Cytometry-based method for counting microorganisms and associated microbiological-control methods

Abstract

The invention in particular relates to a counting method for microbiological control, comprising: cytometry analysis (101) of a matrix comprising a given volume of a sample, said sample having been labelled beforehand using a labelling protocol; and obtaining (102) a multidimensional data table collating objects detected during the cytometry analysis, characterised in that the method comprises a step of applying a counting algorithm comprising at least the following steps: assigning (105) a quantitative score contained in a range of values ranging from [0 to 1] to each object detected during the analysis, depending on its probability of belonging to a defined "microorganism" class; and summing (108) the scores obtained in the preceding step, and in that the method comprises a step of: counting (109) microorganisms based on the sum of the scores.

Description

[0001] Title: Method for enumerating microorganisms based on cytometry and associated microbiological control methods.

[0002] Technical field of the invention

[0003] The invention relates to the technical field of flow cytometry, whether flow cytometry or solid-phase cytometry, and more particularly, the invention concerns the enumeration of microorganisms and associated microbiological control methods corresponding to either a quantitative or a qualitative analysis method for the matrices analyzed. The invention is primarily intended for processing flow cytometry data acquired on "complex matrices" but can also be applied to "simple matrices" without limitation.

[0004] The invention is of particular interest in the context of the enumeration of microorganisms in complex matrices by solid phase cytometry, and preferably for the microbiological control of cell suspensions found in cell and gene therapy products (lymphocytes, various types of stem cells, dendritic cells, fibroblasts, myoblasts, chondrocytes).

[0005] Definitions

[0006] In the present invention, "flow cytometry" refers to a cytometry method in which the sample, already fluorescently labeled using a labeling protocol, flows through a capillary-type fluidic channel. This capillary is often sheathed by an external liquid (called the sheathing liquid) according to the principle of hydrodynamic focusing. At a specific point in the capillary, a laser excites a portion of the flowing sample and several point detectors, oriented at 90° or 180° to the excitation beam and associated with specific collection objectives and optical filters, continuously detect the signals emitted by the interaction between light and the sample (primarily side-scatter scattering, forward scattering, and fluorescent emissions). The intersection between the laser beam, the sample capillary, and the volume interrogated by the collection optics (called the interrogation point) corresponds to a very small volume, on the order of a picoliter.This configuration, with a suitable fluid regime, allows for the simultaneous detection of all suspended particles in the sample. Flow cytometry typically allows the analysis of volumes between 100 µL and 500 µL with a Limit of Detection (LoD) around 10. 2 CFU / ml which depends on the matrix. An example of an instrument implementing flow cytometry is the D-COUNT® system marketed by bioMérieux, the operation of which will be explained later.

[0007] In the present invention, "solid-phase cytometry" refers to a cytometry method in which the sample to be tested is first filtered through a membrane. This technique requires that the sample be filterable and may therefore necessitate specific lysis preparatory steps before the filtration stage. The original sample volume can be very large, up to several tens of milliliters. Furthermore, because the remaining particles are captured on the membrane surface, this technique allows for a more detailed analysis and, for microbiological testing applications, an ultimate detection limit of one CFU per filtered volume, enabling sterility testing (qualitative microbiological control). The filtration membrane used is a so-called "surface" membrane, where particles are retained on the upper surface and not within the membrane itself.Once the objects, initially suspended in the original sample, are retained on the membrane, various operations can be performed by applying reagents either via the upper surface or even via the inner surface of the membrane, with the reagent rising by capillary action on the upper surface. Examples include counterstaining to increase contrast, activation and resuscitation of spore-forming organisms, and fluorescent labeling. For microbiological control, the fluorescent labeling process used is similar to that used for flow cytometry, namely viability labeling using fluorogenic substrates. Once prepared, this membrane can then be analyzed on a specific instrument to detect all objects present on its surface and measure various physical parameters for each object (size, shape, fluorescence intensity).This analysis comprises a first detection step, also called "screening," and an optional second, complementary confirmation step. During the first detection or "screening" step, the entire membrane is scanned using fluorescence with a specific system such as the SCANRDI® system, marketed by bioMérieux. Objects are detected, and several parameters are extracted for each, corresponding to criteria such as fluorescence intensity, the fluorescence ratio between two emission channels, and the size and shape of the objects. Business rules for discrimination are then applied to each object to classify them binary as "microorganism" or "non-microorganism." For enumeration or qualitative applications that do not require an ultimate detection limit, the analysis stops at this first step, and a result is provided based on the count of these objects classified as "microorganism."

[0008] For enumeration or qualitative applications requiring a detection limit, "microorganism" objects are individually confirmed in the second confirmation step. They are confirmed sequentially using fluorescence microscopy with a high-magnification objective (greater than 40X) and possibly on a stack of z-images to ensure proper focus. For sterility testing applications where this second step is planned, the settings of the first step are generally adjusted to increase sensitivity, even if it means detecting many non-microorganisms, knowing that these will be eliminated in the second step. This two-step detection analysis is presented in particular in the article: "Carro et al. iMSRC: Converting a Standard Automated Microscope into an Intelligent Screening Platform. Sci Rep 2015, 5 (1), 10502".

[0009] This analysis ensures excellent sensitivity and specificity and allows solid phase cytometry to be used in the context of sterility tests required by the pharmacopoeia before the release of injectable products in particular.

[0010] In the present invention, "sensitivity" refers to the ability of a qualitative method to detect contamination when it is present: no sample is falsely declared sterile. Furthermore, "specificity" refers to the ability of a qualitative method to accurately identify an uncontaminated sample: no false alarms with a sample that is falsely declared contaminated. The threshold of a test, that is, the value at which it is considered positive, influences both the sensitivity and specificity of the test. Thus, if this threshold is lowered, the test will be more sensitive but less specific. Highly sensitive tests are particularly useful for ensuring that contamination is not present (few false negatives), while highly specific tests are useful for confirming that contamination is present (few false positives).

[0011] In the present invention, "non-microorganism" means any object detected by the cytometer that does not correspond to a predetermined class of "microorganism." In the general context and without further specification, non-microorganisms correspond to reading artifacts, air bubbles, or matrix debris. In this application, the terms "point" and "object" are used interchangeably depending on the context: "point" for the cytogram, "object" for the equivalent data table. A point is an object.

