a bacterial outbreak, and a second step 20 of the method of the invention is performed, the parameters of which are the thresholds learned in step 10. More specifically, the method is based on a genome distance comparison between two strains named BSi and BSj, named D g (BSi, BSj).

[0071] Step 10 begins with the creation of a learning database for the species in question in step 12, including:

[0072] - Digital genomes of the different strains BS1, BS2, BS3…BSN belonging to this species;

[0073] - Figure 2 The correspondence table connecting each strain of the database to all other strains is shown, wherein each association between two strains in the database can be inferred to be a "related" state (black box) when the two strains were previously determined to belong to the same bacterial outbreak, and an "unrelated" state (white box) when the two strains were previously determined not to belong to the same bacterial outbreak, and the association state between the two strains was determined, for example, during a previous epidemiological study. In addition, the association of the strains relative to themselves is fixed to the "related" state. Figure 2 It can be seen that several infectious foci of the species in question can be taken into account to determine the "related" and "unrelated" status of strains in the learning database. As described below, the learning database may also contain strains determined to be "related" but not diagnosed as belonging to any bacterial outbreak. Preferably, the table also stores the genomic distance D between each strain pair BSi and BSj in the learning database g (BSi, BSj);

[0074] - a table listing all foci of infection identified with their associated strains;

[0075] - The resistome (the set of genetic markers that contribute to the sensitivity or resistance of bacteria to antibiotics) and virulenceome (the set of genetic markers that contribute to the virulence of bacteria) of strains BS1, BS2, BS3…BSN.

[0076] The genome of the strain is preferably obtained as follows:

[0077] - taking a sample from a patient containing said strain;

[0078] - preparing an isolate of the strain, for example by spreading the sample b on an agar medium and incubating to allow colonies of the strain to grow;

[0079] - taking a portion of the colony and preparing the amount for sequencing (e.g. lysing to release bacterial DNA, if necessary, amplifying the released DNA and preparing a library for the sequencing technology that requires it);

[0080] - sequencing said DNA, preferably completely sequencing (or WGS sequencing) to generate digital sequences, usually referred to as "reads", for example using "next generation sequencing" type technologies, such as the "MiSeq" sequencing platform from Illumina Inc., San Diego, California;

[0081] - optionally, assembling said reads so as to generate a combined sequence, known by the term "contig";

[0082] - Characterization of the genome in the form of contigs or reads, usually called "wgMLST profiles", by the wgMLST ("whole genome multilocus sequence typing") technique. As is known, this characterization consists of locating loci within the genome from a predetermined set of loci and, for each locus identified, determining the allele representing this locus. The wgMLST technique is described, for example, in the document "MLST revisited: the gene-by-gene approach to bacterial genomics", Martin C. J. Maiden, Nature Reviews Microbiology, 2013.

[0083] Learning continues by calculating thresholds S1 and S2 from the learning database. More specifically, this calculation consists of the following transformations:

[0084] - Two strains are or are not the first predictor of a bacterial outbreak T , based on the genome distance D g A single threshold T in (BSi, BSj) divides the genomic distance space into only 2 intervals:

[0085]

[0086] - A second predictor of whether the two strains are part of a bacterial outbreak or not S1，S2 , based on the genome distance D g The two thresholds S1 and S2 in (BSi, BSj) divide the genome distance space into three intervals:

[0087]

[0088] In a preferred variant, the first prediction indicator f T Defined as:

[0089]

[0090] And the second prediction indicator is defined as:

[0091]

[0092] Preferably, the genomic distance D g (BSi, BSj) is the normalized distance and is thus between 0 and 1, calculated as follows:

[0093] a. Identification of common loci in the wgMLST profiles of the two strains BSi and BSj;

[0094] b. For each common locus, determine whether there is an allelic difference between the two strains, and in this case, if at least one allelic difference is found, increase the counter Compt of the allelic difference by 1;

[0095] c. Calculate D from the following formula g (BSi, BSj), where N lc is the number of common loci:

[0096]

[0097] At 14, by calculating each value Ti (for example, in increments of 10) of a set {T1, T2, ..., TM} of a threshold value T between 0 and 1 -4 ) is the confusion matrix MC(Ti) of the binary prediction indicator, and the thresholds S1 and S2 are calculated. Figure 3 The calculation of the confusion matrix for the threshold Ti shown is as follows Figure 4 is shown and consists of the following counts:

[0098] -True positives, named "TPis", equal to the total number of related strain pairs in the database, so that D g (BSi, BSj)≤Ti;

[0099] - False negatives, named "FNi", equal to the total number of related strain pairs in the database, so D g (BSi, BSj)>Ti;

[0100] - False positives, named "FPi", equal to the total number of unrelated strain pairs in the database, so D g (BSi, BSj)≤Ti; and

[0101] -True negatives, named "TNi", equal to the total number of unrelated strain pairs in the database, so that D g (BSi, BSj)≤Ti.

