Method for detecting the presence, identification and quantification of anticoagulants in a blood sample that are inhibitors of blood clotting enzymes, and means for carrying this out

Description

TECHNICAL FIELD OF THE INVENTION

[0001] This application relates to the field of hemostasis and in particular that of blood coagulation and provides methods and means allowing "blind" detection, and subsequently and where appropriate, identification and qualitative then quantitative characterization (dosage), of anticoagulants, inhibitors of blood coagulation enzymes in an analyzed sample.

[0002] The invention implements artificial intelligence tools, in particular supervised machine learning models, for the purpose of post-processing kinetic measurements carried out on blood samples.

[0003] The invention thus relates more particularly to a method for detecting the presence of an inhibitor of a blood coagulation enzyme, the latter being chosen from factor Xa and factor IIa, as defined in claim 1, then, in the event of presence, identification of said inhibitor, this identification including its category as for example defined in claim 2, its nature or its effective characterization as defined in claims 3, 5 or 6, then after this identification, its quantitative dosage as defined in claims 4, 7 or 8. The invention is firstly intended for application for the detection in vitro of the presence of the inhibitors in question here in biological samples taken from human subjects. Claims 9 to 12 specify characteristics of the steps carried out in vitro.

[0004] The invention also relates to useful means, and adapted for these purposes: data processing system or appropriate device, computer program, recording or data medium, and kits allowing the implementation of all the steps of the methods described here, whether experimental or in silico. Said means are as defined in claims 13 to 17. STATE OF PRIOR ART

[0005] Coagulation is a complex physiological phenomenon, occurring in a cascade and involving several plasma proteins. Once this process is initiated, it results in the formation of a clot (a platelet plug), which in its non-pathological physiological function helps to slow or stop hemorrhages. Conversely, coagulation disorders that lead to a greater risk of bleeding are called hemophilia. Natural blood clotting is regulated by the presence of various clotting factors that work together to maintain a normal balance between the natural tendencies to clot and those to bleed.

[0006] It is known, for example from EP1367135A1, methods in vitroto determine the dynamics of thrombin formation, i.e., factor IIa, in a blood sample over time. This type of method makes it possible to obtain the kinetics of thrombin formation over time, and to see the impact of different parameters on these kinetics.

[0007] It is also known from WO2014 / 134223A1 to use a test for the dosage of factor X, which is involved in the common coagulation pathway and is activated to factor Xa by different agents, as a biomarker for diagnosing kidney disease in a patient. WO2014 / 134223A1 discloses that the diagnosis can be made using thresholds or classification models which can be rule sets, decision trees, Bayesian methods or neural networks. WO2014 / 134223A1 however does not seek to detect the presence of an inhibitor of a blood clotting enzyme selected from factor Xa and factor IIa in a blood sample of a patient.

[0008] In some cases, especially in pathological situations, the coagulation process must be corrected or modified by administering an anticoagulant agent, in the treatment or prevention of certain disorders or pathologies. The degree of anticoagulant effect achieved when using these agents in the form of medications depends on a large number of factors, often little known. In the event of an overdose, however, unexpected bleeding from natural orifices may occur. Such treatments therefore involve regular monitoring of the degree of anticoagulation, generally by means of a blood test.

[0009] Non-physiological anticoagulants, for example synthetic, i.e., drug anticoagulants, are thus commonly prescribed for the prevention and / or treatment of established disorders.

[0010] The available medicinal anticoagulant agents that inhibit coagulation enzymes belong to two main families of anticoagulants: 1 / heparins, and 2 / direct oral anticoagulants (DOACs), the agents of both families corresponding to inhibitors of factor Xa and / or factor IIa of coagulation.

[0011] Heparins are irreversible indirect inhibitors of factors Xa and IIa. They are divided into two classes: unfractionated heparins (UFH) and low molecular weight heparins (LMWH). UFH have a ratio (anti-Xa potency / anti-Ila potency) close to 1, while LMWH have a ratio (anti-Xa potency / anti-Ila potency) close to 2 or 3.

[0012] Direct oral anticoagulants (DOACs) are direct reversible inhibitors specific to factor Xa or factor IIa. The main factor Xa-specific DOACs are rivaroxaban, apixaban, and edoxaban. The main factor IIa-specific DOAC is dabigatran. Tables 18 and 20 provide more complete details of other known synthetic factor Xa and IIa inhibitors.

[0013] In order to determine the hemostatic status of subjects in clinical situations requiring it and to monitor the hemostatic status of patients under anticoagulant treatment, different tests aim to detect and / or measure coagulation parameters.

[0014] For example, there are clinical tests that can determine the presence of hypocoagulation, such as an increase in prothrombin time or prothrombin time. However, these medical biology tests require laboratory equipment that is not always available in all locations.

[0015] There are also known tests that can detect the presence of a specific, most often unique, anticoagulant whose presence is suspected. For example, the STA ®< - Liquid Anti-Xa kit marketed by the Applicant is intended for use with the STA-R ®< range of devices, for quantitative determination of plasma levels of unfractionated heparins (UFH) or low molecular weight heparins (LMWH) by measuring their anti-Xa activity in a competitive test using a synthetic chromogenic substrate. Depending on experience, equipment availability, and operator responsiveness, this test, which can be performed in about ten minutes, is of course relevant, especially when you know what you are looking for. This type of test, however, always requires a calibration step.

[0016] The Stago STA ®< - Multi-Hep Calibrator kit is also known, which allows the measurement of both UFH and LMWH using a common methodology and hybrid calibration. However, this kit does not allow the detection of DOAs.