[0012] In the present invention, the term "data table" refers to a table generated by flow cytometry analysis, presenting the detected objects in rows and the measured parameter values ​​in columns. This data format is standardized with the "fcs" (Flow Cytometry Standard) format (latest version fcs 3.1). Mathematically, each object can be considered a point in a multidimensional space, with each variable representing a dimension. Graphically, these points can be represented on a cytogram whose axes are the measured variables. Limited to a single variable (a single axis), the cytogram is equivalent to a classic histogram. Beyond three dimensions, graphical representation becomes difficult, but conceptually the notion of a cytogram can be preserved as a geometric analog for representing the data table, using the concepts of points, point density, and clusters.Traditionally, cytograms are represented in two dimensions, particularly when multiple fluorescent markers are used, allowing each marker to be represented on a separate axis. Even with a single marker (and its associated fluorescence emission spectrum), detection can be performed on two fluorescence channels, each with a specific spectral band, and the information related to the fluorescence emission profile of the detected objects can be used on a 2D cytogram. Cytograms can be used to identify cell populations, assess data quality, and optimize acquisition parameters. In this case, the data table (or its geometric analog, the cytogram) is used as input data in the method according to the invention.We can therefore speak indistinctly of an object (individual element detected by the cytometer = a line of the data table) or of a point, which is the mathematical representation of this object in the cytogram.

[0013] In the present invention, "business rules" refer to the flow cytometry variables that generally have a physical meaning and are directly interpretable. For example, the "Side Scatter" parameter (laser scattering measured at 90° to the excitation) is directly related to the surface texture and granular appearance of the objects, while the "Forward Scatter" parameter (laser scattering measured at 180°) is directly related to the size of the detected objects. Fluorescence parameters are also directly interpretable based on the labeling protocol. Thus, with viability labeling using a fluorogenic substrate, the fluorophore's emission spectrum is known a priori. If the fluorescent emission is measured on separate channels, one corresponding to the emission maximum and the other to the "tail" of the emission spectrum, the intensity ratio measured between these two channels will remain constant around a predefined value for the labeled objects.In a 2D cytogram with each channel on an axis and logarithmic scales, microorganisms will preferentially be positioned along the diagonal. The rules of procedure correspond to the direct physical interpretation of the cytogram based on the absolute positions of the detected points within the considered coordinate systems.

[0014] Technological background of the invention

[0015] Classically, in cytometry applications for microbiological control, detected objects are first categorized into various types (e.g., microbial cells, control beads, or particles), and then the "microorganism" objects corresponding to microbial cells are counted to establish the enumeration result.

[0016] For qualitative applications such as presence / absence (food or cosmetic applications) or sterility testing (pharmaceutical applications), this enumeration result is then converted, via a contamination threshold defined according to the requirements specific to the field of application, into a binary result: contaminated or uncontaminated sample.

[0017] For the classification phase of these objects, there are currently two main approaches, each corresponding to a way of defining regions of the cytogram resulting from the analysis. The approach differs depending on the type of cytometry (flow or solid phase).

[0018] So,

[0019] The first approach consists of defining these regions, also called "gates" or "windows", from predefined business rules and / or from a training set, and then classifying each of the objects according to its position with respect to these determined and fixed regions.

[0020] A first example of the application of the first approach for microbiological control is explained below. The D-COUNT® system is a flow cytometry system for the detection and enumeration of previously fluorescently labeled microbial cells. The D-COUNT® system measures the fluorescence signature of these targets at two different wavelengths (FL1 and FL2 channels) for the same excitation and the same fluorophore. A unique feature of the D-COUNT® system is that the FL2 channel actually measures, in addition to a fluorescence signal, a fraction of the light scattered by the system's laser, via a suitable selection of optical filters (a dichroic mirror without a bandpass emission filter in front of the detector). The algorithm counts, individually and as they are detected, the number of fluorescent objects within a predefined and fixed window that corresponds to the theoretical fluorescence signature of the microbial cells.This enumeration is reported to the analyzed volume, and is multiplied by a possible dilution factor to give the final enumeration result in CFU / ml of analyzed product.

[0021] While this first approach is robust and efficient, it presents problems with complex matrices. Indeed, despite calibration of the response, a complex matrix can alter the absolute positions of individual points and sometimes cause objects that should be considered non-microorganisms to be included in "microorganism" windows, or conversely, exclude objects that should be considered microorganisms from their "microorganism" window. These errors often necessitate an increase in the positivity threshold for qualitative applications, and therefore a reduction in sensitivity.Furthermore, in the case of double contamination by two distinct populations, for example bacteria and molds, the first approach does not allow these two different populations to be counted separately, although they may appear distinctly in a point cloud, because this approach is based on an a priori window that is wide enough to encompass all contaminants, whatever they may be.

[0022] A second example of applying the first approach to solid-phase cytometry data is the SCANRDI® system marketed by bioMérieux, specifically designed for sterility testing of pharmaceutical products. The membrane, on which fluorophore-labeled objects are present, is scanned. Several parameters (fluorescence and / or morphology) are extracted for each detected object. Business rules are applied to each object to retain only those that are potentially microorganisms. Finally, a confirmation step is performed for each retained object by capturing high-magnification (>40X) fluorescence microscopy images.

[0023] In the case of complex matrices, particularly cellular debris matrices, a very large number of objects can be detected during the initial stage, which can complicate and tedious the confirmation step. Furthermore, it is very difficult to establish discrimination rules that ensure a good compromise between sensitivity and specificity. Finally, even with high-resolution confirmatory images, there are many cases where it is difficult to definitively assign a binary classification to objects as "microorganism" or "non-microorganism."

[0024] The second approach involves directly dividing the initial point cloud, containing a plurality of objects, into subgroups using a clustering method. This clustering can be performed with various machine learning algorithms and is based on the observed distribution of objects represented by points on the cytogram, without any prior knowledge of the regions. Once the subgroups are formed, they can then be assigned, for example, into "microorganisms" or "particles" clusters, either according to predefined business rules or via a training dataset.