[0102] Once the set of confusion matrices {MC(T1), MC(T2, ..., MC(TM))} is calculated, the method continues at step 16 by calculating Figure 5 Different thresholds shown:

[0103] -Threshold S3, thus predicting the index f T The specificity of is the largest, so the prediction specificity of the correlation between the two strains is the largest, that is, Where N is the number of unrelated strain pairs;

[0104] -Threshold S4, thus predicting the index f T The sensitivity of is the largest, so the prediction sensitivity of the two strains being unrelated is the largest, that is, where P is the number of related strain pairs;

[0105] -Threshold S1, optimized prediction index f T The first quality index, different from sensitivity and specificity, explicitly considers the imbalance between the numbers P and N, preferably the Matthews correlation coefficient (MCC), i.e.

[0106]

[0107] -Threshold S2, optimized prediction index f T The second quality index, different from sensitivity and specificity, is preferably the Youden index, i.e.

[0108]

[0109] Then, a step 18 of checking the quality of the thresholds S1 and S2 is performed. More specifically (the symbol "\" means "thereby"):

[0110] - If the thresholds S1 and S2 are less than or equal to 0.1, save, indicating that the learning database is suitable for its calculation and its subsequent application;

[0111] If the thresholds S1 and S2 are higher than 0.1 or differ by less than 1%, their values ​​are fixed at 0.1 and M=max(D g (BSi, BSj)\(D g (BSi, BSj)<0.2), where max(D g (BSi, BSj)\(D g (BSi, BSj) < 0.2) in this case is the maximum genomic distance closest to 0.2 among the related strain pairs in the learning database.

[0112] If one of the thresholds S1 or S2 is greater than 0.1, then this threshold is fixed at the value 0.1 and max(D g (BSi, BSj)\(D g (BSi, BSj)<0.2). If this minimum value is different from another threshold value (such as the difference exceeds 1%), the threshold value is fixed at the maximum value of these two values.

[0113] For the sake of simplicity, it is inferred that the threshold S1 is lower than the threshold S2 below, so that Figure 4 As shown, these thresholds divide the genomic distance space into 3 intervals:

[0114] If the genomic distance between two strains is contained within this interval, these strains are predicted to be “related” (g s1，S2 =1);

[0115] If the genomic distance between two strains is contained within this interval, these strains are predicted to be “unrelated” (g S1，S2 = -1); and

[0116] If the genomic distance between two strains is contained within this interval, these strains are predicted to be “potentially related” (g S1，S2 =0).

[0117] The threshold values ​​S1 and S2 are then stored in the computer memory of a computer system for carrying out step 20, which is now described, said system also comprising a learning database. Step 20 is carried out in a hospital to detect and monitor the prevalence of bacterial species, for example, systematically as soon as a patient is affected by a bacterial infection, an environmental sample contains pathogenic bacteria, or a patient presents symptoms identical or similar to another patient in the hospital. Other criteria can of course be used to initiate this step.

[0118] Step 20 begins at 22 with taking a sample containing the pathogenic strain, and if this collection has not occurred, continues at 24 with sequencing the strain and building its wgMLST profile, as described with respect to step 12. At 26, the genomic distances between the collected strains and the strains in the learning database are then calculated. A first epidemiological diagnosis is then issued at 28. More specifically:

[0119] - If the collected strain is not related to any strain in the database, that is, no matter which strain in the database, the genome distance D from the collected strain g If (BSi, BSj) is higher than the threshold S2, it is determined that the collected strains do not belong to any bacterial outbreak;

[0120] - If the collected strain is related to the strain in the database, the genome distance between the two strains is Dg (BSi, BSj) is less than or equal to the threshold S1, an alarm is triggered to draw the user's attention and initiate further epidemiological studies 30 and, if applicable, preventive measures to combat the spread of hospital-acquired strains;

[0121] - If the collected strains are potentially related to strains in the database ( Figure 1 ), that is, if the genome distance D between the two strains g (BSi, BSj) is between thresholds S1 and S2, and a supplementary analysis is performed at 34 to eliminate uncertainty about the association between the two strains. Preferably, the resistome and virulence group of the collected strain are determined, and then compared with the resistome and virulence group of the strain that is potentially related to it. If the resistome and virulence group are consistent, the strain is then determined to be related, an alarm is triggered and a deeper study is completed 30. Otherwise, the strain is determined to be unrelated. Ultimately, in the event that the comparison does not resolve the issue, a deeper study 30 is performed. Other data can be used for this supplementary study, such as the time elapsed between taking the sample and taking the strain in the database, the number of different SNPs in the plastic gene, etc.