[0017] There are still situations, particularly clinical situations, in which the medication regimen of the patient in question is not known with certainty or cannot be determined (for example, if the patient is unconscious), which complicates a detection task. However, it may be necessary or preferable to know the anticoagulant regimen of a patient by identifying the anticoagulant or anticoagulants administered to him and their concentration in the patient, possibly immediately after medical care of the patient, if necessary in emergency conditions, for example in view of drug treatment, biological or clinical examinations, or a surgical intervention.Generally speaking, detecting the presence of one of the known drug anticoagulants in a blood sample or their identification can be very useful in an emergency context (in the context of a hemorrhagic phenotype, the implementation of an antidote, etc.). The tests available for measuring anticoagulant agents are suitable for measuring anticoagulants whose identity is known and could be communicated in good time during the patient examination and for which it has been verified that they were prescribed or administered to the patient examined.On the contrary, "blind" dosage tests which would allow in a first step to identify the presence of anticoagulant(s) in a biological sample, then to determine their category and / or their identity and finally, if applicable, their concentration, in an effective test for collecting biological or clinical information are not available even though they would be of obvious use in responding to certain clinical patient management situations.

[0018] Finally, although there are currently several commercial solutions for measuring the various anticoagulants mentioned above, all of these solutions necessarily require the use of a dedicated kit with its own calibrators and controls as well as its own experimental methodology.

[0019] In fact, in the example of measuring the presence of an anticoagulant inhibitor of factor Xa, the use of current methods requires a minimum of four dosages (for example, one dosage of heparins in hybrid methodology and three dosages of anti-Xa DOACs). Identification without prior information of the anticoagulant molecule is currently impossible.

[0020] The invention thus proposes in particular to solve all or part of the problems identified in the preceding paragraphs, in particular the problem of "blind" detection of the presence or absence of an anticoagulant inhibiting a blood clotting enzyme in a blood sample. The proposed solution allows in particular, for the first time, the development of a test particularly suited to a clinical management situation, in particular in an emergency context, in the context of uncertainty about the patient's treatment regimen or in the context of an unconscious patient. If such a test is to provide comfort and save time for the user in the context of clinical management, it must naturally also be reliable, given the potentially life-threatening nature of the issue. Such a test should allow the reliable identification of an agent contained in a blood sample, the presence of which is unknown, or the presence of which cannot be suspected.It should be noted that no such test currently exists.

[0021] According to a particular aspect, the invention relates to the detection of exogenous anticoagulant agents, also called "synthetic" in the present text, in particular the detection of anticoagulant drugs and preferably also allows detection in a single test, making it possible to reach a conclusion without repeating a multitude of separate tests. DETAILED DESCRIPTION

[0022] Thus, in response to these problems, a method for detecting in a biological sample, in particular a blood sample, the presence of an inhibitor of a blood clotting enzyme chosen, independently, from factor Xa (FXa) and factor IIa (FIIa) has been designed and implemented, the method as defined in claim 1 comprising the following steps: a. Measuring one or more competitive kinetics by performing a competitive enzymatic assay on a blood sample previously obtained from a subject, said assay being suitable for performing competitive kinetics with respect to either a factor Xa inhibitor or a factor IIa inhibitor (independently of one another), then b. Providing as input the kinetics obtained in step a. to a classification decision model A obtained by training a supervised machine learning model, for example a model chosen from one of the following families: support vector machine, neural networks, decision trees, ensemble methods, and k-nearest neighbors model, in particular the parameters of which have been calculated by training, then biIf decision model A excludes the presence of an inhibitor of the targeted blood clotting enzyme in the sample analyzed, conclusion of the absence of said inhibitor, optionally allocation as output by model A, for example in a variable, of the corresponding information, or b.ii. If decision model A attests to the presence of an inhibitor of the targeted blood clotting enzyme in the sample analyzed, conclusion of the presence of said inhibitor, optionally allocation as output by model A, for example in a variable, of the corresponding information.

[0023] Claims 2 to 12 detail more specific features of this method. The blood sample tested is, for example, a blood sample or a plasma sample. It may be diluted or not, as required.

[0024] The invention serves the problem of so-called "blind detection". By "blind detection", it is indicated that according to the invention, it is not known in advance whether an inhibitor of the targeted enzyme is present in the sample, nor a fortiori which inhibitor could be present, and that the method described here is thus suitable and, depending on the models made available within its framework, can be used for the detection of any coagulation inhibitor acting on factors Xa, and / or IIa.

[0025] The invention relates more particularly to the detection of an inhibitor here called medicinal (also called "synthetic" in Tables 17 to 20) of factor Xa (FXa) or factor IIa (FIIa), as opposed to a so-called "natural" or physiological inhibitor (the distinction being illustrated in the form of lists in Tables 17 to 20), the inhibitor having a selective anticoagulant function - due to the targeted enzyme. In a particular embodiment of the invention, a sought inhibitor of a coagulation enzyme is thus a non-physiological inhibitor. Such an inhibitor is in particular a medicinal inhibitor. According to a particular aspect, a medicinal inhibitor may be found in a patient sample at a concentration substantially higher than a "natural" physiological concentration. According to particular embodiments, said inhibitors belong to the heparin family or that of DOACs.

[0026] Examples of drug (synthetic) factor Xa inhibitors designated by their active substance or INN are: Rivaroxaban, Apixaban, Edoxaban, Betrixaban, UFH, for example calcium UFH, sodium UFH, LMWH, Pentasaccharide, Danaparoid sodium.The proof of concept reported in the present application is aimed more particularly at: heparins which are unfractionated heparins (UFH) such as calcium UFH or sodium UFH, low molecular weight heparins (LMWH) such as those constituting the active ingredients of Fragmine ®< , Lovenox ®< , Innohep ®< , or DOACs which are Rivaroxaban, Apixaban, Edoxaban, but it is understood that the method described here can be implemented for any type of factor Xa and / or IIa inhibitors, provided that an appropriate substrate of the enzyme is available, and that adequate models have been generated or can be generated in accordance with the guiding principles set out here and the illustration provided.