[0025] An example of the second approach is presented in patent application WO2021 / 154561. It describes a clustering step followed by a classification step for the clusters formed, taking into account the shape of the point clouds and the density of the points. The approach is refined in that the scores of individual points are considered through the calculation of spillover matrices. This ultimately allows for more flexible adjustment of new cluster recognition envelopes and for accounting for potentially distant points. Nevertheless, despite this approach, the entire cluster thus formed is then assigned a class. The objects (individual points) remain categorically classified according to their membership in these assigned classes.Here again, even though a continuous score (between 0 and 1) is assigned to each object, this is only an intermediate step to define zones, and ultimately, the status of each object remains qualitative, depending on its belonging to a particular region. For both approaches, once the class of each object is assigned (either directly for the first approach, or via an intermediate step of clustering the point cloud and assigning classes to subgroups for the second approach), the objects of the "microorganism" class are counted.

[0026] These approaches have drawbacks. Objects represented by isolated points or overlapping point clouds pose a critical problem, particularly for the microbiological control of complex matrices. The goal of microbiological control systems is to count all microbial cells, without exception, for a given analysis volume in order to obtain a precise quantitative result and achieve the best detection limit while minimizing false alarms. However, for complex matrices, it is not uncommon to observe shifts or distortions in point clouds, with possible overlaps (due to optical scattering phenomena related to the matrix) or even the appearance of new points (debris, air bubbles, interference between labeling reagents and the matrix).In this context, current methods ultimately assign each point (whether assigned to a cluster, overlapping several clusters or individual) a qualitative class ("microorganism" or "particle" for example); and this can significantly bias the enumeration result of the sample from the reality of its microbiological state.

[0027] On the other hand, even if all the marked objects are counted correctly, since the marking principle differs from the principle of traditional microbiology methods (based on culturing for a long incubation period), the enumeration result will not necessarily be well correlated or even comparable or interpretable in terms of "Colony Forming Units" (CFU).

[0028] Object of the invention

[0029] The invention makes it possible to resolve all or part of these drawbacks and to improve existing systems and processes.

[0030] The invention relates to a method for enumerating microbial cells based on flow cytometry or solid-phase cytometry and an algorithm for significantly reducing the influence of particles in the complex matrix, thereby reducing false alarms and improving the sensitivity of the enumeration method, as well as improving the correlation with colony-forming unit (CFU) counting. To this end, the invention relates to an enumeration method for microbiological control comprising:

[0031] Flow cytometry analysis of a matrix comprising a given volume of a sample, said sample having been previously labeled according to a labeling protocol,

[0032] Obtaining a multidimensional data table showing objects detected during flow cytometry analysis, characterized in that the method includes a step of applying an enumeration algorithm comprising at least the following steps:

[0033] Assignment of a quantitative score within a range of values ​​from [0 to 1] to each object detected during the analysis based on its probability of belonging to a determined "microorganism" class, Sum of the scores obtained in the previous step, and in that the method includes a step: Enumeration of microorganisms from the sum of the scores.

[0034] The method according to the invention has many advantages, which are set out below.

[0035] Indeed, assigning a quantitative score to each object allows for greater accuracy in enumeration, particularly in cases where, after fluorescent labeling of the sample, one or more microbial cells are relatively inactive and generate a weak fluorescence signal. The position of this cell(s) as an object may be outside a cluster of objects classified as "microorganisms," despite its proximity. Consequently, the score for this object will be between 0 and 1. For example, summing the scores of three objects outside the classified cluster, each with a score of 0.33, results in a sum of 1. This probabilistic approach is scientifically sound because it is known that not all microbial cells detected by cytometry would necessarily grow on a culture medium.

[0036] Furthermore, this method is also particularly advantageous in the case of a cytometry system combining two different types of information on the same detection channel, especially a fluorescence signal and a scattering signal, as is the case for the D-COUNT® system. Indeed, depending on the matrices addressed, the more or less scattering nature of the sample, the position of the center of the point cloud clusters, and their elongation along this axis can vary considerably, with distribution tails, overlapping point cloud clusters, and isolated points.

[0037] Furthermore, the method according to the invention is also of interest for enumerating microorganisms from cytometry systems with multiple detection channels and potential cross-correlations between measurements, for automatic selection of subpopulations, in particular by easing the constraints of compensation correction or even eliminating them, knowing that they are not based on fixed and predefined windows and that the clustering phase of the data table can be carried out on a possibly pre-processed data table.

[0038] According to one feature of the invention, the method comprises a step of:

[0039] Clustering of the objects in the data table into a plurality of clusters comprising at least one cluster called a "point cloud" and / or at least one "isolated point" by means of a calculation of at least one parameter for each cluster or isolated point, the at least one parameter being selected from a list of parameters comprising at least one position and / or shape and / or density of the cluster or isolated point,

[0040] Classification of each cluster point cloud, the class being selected from several classes including at least the class "microorganism", the classification being carried out according to at least one parameter calculated in the previous step and business rules relating to the marking protocol; each isolated point not being classified.

[0041] Advantageously, the method according to the invention applies by default to a situation with a single reference class, the class "microorganism," and therefore a binary classification into "microorganism" and "non-microorganism" (or "particle"). In this case, each object is assigned a quantitative score between 0 and 1, with 1 corresponding to the class "microorganism" and 0 to everything else ("non-microorganism" or "particle"). This is a "one-class" classification.

[0042] According to one feature of the invention, the classification step can comprise a plurality of classes. According to another feature of the invention, depending on the distance of the object from a cluster classified as a microorganism, the score "1" is assigned to objects in the cluster classified as a microorganism or those closest to the cluster classified as a microorganism, the score 0 is assigned to objects furthest from the cluster classified as a microorganism, and a score between 0 and 1 is assigned to objects positioned between objects with a score of 1 and objects with a score of 0.

[0043] According to one feature of the invention, the classification step may include a "non-microorganism" class.Thus, according to the invention, the classification step consists of classifying each point cloud cluster, the class being selected from several classes including at least the "non-microorganism" class and the "microorganism" class, the classification being carried out according to at least one parameter calculated in the previous step and business rules relating to the marking protocol; each "isolated point" cluster not being classified, and the assignment step consists of assigning a score to each object in each cluster detected during the analysis: (i) if the object is in a cluster belonging to the "non-microorganism" class: the score is 0; (ii) if the object is in a cluster belonging to the "microorganism" class: the score is 1; (iii) if the object is not classified: its score is between 0 and 1 according to its distance from a cluster including classified objects.

[0044] Advantageously, the class "non-microorganism" is a class representative of the objects of a cluster and / or isolated point that are not, and this is certain, microorganisms.