[0122] As is well known, one of the purposes of the study 30 conducted by the epidemiology team is to determine whether the different collected strains in the hospital constitute an epidemic. At the end of this study, the association between the different strains is clearly established, that is, "related" or "unrelated". In addition, if an epidemic is detected, the epidemic strain is also marked as a function of this epidemic. The genome, wgMLST profile, resistance genome and virulence group of the collected strains, their association with other strains in the database and the data related to the bacterial outbreak are then stored in the learning database so that they can be used later. The thresholds S1 and S2 can therefore be updated regularly or at each new entry in the database to refine their values.

[0123] Figure 6 A computing and sequencing system 40 for carrying out the method of the invention is illustrated. The system 40 comprises a sequencing platform 42, which is used to sequence the bacterial DNA of a sample 44 and thereby generate a set of digital sequences or "reads". The platform 42 is connected to a data processing unit 46, such as a personal computer, which receives the sequences and optionally applies a program for assembling the reads to generate contigs. In addition, the unit 46 is connected to a remote server 48, using software as a service (or "Saas"), for example in the form of a cloud solution. The unit 46 (on which a "front end" software runs), sends to the server 48 the genome sequenced by the platform 42, in the form of reads or contigs. The server 48 (on which an information service runs in the form of a "back end" and which is connected to a learning database 50), receives the genome and carries out the processing steps of the method of the invention (such as Figure 1The server stores in computer memory the set of instructions necessary to perform the method. The server returns the processing results to unit 46 in the form of a report 52. System 40 includes one or more servers 54 connected to unit 42, specifically, those servers of computer systems storing patient and epidemiological data, which are used for further learning to characterize the epidemiological bacterial outbreak.

[0124] Figures 7 and 9 illustrate respectively Clostridium difficile ( Fig. 7A and 7B ) and Staphylococcus aureus ( Fig. 9A and 9B ) distribution of the number of related strain pairs and unrelated strain pairs. As can be seen in these figures, there are related strain pairs with large genome distances (e.g., Clostridium difficile, more than 0.6) and unrelated strain pairs with small genome distances (Staphylococcus aureus, less than 0.2). Therefore, there is a region where, if a single threshold is used, the genome distance can indicate (coder) a "related" state or an "unrelated" state. This intermediate region exists naturally and corresponds to, for example, strains belonging to the same subtype but not judged to belong to the same bacterial outbreak. In addition, it is observed from 8A-B and 10A-B that the thresholds S3 (maximum specificity, named "specificity") and S4 (maximum sensitivity, named "sensitivity") are selected to divide the genome distance space into 3 segments, and the intermediate region is large enough to make many strains be judged as potentially related. Using the thresholds S1 (such as maximizing the Matthews coefficient MMC) and S2 (such as optimizing the Youden index) that optimize the prediction quality, it is noted that the intermediate region is significantly reduced while still maintaining excellent overall sensitivity.

[0125] An epidemiological application for nosocomial pathogens is described. Of course, the invention is not limited to this application and can be used in the field of industrial (eg food industry), environmental and veterinary microbiological control.

[0126] The use of wgMLST profiles to calculate genomic distances is described. Other profiles such as cgMLST (core genome multilocus sequence typing) profiles, MLST, collections of genes or SNPs may be used.

[0127] The use of Youden index and Matthews correlation coefficient is described. Other quality indices may 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)). Preferably, at least one of these indices takes into account the imbalance of the database.

[0128] A learning database is described, which is also used for comparison with the collected strains. As a variant, a separate database or "epidemiological database" can be used to process the collected strains. For example, such a database is suitable for a hospital, institution, company, etc., and the learning database is then used only to establish the value of the threshold.