[0027] Examples of (synthetic) drug inhibitors of factor IIa designated by their active substance or INN are: Dabigatran, Melagatran, Argatroban, Bivalirudin, HNF, LMWH. The skilled person will be able, on the basis of the literature concerning the artificial intelligence algorithms and their implementation at his disposal, and according to the content of the experimental part of the present description, to adapt the protocols described here to substitute, integrate or add other models allowing decision-making.

[0028] By "selected blood clotting enzyme" are thus meant factor Xa and factor IIa, the inhibitors of which are sought independently of each other, respectively, in the same test. The method described here proposes to detect the presence (and in subsequent steps, if necessary, to detect the identity and preferably to carry out a quantitative assay, for example to give the concentration), either a factor Xa inhibitor or a factor IIa inhibitor, to the exclusion of a concomitant detection of both.

[0029] By "measurement of several competition kinetics" in step a., it is indicated that, according to a particular embodiment, the measurements can be repeated several times on the basis of the same blood sample taken from a subject (preferably, a human, in particular, a patient), in order to offer better reliability. Several kinetics can thus be provided at all the steps which require it, or a single kinetic integrating all the others can be provided.To carry out an enzymatic assay so as to obtain competition kinetics within the framework of the invention, the conditions specific to carrying out the following biochemical reactions must be met: the blood sample likely to contain the inhibitor of the targeted coagulation enzyme is placed in the presence of a chosen substrate, of the targeted enzyme, preferably a substrate specific to the targeted enzyme, and an incubation step is carried out before adding the targeted enzyme to the reaction mixture under conditions allowing the triggering of competition between the inhibition reaction of the enzyme and the enzymatic reaction of said enzyme with its substrate.The substrate used is normally labeled so as to allow the release of the label when the enzyme interacts with this substrate to transform it into a product of the enzymatic reaction and to subsequently allow the measurement of said released label and the recording of its variation over time in the form of a kinetics of release of the label which is impacted when the inhibitor is present in the sample, this impact being further modulated by the concentration of the inhibitor. Kinetic diagrams of the competitive assay reactions are illustrated in the Examples.Substrates indicated for the enzymatic reaction with the targeted enzyme are known to those skilled in the art; examples suitable for carrying out the invention include: the MAPA-Gly-Arg-pNA substrate specific for factor Xa which, once transformed by FXa, induces the release of paranitroalanine (pNA) detectable by colorimetry when measured at 405nm, or the MAPA-Gly-Arg-AMC substrate or any known variant, which induces the release of AminoMethylCoumarin (AMC) detectable by fluorometry, or the EtM-SPro-Arg-pNA substrate specific for factor IIa, or any variant known to those skilled in the art.

[0030] The measurement of a competition kinetics according to step a., is a measurement which is carried out in vitro,according to standard practice in the field of hemostasis exploration, and according to well-known techniques, examples of which are given below. This experimental step is carried out according to conventional practices well known to those skilled in the art.

[0031] It will nevertheless be understood that there is an inseparable link for the reliability of the results throughout the classification chains proposed in the present invention, between the data sets which were used for training the models and the kinetic data obtained from the blood sample of the subject considered since said data sets and said kinetic data of the test carried out on the sample of the subject must have been obtained under similar, if not identical, experimental conditions. Thus, it will be said that the decision models used are linked / associated with the reagents and instruments used in the context of the collection of test data from the subject.

[0032] That being said, this does not mean that the invention must be carried out on a particular model of instrument or measuring device, or with a particular model of reagents. It is simply necessary to compare the results of tests or experiments conducted under similar experimental conditions so that the result is not distorted, the algorithmic methods described here replacing an inter-instrument calibration and making it possible to dispense with the need for calibration for each test or batch of tests carried out.

[0033] By "decision-making classification model A obtained by training a supervised machine learning model", for example a model chosen from one of the following families: "support vector machine, neural networks, decision trees, ensemble methods, and k-nearest neighbors model", reference is made to the multiple possibilities for implementing a supervised machine learning model, as documented in particular in the literature. The person skilled in the art, on the basis of his general knowledge, will be able to modulate the type of model used to adapt the observed performances to his needs and the objectives sought. See for example, non-exhaustively [Géron, 2017] or Bonaccorso, G. (2017).

[0034] A supervised machine learning model has, by definition, its parameters which have been calculated by training, in a conventional way in the field. See for example, non-exhaustively [Géron, 2017] or Bonaccorso, G. (2017).

[0035] According to a particular embodiment, the classification decision model A is a support vector machine, in particular was obtained by training a support vector machine. A more particular embodiment applied to the search for a factor Xa inhibitor is as described in the experimental part.

[0036] It follows from the essence of the invention that the method described here takes into account the result of training a machine learning model on previously collected and established data sets. Typically, such a classification decision model may consist of a matrix corresponding to an empirically determined mathematical function, which, applied to an input data set (the composition of which is, prior to carrying out the method according to the invention, unknown), makes it possible to return a solution. With regard to model A, the solution may, for example, be either "yes" or "no" reflecting the presence or absence of a sought inhibitor (which may be coded by an integer, such as "1" or "0").

[0037] In fact, step b. reported above is a step necessarily implemented by computer, due to the complexity of the mathematical function empirically determined by learning.

[0038] According to another particular embodiment, the classification decision model A is a neural network, in particular a multi-layer perceptron, in particular was obtained by training a neural network, in particular a multi-layer perceptron. A more particular embodiment of such a model A, applied to the search for a factor Xa inhibitor is as described in the experimental part.