[0045] According to one feature of the invention, the classification step may include a "control beads" class. Indeed, the sample can be mixed with internal control beads before analysis, and thus a point cloud can correspond to these control beads, which are not non-microorganisms.

[0046] Furthermore, the method according to the invention is advantageous when two types of microbial contamination are present in the same sample, for example, bacteria and yeasts, or molds, etc. The method according to the invention can include several specific classes replacing the generic class "microorganism," which can make it possible to distinguish two different, separate point cloud clusters that would each be classified in the "microorganism" class if the method included only one class, but which, since there are specific classes, are specifically classified into two different specific classes.

[0047] Therefore, according to the invention, in the case where several classes (other than "microorganism" or "particle / non-microorganism") need to be identified, for example two different types of microorganisms as developed above, or two different types of objects of interest, for example microorganisms and beads, the classification is said to be "multi-class" and several scores are calculated for each object, according to a complete disjunctive coding ("one hot encoding").

[0048] For example, for a microorganism class Mi, a microorganism class M2, and a bead class B, three scores Si, S2, and S3 are calculated respectively for each object, based on three separate classification models, each one-class:

[0049] A first model allows us to distinguish class Mi from the rest, with a score Si falling within the range of values ​​[0;1] depending on its probability of belonging to class Mi.

[0050] A second model allows us to recognize class M2 in relation to the rest, with a score S2 therefore falling within the range of values ​​[0;1] depending on its probability of belonging to class M2)

[0051] A third model allows class B to be recognized in relation to the rest, with an S3 score within the range of values ​​[0;1] depending on its probability of belonging to class B).

[0052] According to one feature of the invention, the classification models are independent; the sum of the scores for each object is not necessarily equal to 1. For example, if an object is an isolated point that has a probability of belonging to each of the three clusters with a score of 0.5, the scores of said object will be 0.5; 0.5; 0.5. On the other hand, if an object is not isolated and obtains a membership score of 1 for one cluster, the other scores will be 0.

[0053] The enumeration is then performed for each category by separately summing the Si, S2, and S3 scores of each object in the cytogram. Similarly, if an activation function is applied to the scores before the summing step, the functions are not necessarily the same for the different classes.

[0054] According to one feature of the invention, the matrix is ​​a complex matrix.

[0055] In the present invention, a "complex matrix" is defined as a matrix corresponding to a sample whose composition is not controlled and / or which contains numerous particles that can be detected and possibly mistaken for microbial cells. More generally, "matrix" refers to the sample to be analyzed after any preparation steps such as dilution, centrifugation, or lysis. Thus, some samples that are initially "complex matrices" become "simple matrices" after the preparation protocol. Other samples remain complex matrices even after the preparation steps. For example, drinking water and certain physiological fluids such as urine and blood plasma are not considered "complex matrices."Some food products, such as milk or conventional yogurt, are also considered "simple matrices" (after application of a preparation protocol). Conversely, dessert creams, soups, or beverages enriched with suspended particles are "complex matrices" because, even after a preparation protocol, numerous particles remain suspended in the sample analyzed by cytometry. In the pharmaceutical field, cell matrices treated to make them filterable (particularly lysed) for solid-phase cytometry analysis contain numerous debris and are considered complex.

[0056] According to another feature of the invention, the method according to the invention includes a step of applying an activation function to the scores of each object, said activation function being configured to optimize the correlation between the enumeration of detected microorganisms and the sum of the scores. Indeed, below a certain score value, it is not relevant to include it in the sum: ten objects scored at 0.1 should not be equivalent to 1, whereas two objects scored at 0.5 should be.

[0057] According to one feature of the invention, the activation function is a "ReLu" (Rectified Linear Unit) type activation function. This activation function may include at least one adjustable parameter allowing optimization of the correlation from a training set.

[0058] According to another feature of the invention, the method according to the invention comprises one or more preprocessing steps of the data array before the clustering step. According to another feature of the invention, the preprocessing step involves sizing and / or rotating and / or distorting and / or binning the data array, which allows the data in the data array to be correctly proportioned relative to each other and facilitates the subsequent steps.

[0059] According to one feature of the invention, the preprocessing step can also be a dimensionality reduction process. For example, a Principal Component Analysis (PCA) technique, a t-distributed stochastic neighbor embedding (t-SNE) technique, or a Uniform Manifold Approximation and Projection (UMAP) technique. Advantageously, this reduction process is configured to reduce the dimensionality of the data and thus to mitigate collinearity problems, and is particularly relevant in the case of a high-dimensional dataset. The algorithm of the method according to the invention therefore allows, when applied to datasets with multiple collinear variables, a robust and unbiased classification of the different subpopulations.

[0060] According to one feature of the invention, the clustering of the data table is carried out using learning methods based on a density calculation of the objects in the data table.

[0061] According to one feature of the invention, the analysis step is a step in which fluorescence detection is performed, with the system according to the invention, on each object of the sample, and a plurality of parameters, such as the fluorescence intensity and / or fluorescence ratio between two emission channels of the system and / or size and / or shape of the objects are extracted for each object.

[0062] According to another feature of the invention, the cytometry analysis is a solid-phase analysis. In the case of a solid-phase cytometry analysis, the sample is first filtered on the surface of a membrane, said membrane then being analyzed by fluorescence imaging to detect objects.

[0063] Alternatively and according to the invention, the cytometry analysis is a flow cytometry.

[0064] According to one feature of the invention, the preparation step may further include substeps of lysis and / or centrifugation depending on the matrix, and / or filtration. According to another feature of the invention, the labeling protocol is a fluorescent labeling protocol for microorganisms potentially present in the sample. The labeling protocol is performed during sample preparation prior to flow cytometry analysis.

[0065] According to one feature of the invention, the clustering of objects in the data table can be performed using a learning method such as density analysis, and preferably HDBSCAN analysis, which is a hierarchical density-based spatial clustering of applications with noise. This technique allows, in a single phase, the establishment of both the clusters and the individual scores of each object.

[0066] The advantage of using HDBSCAN analysis for clustering is that this type of density-based clustering algorithm eliminates the need to consider the number of clusters beforehand, as well as their convexity. The principle of HDBSCAN analysis is based on the fundamentals of DBSCAN, but takes into account the different data densities. The HDBSCAN documentation provides access to numerous hyperparameters; however, we have chosen to retain only the most relevant ones to avoid an excessively lengthy hyperparameter tuning phase.