Claims

1. A method for detecting and monitoring bacterial outbreaks associated with a bacterial species within a geographic area, the method comprising: - obtaining digital genomes of strains collected in said geographical area and belonging to said bacterial species; - calculating the genomic distance between the obtained digital genome and the digital genome of an epidemiological database containing at least one digital genome of a strain belonging to said bacterial species; -predict: o If their genomic distance is below a first predetermined threshold, the collected strain and the database strain belong to a bacterial outbreak; or o If their genomic distance is above a second predetermined threshold, which is strictly higher than the first predetermined threshold, then the collected strain and the strain in the database do not belong to a bacterial outbreak; or o If their genome distances are between the first and second predetermined thresholds, the collected strains and the strains in the database may belong to a bacterial outbreak; And according to the method: - the first predetermined threshold is greater than or equal to 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; and - the second predetermined threshold is less than or equal to a fourth threshold, so that the prediction that two strains having a genomic distance higher than the fourth threshold do not belong to a bacterial outbreak has maximum sensitivity.

2. The method of claim 1, wherein the first and second predetermined thresholds are equal to two genomic distances calculated as follows: - constructing a learning database of digital genomes of strains belonging to said bacterial species, said learning database comprising: o Pairs of strains previously identified as belonging to the same bacterial outbreak and labeled as "related strain pairs"; o strain pairs that were predetermined not to be part of the same bacterial outbreak and were labeled as "unrelated strain pairs"; - selecting a binary prediction indicator configured for predicting whether two strains are related or unrelated by comparing the genomic distance of the two strains to a fifth threshold; - for each value of the fifth threshold belonging to the set of predetermined values ​​of the fifth threshold, o Calculate the confusion matrix of the prediction indicators based on the learning database; o calculating a first quality index of the predictor based on the confusion matrix, the first quality index being different from the sensitivity and specificity of the predictor; o calculating a second quality index different from the first quality index based on the confusion matrix, the second quality index different from the first quality index, the sensitivity and specificity of the predictor; - finding a first value of the fifth threshold that optimizes the first quality index and a second value of the fifth threshold that optimizes the second quality index; - setting a first predetermined threshold value equal to a minimum value of the first value and the second value of the fifth threshold value, and setting a second predetermined threshold value equal to a maximum value of the first value and the second value of the fifth threshold value.

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

5. The method of any one of claims 2 to 4, wherein the second quality index is the Youden Index.

6. The method according to any one of claims 2 to 4, wherein the prediction indicator is selected in the following manner: - True positives correspond to related strain pairs whose genomic distance is below the fifth threshold; - False negatives correspond to related strain pairs whose genomic distance is above the fifth threshold; - False positives correspond to unrelated strain pairs whose genomic distance is below said fifth threshold; and -True negatives correspond to unrelated strain pairs with a genomic distance above the fifth threshold.

7. The method according to any one of claims 2 to 4, wherein the epidemiological database comprises the learning database.

8. The method of any one of claims 1-4, wherein the genomic distance is a normalized distance.

9. The method of claim 8, wherein the genomic distance between the two strains is calculated as follows: - selecting the loci common to the digital genomes of the strains from a set consisting mainly of loci; - counting the number of allele differences at common loci between the two digital genomes of the strains; - Divide the number of differences by the number of common loci.

10. The method of claim 4 or 9, wherein if the first value and the second value of the fifth threshold value are higher than 0.1, then: - the second predetermined threshold is set equal to 0.1; - The first predetermined threshold is set equal to max(D g \D g <0.2), where max(D g \D g <0.2) is the largest genomic distance among related strain pairs, strictly below 0.

2.

11. The method according to any one of claims 1 to 4, wherein the distances between the digital genomes are calculated based on a marker database of genes or SNPs.

12. The method of claim 11, wherein the database is a database wgMLST, cgMLST or MLST.

13. The method of any one of claims 1 to 4, wherein when the collected strains are predicted to belong to a bacterial outbreak, they are marked in the epidemiological database as "related" to the strains of the bacterial outbreak and "unrelated" to other strains.

14. The method of any one of claims 1-4, wherein when a collected strain is predicted to possibly belong to a bacterial outbreak, additional characterization of the strain is performed to determine whether it actually belongs to the outbreak, and if so, the collected strain is marked in an epidemiological database as "associated" with strains of the bacterial outbreak and "unassociated" with other strains.

15. The method of any one of claims 1-4, wherein the first and second predetermined thresholds are recalculated periodically and / or once N new strains are added to the epidemiological database, wherein N is an integer greater than or equal to 1.

16. The method of any one of claims 1-4, wherein when a strain is predicted to be part of a bacterial outbreak, preventive measures are implemented to stop the outbreak.

Images