[0039] Examples of implementation, in particular implementation models of a support vector machine or a neural network, such as a multi-layer perceptron, are known to those skilled in the art, if necessary by reference to the literature in the field (See for example, non-exhaustively [Géron, 2017] or Bonaccorso, G. (2017)). More precisely, the LIBSVM package (https: / / www.csie.ntu.edu.tw / -cjlin / libsvm / ) implemented in the scikit-learn library (https: / / scikit-learn.org / ) can be used. The parameters that the skilled person will think of setting are, for example, and independently or not of each other, the kernel (linear, polynomial, Gaussian, sigmoidal ...), the kernel coefficients and the bare parameter associated with the SVM model. The person skilled in the art may refer in particular to Géron, 2017 or Bonaccorso, G.(2017), for the implementation of such models, including with regard to the methods for setting the aforementioned parameters, including, where necessary, the directives contained explicitly or implicitly in the present description. In the event of the proven presence of an inhibitor, a more advanced embodiment of the present invention subsequently includes a step of identifying the inhibitor of said detected enzyme.

[0040] The invention thus also relates to a method for identifying in a biological sample, in particular a blood sample, an inhibitor of a blood coagulation enzyme chosen, independently, from factor Xa (FXa) and factor IIa (Flla), the method comprising the following steps: 1. Implementation of the steps of the detection method described above or of any particular mode of this method described and in the present description, then 2. Supply as input the kinetics(s) obtained in the step defined in point a. above (in the context of the detection) and the result obtained at the end of step b. ii.above (in the context of detection), to a classification decision model B obtained by training a supervised machine learning model, for example a model chosen from one of the following families: support vector machine, neural networks, decision trees, ensemble methods, and k-nearest neighbor model, in particular whose parameters have been calculated by training, and allocation at output by model B of the inhibitor category to one of the following categories: irreversible indirect inhibitor (heparins), or reversible direct inhibitor (AOD), in particular when the data sets used for training model B include data relating to these two categories of inhibitors, and provision at output, for example by allocation in a variable, of the inhibitor category determined by model B.

[0041] By "classification decision model B" is meant the same thing as the above in relation to "model A", by analogy.

[0042] According to a particular embodiment, the classification decision model B is a k-nearest neighbor model, in particular was obtained by training a k-nearest neighbor model. A more particular embodiment applied to the search for a factor Xa inhibitor is as described in the experimental part.

[0043] By "particularly when the data sets used for training model B include data relating to these two categories of inhibitors, i.e., irreversible indirect inhibitor (heparins), and reversible direct inhibitor (DOA)" it is understood that it follows from the essence of the invention that the data sets used for training must be correlated with the type of information that one seeks to obtain blindly. Guidelines for the skilled person naturally follow from the examples provided in the experimental part.

[0044] In fact, the implementation step of model B is a step necessarily implemented by computer, due to the complexity of the mathematical function empirically determined by learning.

[0045] According to another particular embodiment, the classification decision model B is a neural network, in particular a multi-layer perceptron, in particular was obtained by training a neural network, in particular a multi-layer perceptron. A more particular embodiment of such a model B, applied to the search for a factor Xa inhibitor, is as described in the experimental part.

[0046] Examples of implementation, in particular models for implementing a k-nearest neighbor model or a neural network, such as a multi-layer perceptron, are known to those skilled in the art, if necessary by reference to the literature in the field (See for example, non-exhaustively [Géron, 2017] or Bonaccorso, G. (2017)). More specifically, the scikit-learn library (https: / / scikit-learn.org / ) can be used, for example. The parameters that the skilled person will consider setting are, for example, and independently or not of each other, the number of neighbors as well as the distance metric. The skilled person may in particular refer to Géron, 2017 or Bonaccorso, G.

[0047] (2017), for the implementation of such models, including with regard to the methods of setting the aforementioned parameters, including where necessary the directives contained explicitly or implicitly in this description.

[0048] The implementation of model B returns a response relating to an "inhibitor category" chosen from two: reversible direct inhibitor (alternatively called "DOA" in this text), irreversible indirect inhibitor (alternatively called "heparins" in this text) (see explanations of these concepts in Figure 1 ). In fact, since heparins are all irreversible indirect inhibitors, and DOACs (direct oral anticoagulants) are all reversible direct inhibitors, when, according to a particular embodiment corresponding to the proof of concept described in the experimental part, the data sets provided for learning involve active ingredients belonging to these two families only, the method according to the invention will be able to distinguish between these two particular families, then indicated as "heparins", and "DOACs".

[0049] In another aspect, the identification method described in the preceding paragraphs (as opposed to the "detection" method described above), partly carried out in vitro,may further comprise an additional step of characterizing the inhibitor whose presence was detected in step b. ii above (as part of the detection), as follows: providing as input the kinetics(s) obtained in step a. above (as part of the detection), equivalent to step 1. above, and the data determined, for example the allocated variable, in step 2.above implementing model B concerning the category of inhibitor whose presence has been detected, to a classification decision model C obtained by training a supervised machine learning model, for example a model chosen from one of the following families: support vector machine, neural networks, decision trees, ensemble methods, and k nearest neighbors model, in particular whose parameters have been calculated by training, and output characterization by model C of the inhibitor sought, the latter being identified from: . a. In the case where the category of the inhibitor sought is that of heparins: UFH or LMWH (for factor Xa and / or IIa inhibitors), or b. In the case where the category of the inhibitor sought is that of DOACs: Rivaroxaban, Apixaban, Edoxaban (for a factor Xa inhibitor) or Dabigatran (for a factor IIa inhibitor), and provision as output, for example by allocation in a variable, of the characterization of the inhibitor determined by model C.