[0067] According to one feature of the invention, the hyperparameters of the algorithm for clustering are:

[0068] ■ n bins ■ P as binning grid discretization.

[0069] - minciuster size: minimum cluster density: proportion between 0 and 1 across all points of the minimum population of a cluster. The smaller this number, the more clusters with low density can appear.

[0070] - thr outUers probability threshold for an isolated point to belong to a cluster.

[0071] Advantageously, the algorithm outputs different clusters that have been detected and objects that do not belong to any cluster and are considered isolated points. The isolated points are assigned a probability of belonging to the different detected clusters by comparing this probability to a threshold. outUers If the probability is greater than the threshold outUersThen, the isolated point is assigned to the cluster. Alternatively, or in addition, the clustering of objects in the data table into point clouds can be performed using unsupervised learning methods such as parametric methods based on distribution assumptions, such as the "FlowPeaks" approach, which combines k-means and Gaussian mixtures to form proto-clusters that can then be merged. The FlowPeaks approach is described in an article in the journal Bioinformatics, Volume 28, Issue 15, August 2012, Pages 2052-2058. This method groups the data with k-means for a high number k, then correlates these groups with a multivariate Gaussian model, and finally merges the Gaussians that are close.

[0072] Alternatively or in addition, the clustering of objects in the data table into point clouds can be carried out using deep learning methods, for example in the form of a self-adaptive map.

[0073] The invention also relates to a cytometry system for microbiological analysis purposes, configured to implement the enumeration method for the microbiological control of samples according to the invention.

[0074] According to one feature of the invention, the cytometry system comprises at least one detection channel. Preferably, the cytometry system comprises only two detection channels, both measuring a fluorescence signal, which allows for a simple system with easy channel management, particularly because it is not necessary to use orthogonalization compensation methods, while maintaining equivalent performance, notably thanks to the method according to the invention.

[0075] According to one feature of the invention, object detection is achieved by analyzing the signal collected on the first detection channel, also called the "trigger channel".

[0076] According to one feature of the invention, the first detection channel, referred to as the "trigger channel," corresponds to the fluorescence measurement at the maximum emission wavelength of the fluorophore used to label microorganisms. Advantageously, the method according to the invention is implementable in a cytometry system comprising more than two channels, said channels being correlated with each other.

[0077] According to one feature of the invention, the second detection channel (FL2) is optically configured, by a suitable choice of filters, to combine, in a weighted sum for example, at least two optical parameters, such as fluorescence and scattering.

[0078] According to one feature of the invention, the optical parameters can be fluorescence parameters.

[0079] According to one feature of the invention, the second detection channel (not used for object detection) simultaneously collects at least two optical parameters. The use of such a second channel positioned on the Y-axis and this simultaneous measurement of two optical parameters is particularly advantageous because it simplifies the optical design with a single detector. However, it is generally along this axis that the relative position of the points can vary depending on the matrix. This variability prevents the accurate definition of gates / windows using a conventional method; but with the method according to the invention, where the classification model is derived from the relative positions of points, and quantitative scores are calculated for each object, this configuration is compatible with applications for enumerating microorganisms.

[0080] According to one feature of the invention, the second detection channel measures both fluorescence over a range beyond 540 nm and a proportion of light scattered by the laser at 488 nm. Indeed, there is intentionally no bandpass filter in front of the second channel, and since the dichroic mirror allows laser light to pass through in transmission, this information is combined with the fluorescence on the second channel.

[0081] According to one feature of the invention, the first channel detects fluorescence in a specific band corresponding to the maximum intensity of the fluorophore. For example, if the fluorophore is fluorescein, the emission peak is around 520 nm. According to the invention, object detection is considered to have occurred as soon as the intensity measured on the first channel exceeds a threshold; the intensity peak value is captured as an area under the peak, along with the peak height and optionally the peak width.

[0082] According to one feature of the invention, the values ​​measured on the first channel and on the second channel are correlated.

[0083] According to one feature of the invention, during object detection the signal on the second channel is acquired.

[0084] According to one feature of the invention, the trigger channel of the cytometry system does not correspond to the measurement of the diffusion of objects on this system (Side Scatter or Forward Scatter), because as the system is dedicated to the analysis of a priori complex matrices, with many objects in suspension, capturing all the objects which have a diffusion signal would saturate the detector very quickly with the risk of missing the fluorescent objects.

[0085] In fact, points positioned in a Y shape on the cytogram can be found at varying heights depending on both: their diffusion (related to the size of the objects) and their fluorescence signature (emission in the yellow / orange range).

[0086] Therefore, there is significant individual variability among the points on this Y-axis, and their position can even vary from one matrix to another. The fixed-window counting strategy can thus be flawed, with points that "fall" within the "microorganism" window but are actually "particles," due to overall matrix-related diffusion that attenuates the individual diffusion value of each point.

[0087] The enumeration method according to the invention is of interest for its implementation in a more specific method used for sterility tests applied to complex matrices by solid phase cytometry.

[0088] Thus, the invention also relates to a qualitative microbiological control method characterized in that it comprises at least the following steps: implementation of the steps of the enumeration method according to the invention, said microbiological control method further comprising at least the following steps: ■ comparison with a determined contamination threshold (110), the comparison being carried out between the enumeration result obtained in the enumeration step

[0089] (109) and said contamination threshold, the comparison step being implemented after the enumeration step (109),

[0090] ■ Determination of sample contamination (111) following comparison

[0091] (110) if the result of the enumeration is greater than the threshold, then the sample is considered contaminated; if the result of the enumeration is less than the threshold, then the sample is considered uncontaminated.

[0092] According to one feature of the invention, the threshold determined for the control method is a threshold determined as a function of the desired false positive rate and negative rate, either by optimizing the false positive rate or the false negative rate, or by combining the false positive and false negative rates by calculating the precision or the Fl-score, or by calculating from the ROC (Receiving Operating Characteristics) curve.

[0093] The Fl-score is the harmonic mean of accuracy and recall. It combines the two measurements into one, taking into account both false positives and false negatives. Thus, if false positives are higher than false negatives, the impact is felt more in terms of accuracy than recall.

[0094] According to one feature of the invention, when the analysis is a solid-phase cytometry analysis, the solid-phase cytometry analysis includes a sample screening step. This screening step is equivalent to a step for detecting objects on the surface of the membrane containing the sample.