[0050] By "classification decision model C" is meant the same as the above in relation to "model A" or "model B", by analogy.

[0051] According to a particular embodiment, the classification decision model C is a neural network, in particular a multi-layer perceptron, in particular was obtained by training a neural network, in particular a multi-layer perceptron. A more particular embodiment applied to the search for a factor Xa inhibitor is as described in the experimental part. In fact, the step of implementing the model C is a step necessarily implemented by computer, due to the complexity of the mathematical function empirically determined by learning.

[0052] Examples of implementation, in particular models for implementing a neural network, in particular a multi-layer perceptron, are known to those skilled in the art, if necessary by reference to the literature in the field (See for example, in a non-exhaustive manner [Géron, 2017] or Bonaccorso, G. (2017)). More precisely, the scikit-learn library (https: / / scikit-learn.org / ) can be used, for example. The parameters that the skilled person will consider setting are, for example, and independently or not of each other, the number of layers of neurons, the number of neurons per layer and the activation functions of the different neurons. But the user has a certain freedom of implementation with regard to the learning algorithm or the regularization parameters. The skilled person may in particular refer to Géron, 2017 or Bonaccorso, G.(2017), for the implementation of such models, including with regard to the methods of setting the aforementioned parameters, including where necessary the directives contained explicitly or implicitly in this description.

[0053] According to another aspect, the identification method described in the preceding paragraphs, partly carried out in vitro,may also comprise, following the implementation of model C, an additional step of quantitative dosage of the characterized inhibitor, in which are provided as input to a regression model D, in particular a supervised machine learning model, for example a model chosen from one of the following families: support vector machines, neural networks, decision trees, ensemble methods, and k-nearest neighbor model, the kinetics(s) obtained in step a. above (in the context of detection), equivalent to step 1. above, and the characterization data obtained following the implementation of Model C identifying the inhibitor present in the analyzed blood sample, said regression model having been trained on a data set obtained under measurement conditions identical to those of step a. above (in the context of detection), equivalent to step 1.above, and allowing the concentration of the inhibitor identified in the analyzed sample to be determined at the output and optionally providing at the output, for example by allocation in a variable, the concentration determined by model D.

[0054] "Allowing the concentration of the identified inhibitor to be determined at the output" can also be written as "determining the concentration of the identified inhibitor at the output" or "returning the concentration of the identified inhibitor at the output".

[0055] By "regression model D", or "regression model type machine learning model", the same is meant as the above in relation to "model A", "model B" or "model C", by analogy.

[0056] According to a particular embodiment, the model D is a neural network, in particular a multi-layer perceptron, in particular was obtained by training a neural network, in particular a multi-layer perceptron. A more particular embodiment applied to the search for a factor Xa inhibitor is as described in the experimental part.

[0057] In fact, the implementation step of the D model is a step necessarily implemented by computer, due to the complexity of the mathematical function empirically determined by learning.

[0058] Examples of implementation, in particular models for implementing a neural network, in particular a multi-layer perceptron, are known to those skilled in the art, if necessary by reference to the literature in the field (See for example, non-exhaustively, [Géron, 2017]). More specifically, the scikit-learn library (https: / / scikit-learn.org / ) can be used, for example. The parameters that the skilled person will consider setting are, for example, and independently or not of each other, the number of layers of neurons, the number of neurons per layer and the activation functions of the different neurons. But the user has a certain freedom of implementation with regard to the learning algorithm or the regularization parameters. The skilled person may in particular refer to Géron, 2017 or Bonaccorso, G.(2017), for the implementation of such models, including with regard to the methods for setting the aforementioned parameters, including where necessary the directives contained explicitly or implicitly in this description. It goes without saying that the references “A”, “B”, “C” and “D” for the models discussed in this description have been indicated for ease of reading, but that these letters may be omitted without changing the meaning of what they indicate. In particular: . A model referred to as "A" is used to exclude or attest to the presence of an inhibitor of the targeted blood clotting enzyme in the analyzed sample, according to any of the definitions contemplated in this description for this inhibitor; A model referred to as "B", if implemented, is used to determine an "inhibitor category" chosen from two: reversible direct inhibitor (alternatively called "DOA" in this text), irreversible indirect inhibitor (alternatively called "heparins" in this text);A model referred to as "C", if implemented, is used to specify the characterization started with model "B": among heparins, the inhibitor sought can be classified between UFH or LMWH (whether the inhibitor sought is a factor Xa and / or IIa inhibitor), and among DOACs, the inhibitor sought can be classified between: Rivaroxaban, Apixaban, Edoxaban (for a factor Xa inhibitor) or Dabigatran (for a factor IIa inhibitor); A model referred to as "D", if implemented, is used to determine the output concentration of the identified inhibitor. This is also valid for a model called "D2" in this description, detailed below: this model is also used to determine the output concentration of an inhibitor. ;

[0059] At this stage, it will be noted that the method(s) described herein, whether the chain of steps detailed above is implemented in whole or in part, and as demonstrated by the proof of concept which is the subject of the examples, provides a solution to the problems mentioned above through the realization of a competitive enzymatic test combined with a universal algorithmic processing methodology. The method(s) of the invention described herein, are made sensitive to the presence, where appropriate at adjustable concentrations, of anti-Xa or anti-Ila anticoagulant agents, the competitive enzymatic test being coupled with an Artificial Intelligence implemented, according to various particular embodiments, by a cascade of machine learning models (carried out in whole or in part). The advantages which result therefrom are: the provision of a universal methodology: a single test can be sensitive to the presence, and at various concentrations, of several types of anticoagulants in a sample (anti-Xa and / or anti-Ila anticoagulants), in particular, the anticoagulants more particularly described according to any embodiment which is the subject of the present description; for the detection of this presence: a single test is to be carried out versus a plurality (five for a dosage of UFH, LMWH, Rivaroxaban, Apixaban, Edoxaban in a conventional manner, or four in hybrid methodology as discussed above - a single kit for heparins); the obsolescence of the calibration: ease of use for the user; the identification of the extrinsic anticoagulant molecule sought among a certain number: this is impossible today, the clinical gain for the patient is obvious.