[0095] According to one feature of the invention, when the analysis is a solid-phase cytometry analysis, the control method includes a confirmation step for the objects detected during the screening step. This confirmation step makes it possible to distinguish even the smallest microbial cell, however faintly labeled, from a particle.

[0096] According to one feature of the control method, in which the analysis is a solid-phase cytometry analysis, the confirmation includes a selection step of at least one object to be confirmed, the selection step being either partial or iterative. According to one feature of the invention, the selection step is performed using the clusters and scores obtained from the preceding steps.

[0097] According to a feature of the control method in which the analysis is a solid phase cytometry analysis, the selection is a partial selection of a number of objects based on the point clouds, preferably by a selection of at least one object per cluster formed in the clustering step.

[0098] Alternatively, the selection is an iterative selection, the objects to be confirmed are selected one by one, the objects selected first being those belonging to a microorganism cluster or whose probability of belonging to a microorganism cluster is the most likely.

[0099] According to a feature of the control method in which the analysis is a solid-phase cytometry analysis, the confirmation further includes, after the iterative and / or partial selection step, a step of acquiring and examining one or more images for each selected object. Advantageously, the examination of each image is a high-magnification microscopic examination of the detected objects.

[0100] According to a characteristic of the control method in which the analysis is a solid-phase cytometry analysis, the confirmation step further includes a step of applying a classification model from one or more images acquired for each object in the acquisition step. Preferably, the model used in the classification step of the confirmation step is a supervised model.

[0101] According to a feature of the control method in which the analysis is a solid-phase cytometry analysis, the confirmation further includes a step of applying a score to each of the selected and classified objects. Advantageously, the score is a quantitative probabilistic score.

[0102] According to one feature of the invention, an optional activation function can be applied to the score obtained in the previous step.

[0103] According to one feature of the invention, the confirmation step is implemented after the scoring application step of the enumeration method and before the activation application step or the summation step of the enumeration method. According to another feature of the invention, the summation step corresponds to the summation of the scores obtained during the confirmation step.

[0104] According to a characteristic of the control method in which the analysis is a solid phase cytometry analysis, and when the selection step is an iterative selection, the enumeration score is compared with the determined threshold, if the score is above the threshold then the method is stopped if the score is below the threshold, the selection, classification model application, scoring application, and summation steps are repeated with a second object, the selection, classification model application, scoring application, and summation steps being repeated for n objects until the threshold is exceeded or all objects are confirmed.

[0105] Brief description of the figures

[0106] The invention will be better understood from the following description, which relates to an embodiment of the present invention, given by way of non-limiting example and explained with reference to the accompanying schematic figures. The accompanying schematic figures are listed below:

[0107] Fig. 1 is a 2D graphical representation of a cytogram

[0108] Fig. 2 is a diagram of the flow cytometry system according to the invention

[0109] Fig. 3 is a diagram illustrating the enumeration method according to the invention,

[0110] Fig. 4 is a diagram illustrating the qualitative microbiological control method according to the invention in a first embodiment,

[0111] Fig. 5 is a diagram detailing step 103 of the application of the algorithm in the enumeration method according to the invention,

[0112] Fig. 6 is a diagram illustrating the qualitative microbiological control method according to the invention in a second embodiment,

[0113] Fig. 7 is a diagram illustrating the qualitative microbiological control method according to the invention in a third embodiment,

[0114] Fig. 8 is a presentation of the processing of a cytogram during the method according to the invention

[0115] Fig. 9 is a comparison of the interpretation of the algorithm according to the present invention with respect to the prior art algorithm.

[0116] Fig. 10 graphically represents examples of the activation function used in the enumeration method according to the invention. Detailed description

[0117] The cytometry system for microbiological analysis, configured to implement the enumeration method according to the invention, is partially shown in Figure 2. This system 1 comprises only two fluorescence measurement detection channels, FL1 and FL2, thus maintaining a simple system with easy channel management. System 1 further includes a plurality of mirrors 2, lenses 3, a laser 4, a filter 5, a CCD (Charged Coupled Device) sensor 8, a light source 7, and an aperture 6. The laser beam 4a is directed onto a matrix 10 carrying a sample, above which a lens 3 of system 1 is positioned.

[0118] In system 1 according to the invention, the FL1 channel detects fluorescence in a specific band corresponding to the maximum intensity of the fluorophore (for example, fluorescein with an emission peak around 520 nm), and it is on this FL1 channel that object detection is performed: as soon as the intensity measured on FL1 exceeds a threshold, the peak intensity value is captured (area under the peak, peak height, and possibly peak width), and simultaneously the signal on the second channel, FL2, is also acquired. There is no channel related to object scattering measurement (Side Scatter or Forward Scatter) on system 1 according to the invention, whereas this is the case in most cytometers, which also use the FL1 channel to trigger the capture of information on the other detectors.Since system 1 according to the invention is dedicated to the analysis of a matrix 10 containing a potentially complex sample with numerous suspended objects, capturing all objects exhibiting a scattering signal would quickly saturate the detector, risking the loss of fluorescent objects. However, object size information remains valuable, and therefore system 1 according to the invention integrates this size information into the FL2 channel, which measures both fluorescence over a range beyond 540 nm (yellow / orange) and a small proportion of light scattered by the laser 4 at approximately 488 nm. Indeed, there is intentionally no bandpass filter in front of the FL2 channel, and since the dichroic mirror 2 transmits laser light, this information is combined with the fluorescence on the FL2 channel.

[0119] System 1 according to the invention is configured to implement several methods according to the invention, an enumeration method illustrated and described in Figure 3 and a microbiological control method illustrated and described in Figures 4, 6 and 7. Figure 5 being a detail of a step common to the two methods according to the invention.

[0120] The enumeration method according to the invention will now be described with reference to Figure 3.

[0121] The enumeration method includes an analysis step (101) of flow cytometry or solid-phase cytometry of a matrix comprising a given volume of a sample. In the case of enumeration, the solid-phase cytometry analysis does not require a specific confirmation step as described with reference to Figures 6 and 7.

[0122] During cytometry analysis, points are acquired on the fly, one after the other, depending on their passage through the "interrogation zone" of system 1, which corresponds to the intersection between the sample vein 10, the excitatory laser beam 4, and the optical collection axis ZZ. The data can therefore be processed "on the fly" as each event is detected and / or all the acquired points can be processed simultaneously across the entire cytogram.

[0123] Following this analysis step (101), a multidimensional data table (step 102) is obtained, showing objects detected during the cytometry analysis (flow or solid phase). Obtaining a data table (102) is graphically represented as a cytogram, as illustrated in Figure 1. Each point of the cytogram can be graphically represented in 2D, with each axis representing a measured parameter. If more than two parameters are measured for each point, several 2D graphs or even a 3D view can be used. In a 2D graphical view, as in Figure 1, "windows" F can be defined. These are regions with predefined boundaries (polygons or ellipses) that allow the selection of a portion of the cytogram points. Furthermore, and independently of predefined windows F, points can also be observed that form clusters or point clouds Cl.Since the window F is predefined by its boundaries, one can conceive of a window without any points inside; conversely, a point cloud Cl is an area of ​​the cytogram necessarily containing points, more or less densely distributed. It is perfectly possible to define, based on a displayed cytogram and according to the visible point clouds, a window around a point cloud. The cytogram may also contain isolated points Pt, not belonging to any visible group, or even points at the intersection of several point clouds Cl.From the data in the data table obtained in step 102, an enumeration algorithm is applied, including a clustering step 103 of the objects in the data table into a plurality of clusters comprising at least one cluster called a "point cloud" Cl and / or at least one isolated point Pt. This is achieved by calculating at least one parameter for each cluster Cl or isolated point Pt, the at least one parameter being selected from a list of parameters including at least one position and / or one shape and / or one density. More specifically, and as illustrated in Figure 5, the isolated points Pt are assigned a probability of belonging to the different clustered point clouds Cl: if this probability is greater than the threshold thr. outUers , then we assign this isolated point Pt to the cloud of points (decision Y) otherwise the point remains isolated and does not belong to any cloud of points (decision N), these points are also called "outliers".

[0124] Following this clustering (103), each point cloud cluster Cl is classified (104) in a binary positive / negative manner: the class is selected from several classes including at least the "microorganism" class (positive), and the classification is performed according to at least one calculated parameter and business rules relating to the marking protocol. A core area of ​​the point cloud is then considered. This core is centered on the centroid and extends diagonally (see Figure 8). This allows us to take into account both the position of the point cloud Cl and its spread along the FL1-FL2 axis (diagonally). If part of the core of this point cloud Cl lies within the positivity window, then the entire point cloud will be considered positive, and each object within it will be assigned a score of 1 in the scoring step (105).Otherwise, the point cloud Cl is considered negative, and each object within it is assigned a score of 0 in the score assignment step (105). The score assignment step (105) is a relative score assignment: an object's score depends on the scores of other objects.

[0125] Regarding outliers, also called "unclassified objects," their position relative to a predetermined positivity window is considered: i) if the unclassified object is located within the predetermined positivity window, it is assigned a score of 1, from which the set of probabilities of belonging to negative point clouds is subtracted; ii) if the unclassified object is not located within the predetermined positivity window, it is assigned a score of 0, to which the set of probabilities of belonging to positive point clouds is added. Thus, the score of unclassified objects (outliers) is between 0 and 1, depending on its distance from a cluster containing classified objects.

[0126] This approach allows us to take into account the position of unclassified objects, considering both their environment and their proximity to detected point clouds. Thus, each object, whether part of a cluster or not, is assigned a positivity score between 0 and 1.

[0127] After assigning scores (105), an activation function (107) can be optionally applied. This activation function (107) is shown in Figure 10, with the solid line representing the identity activation function f(x)=x and the dashed line representing the Heavside activation function with a threshold of 0.5, the transformed score on the y-axis and the initial score on the x-axis. The activation function, which is configurable (during the training phase according to the desired performance, it can be viewed as a hyperparameter), optimizes the correlation between the enumeration of microorganisms by flow cytometry and the CFU value (score). Indeed, below a certain score value, the sum is not meaningful: ten objects scored at 0.1 should not be equivalent to 1, whereas two objects scored at 0.5 should be.The activation function is therefore applied before any summation of scores to avoid non-representative residual scores.

[0128] The next step after assigning a score (105) and possibly applying the activation function (107) is to sum the scores (108) obtained previously. Then, the microorganisms are enumerated (109) from the sum of the scores (108). The sum (108) is, by design, equal to the enumeration (109): from the sum of the scores, the number of microorganisms, expressed in CFU, is deduced.

[0129] Finally, an alert signal giving the result of the enumeration (112) is emitted. The signal may be audible and / or visual.

[0130] Figure 8 shows the evolution of the cytogram as a function of its different treatments. Figure 8(a) illustrates the original 2D cytogram with predefined regions. Figure 8(b) illustrates the same cytogram after the clustering phase, based on the density, shape, and position of the point clouds relative to the predefined regions. Figure 8(c) illustrates the cytogram after assigning scores to each point based on the relative distance between each point. Figures 4, 6, and 7 illustrate the steps of the qualitative microbiological control method according to the invention and in several embodiments.

[0131] According to a first embodiment of the qualitative microbiological control method shown in Figure 4, the analysis step (101) in cytometry is a flow cytometry analysis or solid phase analysis without confirmation.

[0132] Steps (100) to (105) and (107) to (109) are identical to those described for the enumeration method described previously with reference to Figure 3.

[0133] Following the enumeration step (109), the control method includes a comparison step (110) with a determined contamination threshold, the comparison being carried out between the enumeration result obtained in the enumeration step (109) and said determined contamination threshold,

[0134] Finally, a step of determining the sterility or contamination of the sample following the comparison (110) is implemented: if the result of the enumeration is greater than the determined contamination threshold, then the sample is considered contaminated; if the result of the enumeration is less than the determined contamination threshold, then the sample is considered uncontaminated.

[0135] According to the second and third embodiments of the qualitative microbiological control method shown in Figures 6 and 7, respectively, the analysis step (101) in flow cytometry is a solid-phase flow cytometry analysis with a confirmation step (106). Steps (100) to (105) and (107) to (109) are identical to those described for the enumeration method described previously with reference to Figure 3. According to the second and third embodiments, the control method includes a confirmation step (106) of the objects detected and scored in step (105). This confirmation step (106) distinguishes even the smallest microbial cell, however faintly labeled, from a particle. Confirmation (106) is performed by high-magnification microscopic examination of the detected objects. Confirmation (106) can be performed selectively in two ways.