[0060] According to a particular embodiment of the invention, the method(s) described in the present text are applied to the search for a factor Xa (FXa) inhibitor chosen from: HNF, LMWH, Rivaroxaban, Apixaban, Edoxaban.

[0061] According to another particular embodiment of the invention, the method(s) described in the present text are applied to the search for a factor IIa (FIIa) inhibitor chosen from: HNF, LMWH, Dabigatran.

[0062] According to a more precise embodiment, the method of the invention is applied precisely and specifically to the search for a factor Xa inhibitor.

[0063] It will be readily understood, by referring in particular to the experimental part stating the interest, in the particular case which is the subject of the examples, of a methodology called "methodology optimized for AODs", that in certain cases it may be appropriate to adapt the decision cycles to allow the introduction of a new experimental measurement step, allowing for example, for the continuation of the chain of decisions that the present invention allows, a more reliable validation, or the adaptation to more precise, or different objectives (in particular with regard to the concentration range of inhibitor(s) that one wishes to be able to detect).

[0064] In this case, it was found that the use of said STA ®< - Liquid Anti-Xa kit described above for the search for AOD, under dilution conditions and over a measurement time period adapted for the search for heparins (the kit being initially designed for this), did not make it possible to obtain measurements over the entire range of inhibitor concentration values envisaged: thus, an adjustment was proposed with regard to the dilution and the measurement time, in order to optimize, modulate, the method according to the invention described here to fulfill the particular objective of being able to target certain concentrations. This adjustment was implemented in the form of an “experimental restart” (performing a new measurement of one or more competition kinetics via the performance of a competitive enzymatic assay), integrated into a method already started.The interest of an experimental restart may depend on the inhibitory activity of the anti-Xa anticoagulant (Heparins, DOACs) whose research is targeted, as well as on the concentration window that one wishes to measure, in particular with precision. The invention, however, covers situations where from the outset, the conditions employed for example either in the "universal methodology" or in the "methodology optimized for DOACs" of the examples, are used.

[0065] The guidelines given herein allow the user to determine from any kit used for kinetic measurements, in particular as known in the art, the dilution conditions, if necessary, and measurement time, to be applied consistently both for measurements affecting the training data, and those of the data actually measured blindly.

[0066] Generally speaking, these adjustments as described in the examples show that the person skilled in the art is able to adapt the way of linking the “models” described here or the classification methods, decision methods, and the compared data, to integrate an experimental restart, following the directives implicitly and explicitly detailed in the present application.

[0067] In fact, as implied by the experimental part providing proof of concept, the method of the invention also works when experimental data initially obtained in step a. of "detection" described above are used instead of those from an experimental restart: the range of values of inhibitor concentrations that can be detected is then simply different. This is an optimization that the skilled person can carry out, on the basis of his general knowledge, possibly the guidelines provided here. Furthermore, the adjustments described here, where appropriate with the test methods indicated in the experimental part, can be carried out by the skilled person to adapt the invention described to different known detection or assay kits, without the essence of the present invention being changed.

[0068] According to a particular embodiment, the present invention thus also relates to an identification method, partly carried out in vitro, of a factor Xa inhibitor the presence of which was detected in step b. ii above (as part of the detection), comprising the following additional characterization step: I. if the inhibitor category has been allocated, in step 2. where model B has been implemented, to the heparin category, then: Ii providing as input to a classification decision model C obtained by training a supervised machine learning model, for example a model chosen from one of the following families: support vector machine, neural networks (in particular a multi-layer perceptron), decision trees, ensemble methods, and k-nearest neighbor model, in particular whose parameters have been calculated by training, said determined data, for example the variable allocated, in step 2. above implementing model B concerning the inhibitor category whose presence has been detected, and the competitive kinetics(s) with respect to a factor Xa inhibitor, obtained in step a. above (in the context of the detection) equivalent to step 1. above, and I.ii.output characterization by model C of the desired inhibitor, the latter being identified among: HNF or LMWH, and output provision, for example by allocation in a variable, of the characterization of the inhibitor determined by model C, or, alternatively, II. if the inhibitor category has been allocated, in step 2. where model B has been implemented, to the DOA category, then: II.i.carrying out a new measurement (experimental relaunch) of one or more competition kinetics via carrying out a competitive enzymatic assay on a blood sample obtained from a (same) subject, said assay being suitable for carrying out competitive kinetics with respect to a factor Xa inhibitor, with a sample dilution factor and / or a measurement duration adapted to a competition situation involving the presence of DOAs inhibiting factor Xa, in particular a dilution factor and / or a measurement duration different from that(those) used for the measurement of the kinetics obtained in step a. above (in the context of detection) equivalent to step 1. above, then II.ii.supplying as input to a classification decision model C obtained by training a supervised machine learning model, for example a model chosen from one of the following families: support vector machine, neural networks (in particular a multi-layer perceptron), decision trees, ensemble methods, and k-nearest neighbor model, in particular whose parameters have been calculated by training, said determined data, for example the variable allocated, in step 2. above implementing model B concerning the category of inhibitor whose presence has been detected, and the kinetics(s) obtained in the preceding step i., and II.iii. characterization as output by model C of the desired inhibitor, the latter being identified among: Rivaroxaban, Apixaban, or Edoxaban, and supplying as output, for example by allocation in a variable, the characterization of the inhibitor determined by model C.By "classification decision model C", however, the same thing is indicated as the above in relation to "model A" or "model B", by analogy.