[0136] According to the second embodiment shown in Figure 6, confirmation (106) is performed by a partial selection (106a) of a number of objects based on the point clouds, preferably one object per cluster (the cluster being created in the clustering step (103) and classified in step (104) from the data table (102) resulting from the solid-phase cytometry analysis (101) which includes a screening step). From this partial selection (106a), all the selected objects are confirmed (106c) by acquiring one or more confirmation images for each selected object and by applying a classification model (106d) based on one or more images acquired for each object, the model being a supervised model. From this classification, a probabilistic quantitative score (106e) is derived for each object.An optional activation function (107) can be applied to the quantitative probabilistic score, then the scores (108) are summed, and an enumeration result (109) is derived. The enumeration score is then compared (110) with a threshold. The threshold chosen for the control method is determined based on the desired false positive and false negative rates, either by optimizing the false positive or false negative rate, or by combining the false positive and false negative rates by calculating the precision or the Fl-score, or by calculating from the ROC (Receiving Operating Characteristics) curve. Whether the sample is contaminated or not is then determined (step 111).

[0137] According to the third embodiment shown in Figure 7, confirmation (106) is performed by an iterative selection (106b), starting with objects whose probability of belonging to the microorganism cluster is high. More specifically, a first object belonging to a microorganism cluster is selected (the cluster being created in the clustering step (103) and classified in step (104) from the data table (102) resulting from the solid-phase cytometry analysis (101), which includes a screening step). From this selection (106b), the first selected object is confirmed (106c) by acquiring one or more confirmation images for each selected object and by applying a classification model (106d) based on one or more images acquired for each object. The model is a supervised model. From this classification, a quantitative probabilistic score (106e) is deduced for the first object.An optional activation function (107) can be applied to the probabilistic quantitative score, then the score (108) is summed and an enumeration result (109) is deduced. The enumeration score is then compared (110) with a threshold; if the score is greater than the threshold, the method is stopped; if the score is less than the threshold, steps 106c, 106d, 106e, 107, 108, 109 are repeated with a second object, with steps 106c, 106d, 106e, 107, 108, 109 being repeated for n objects until the threshold is exceeded or all objects are confirmed.Figure 9 represents a comparison between the processing of a sample according to a known algorithm figures (a) and (c) and the algorithm according to the invention figures (b) and (d) We see that in the case of a known algorithm different point clouds Cil, CI2 are detected depending on whether they are positioned in the determined positivity window F, while for the algorithm according to the invention only one point cloud Cil is detected.

[0138] With such an approach, the known algorithm would have wrongly detected contamination, while the algorithm according to the invention, which is much more precise, avoids the false positive.

[0139] Of course, the invention is not limited to the embodiments described and shown in the accompanying figures. Modifications remain possible, particularly with regard to the composition of the various elements or by substitution of technical equivalents, without departing from the scope of protection of the invention.

Claims

DEMANDS 1. Enumeration method for microbiological control comprising: Analysis (101) by flow cytometry of a matrix comprising a given volume of a sample, said sample having been previously labeled by a labeling protocol, Obtaining (102) a multidimensional data table presenting objects detected during flow cytometry analysis, characterized in that the method includes a step of applying an enumeration algorithm comprising at least the following steps: Assignment (105) of a quantitative score within a range of values ​​from [0 to 1] to each object detected during the analysis based on its probability of belonging to a determined "microorganism" class Sum (108) of the scores obtained in the previous step and in that the method includes a step: of Enumeration (109) of microorganisms from the sum of the scores.

2. Enumeration method according to claim 1, comprising the following steps implemented before the scoring assignment step (105): ■ Clustering (103) of the objects in the data table into a plurality of clusters comprising at least one cluster called a "point cloud" and / or at least one "isolated point" by means of a calculation of at least one parameter for each cluster or isolated point, the at least one parameter being selected from a list of parameters comprising at least one position and / or shape and / or density of the cluster or isolated point, ■ Classification (104) of each point cloud cluster, the class being selected from several classes including at least the class "microorganism", the classification being carried out according to at least one parameter calculated in the previous step and business rules relating to the marking protocol; each isolated point not being classified.

3. Enumeration method according to claim 2, comprising one or more pre-processing steps (100) of the data table before the clustering step.

4. Enumeration method according to claim 3, comprising at least one pre-processing step of the type sizing and / or rotation and / or distortion and / or binning of the data table.

5. Enumeration method according to any one of claims 3 or 4, comprising at least one dimensionality reduction pre-processing step.

6. Enumeration method according to any one of claims 2 to 5, wherein the clustering of the data table is carried out from learning methods based on a density calculation of the objects in the data table.

7. Enumeration method according to any one of claims 2 to 5, wherein the clustering of the objects in the data table can be achieved from a learning method such as density analysis and preferably HDBSCAN analysis.

8. Enumeration method according to any one of claims 1 to 7, wherein the cytometry analysis is a solid phase analysis.

9. Enumeration method according to any one of claims 1 to 8, comprising a step of applying an activation function (107) to the scores of each object, said activation function being configured to optimize the correlation between the enumeration of detected microorganisms and the sum of the scores.

10. Cytometry system (1) for microbiological analysis purposes, configured to implement the enumeration method for the microbiological control of samples according to any one of claims 1 to 9.

11. Qualitative microbiological control method comprising the following steps: implementation of the steps of the enumeration method according to any one of claims 1 to 9, said microbiological control method further comprising at least the following steps: ■ comparison with a determined contamination threshold (110), the comparison being made between the enumeration result obtained in the enumeration step (109) and said contamination threshold, the comparison step being implemented after the enumeration step (109), ■ Determination of sample contamination (111) following comparison (110) if the result of the enumeration is greater than the threshold, then the sample is considered contaminated; if the result of the enumeration is less than the threshold, then the sample is considered uncontaminated.

12. Qualitative microbiological control method according to claim 11, wherein where the analysis is a solid phase cytometry analysis which includes a screening step of the sample, the control method includes a confirmation step (106) of the objects detected during the screening step.

13. Qualitative microbiological control method according to claim 12, wherein the confirmation (106) comprises a selection step (106a, 106b) of at least one item to be confirmed, the selection step being either partial (106a) or iterative (106b).

Images