[0069] According to a particular embodiment, the classification decision model C is a neural network, in particular a multi-layer perceptron, in particular was obtained by training a neural network, in particular a multi-layer perceptron. A more particular embodiment applied to the search for a factor Xa inhibitor is as described in the experimental part. In fact, the step of implementing the model C is a step necessarily implemented by computer, due to the complexity of the mathematical function empirically determined by learning.

[0070] It will be understood that in the context of carrying out step ii. of the preceding paragraph, the classification decision model C in question has been trained on appropriate training data with regard to the context in which it is used. In particular, these data are adapted to the effective carrying out of a classification taking into account the measurement circumstances involving an experimental restart (with regard to the parameters allowing the carrying out of competitive kinetics in the context of step i., in particular with regard to the sample dilution factor and / or the measurement duration used - in other words and logically, the "context" in which the measurements are carried out coincides / corresponds with regard to the training data used, the measurement parameters used (or measurement kits used), and the measurement area targeted for the target sample tested).

[0071] Examples of implementation, in particular models for implementing a neural network, in particular a multi-layer perceptron, are known to those skilled in the art, if necessary by reference to the literature in the field (See for example, non-exhaustively, [Géron, 2017]). More specifically, the scikit-learn library (https: / / scikit-learn.org / ) can be used, for example. The parameters that the skilled person will consider setting are, for example, and independently or not of each other, the number of layers of neurons, the number of neurons per layer and the activation functions of the different neurons. But the user has a certain freedom of implementation with regard to the algorithms and the regularization parameters. The skilled person can in particular refer to Géron, 2017 or Bonaccorso, G.(2017), for the implementation of such models, including with regard to the methods of setting the aforementioned parameters, including where necessary the directives contained explicitly or implicitly in this description.

[0072] In the context of carrying out a new measurement (experimental relaunch) of one or more competition kinetics via the carrying out of a competitive enzymatic assay specific to this particular embodiment, it will be noted that advantageously: this new measurement is carried out on a blood sample obtained from the same subject as that for which a measurement was initially carried out for the “detection” (but distinct, the one used for the initial measurement cannot be reused), said dosage specific to the experimental restart is adapted to the realization of competitive kinetics with respect to a factor Xa inhibitor, naturally with a dilution factor of the sample and / or a measurement duration which is (are) adapted to a competitive situation involving the presence of DOA inhibiting factor Xa, so that the measurement makes sense (the person skilled in the art can, for example by using the verification methods illustrated in the experimental part, verify this adaptation), and preferably, the dilution factor of the sample analyzed in the experimental restart and / or the measurement duration is (are) different from that (those) used for the measurement of the kinetics (s) obtained in step a.above (as part of detection).

[0073] An assay is considered suitable for "achieving competitive kinetics against a factor Xa inhibitor, naturally with a sample dilution factor and / or measurement duration that is(are) suitable for a competitive situation involving the presence of factor Xa-inhibiting DOACs", provided that it is possible to obtain a kinetic plot that conforms to conventional practice in the field. Examples of plots are shown in Figures 5 to 9 . A person skilled in the art is able to adapt the experimental conditions of a dosage test in order to obtain this type of plot deemed compliant, in the sense of being usable for his needs.

[0074] As indicated below, the following parameters can be taken into account: the substrate used must be specific for factor Xa but have little affinity for the latter so as not to interfere with the reaction between the enzyme and its inhibitor. For example, the substrate used must have a Michaelis constant KM considered high, in particular between approximately 10 µM and 1000 µM; and / or The affinity between the enzyme and the substrate being deliberately chosen to be low, the catalytic constant kcat of the enzyme for the substrate must be sufficiently high, for example greater than approximately 10 s-1, or greater than 10 s-1; and / or the initial substrate concentration [S]0 must be sufficient to allow the generation of the marker during the entire duration d of the measurement, which constitutes a conventional adaptation for those skilled in the art (see also formula below in the experimental part). According to one embodiment, the initial substrate concentration [S]0 is less than KM*10; and / or The duration of the measurement is chosen to be sufficiently long to allow the anticoagulant, if present, to exert its inhibitory action on the enzyme and for this to be observable via the measurement; for example a duration of between 10 and 1000 seconds.

[0075] The elements provided in the preceding paragraph relating to the situation where the substrate used is specific for factor Xa apply identically to the situation where the substrate used is specific for factor IIa. In fact, these elements apply generally to any realization of kinetics envisaged in the present description.

[0076] Although it is indicated that the dilution factor of the sample analyzed in the experimental restart and / or the measurement duration is (are) different from that (those) used for the measurement of the kinetics (kinetics) obtained in step a. above (in the context of detection) equivalent to step 1. above, in both cases the ranges of values possible for these two parameters can, for illustrative purposes only, be: For the dilution factor, the sample is diluted, if necessary, in an interval between 1 / 2 and 1 / 50 (the volumes generally used for the kinetic measurements themselves are as indicated below in this description, without this aspect being limiting, since the measurement volumes may vary according to the devices provided for such measurements), and / or For the measurement duration, it is between 10 and 1000 seconds.

[0077] These parameters are relatively universal parameters, but are not limiting because the person skilled in the art may need to adapt them depending on the equipment used for the dosage.

[0078] The dilution may be carried out in any buffer conventionally used for this purpose, as known to those skilled in the art or indicated in manufacturers' instructions.

[0079] In fact, this application covers situations where the sample is diluted, if necessary and in relation to the basic sample obtained from the individual analyzed, in a ratio of 1 / 2, 1 / 3, 1 / 4, 1 / 5, 1 / 6, 1 / 7, 1 / 8, 1 / 9, 1 / 10, 1 / 11, 1 / 12; 1 / 13, 1 / 14, 1 / 15, 1 / 16, 1 / 17, 1 / 18, 1 / 19, 1 / 20, 1 / 21, 1 / 22; 1 / 23, 1 / 24, 1 / 25, 1 / 26, 1 / 27, 1 / 28, 1 / 29, 1 / 30, 1 / 31, 1 / 32; 1 / 33, 1 / 34, 1 / 35, 1 / 36, 1 / 37, 1 / 38, 1 / 39, 1 / 40, 1 / 41, 1 / 42; 1 / 43, 1 / 44, 1 / 45, 1 / 46, 1 / 47, 1 / 48, 1 / 49, 1 / 50, or any interval taken between any of these values.

[0080] In fact, the present application covers situations where the volume used for the kinetic measurements themselves is 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 150, 175, 200, 225, 250, 275, 300, 325, 350, 375, 400 µL, or any intermediate value within a range of values defined as an interval between any of the values just specified.

[0081] According to another variant of implementation of the invention, the method or one of the methods described herein may be implemented on devices dedicated to the point-of-care, for example systems for carrying out unit tests with miniaturization devices, such as for example devices using microfluidics. In this case, and conventionally, the method or one of the methods described herein may be carried out with reaction volumes adapted to the device used or to the implementation context, said volumes being able, for example, and in a non-limiting manner, to be between 1 and 20 µL depending on the miniaturized device used. The incubation times may also, where appropriate and conventionally for those skilled in the art, be adapted according to the characteristics of said device used or the implementation context implemented.For the measurement duration, and for illustrative purposes, it can be 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 120, 130, 140, 150, 160, 170, 180, 190, 200, 210, 220, 230, 240, 250, 260, 270, 280, 290, 300, 310, 320, 330, 340, 350, 360, 370, 380, 390, 400, 410, 420, 430, 440, 450, 460, 470, 480, 490, 500, 510, 520, 530, 540, 550, 560, 570, 580, 590, 600, 610, 620, 630, 640, 650, 660, 670, 680, 690, 700, 710, 720, 730, 740, 750, 760, 770, 780, 790, 800, 810, 820, 830, 840, 850, 860, 870, 880, 890, 900, 910, 920, 930, 940, 950, 960, 970, 980, 990, 1000 seconds, or any intermediate value within a range of values defined as an interval between any of the values just specified.

[0082] Those skilled in the art will nevertheless understand that according to a variant of implementation of the invention on devices dedicated to the point-of-care described above, said measurement duration may be adapted, in a conventional manner for those skilled in the art

[0083] An experimental relaunch, when considered in accordance with the objectives pursued, can make it possible to optimize the identification and / or dosage of a factor Xa inhibitor, present in a more specific, or at least different, concentration range in the sample analyzed blindly. The experimental part provides principles easily reproducible by those skilled in the art to verify that an experimental relaunch is of interest, or that kinetics can actually be obtained.

[0084] According to another aspect, and in a manner analogous to the description given above, a regression model D may be applied following the classification decision model C implemented on the basis of an experimental relaunch.

[0085] However, and according to another embodiment, the identification method, partly carried out in vitro,described here, where appropriate applied to the identification of a factor Xa inhibitor, and comprising an additional step of quantitative dosage of the characterized inhibitor implementing a regression model D, can take advantage of carrying out a new measurement (experimental relaunch) of one or more competition kinetics via carrying out a competitive enzymatic dosage on a blood sample obtained from a (same) subject, not upstream of a classification decision model C as described above, but upstream of the implementation of said regression model D only.In this case, the steps described above are modified to only integrate the performance of a new measurement (experimental relaunch) - according to the same methods as described above - before the implementation of said regression model D, by providing, in a manner analogous to the processes described here, said model D with the kinetics newly obtained due to the experimental relaunch, and the result of the implementation of the classification model C previously implemented. In the context of such an embodiment, the learning and validation of the model C implemented on the one hand, and the learning and validation of the model D implemented on the other hand, will have respectively been carried out on data sets involving kinetic measurement conditions which allow a comparison with the data obtained from patients during the implementation of the method of the invention with a sample.That is to say, logically, in the context of such an embodiment, model C will have been trained on the basis of kinetic measurements carried out under the same conditions as the “initial” measurements described here, and model D will have been trained on the basis of kinetic measurements carried out under the same conditions as the “experimental relaunch” measurements described here.

[0086] It goes without saying that this need to compare data obtained under comparable conditions (and according to a particular embodiment, identical), or to compare data with respect to models having been obtained and / or validated on the basis of measurements carried out under comparable conditions (and according to a particular embodiment, identical), will be taken into account by any person skilled in the art implementing classification or regression models such as those used in the present invention. The remark in this paragraph therefore applies generically to all the embodiments of the present description, in their different combinations.

[0087] According to another aspect, however, the inventors have also implemented a particular embodiment which takes advantage of a recalculation of the concentration of DOA, chosen from rivaroxaban, apixaban and edoxaban, by implementing a second regression model, herein referred to as regression model D2, using as input a kinetics previously measured by the universal methodology as described herein (rather than a kinetics measured by said optimized methodology (for DOAs), the latter having generally been used during a first calculation of the concentration of DOA chosen from rivaroxaban, apixaban and edoxaban, said calculation employing a first regression model D). This particular embodiment is part of a particular succession of preliminary steps represented in the diagram of the Figure 16, including a step 7.1. “Recalculation of the concentration using the kinetics measured by the universal methodology”. The results associated with such an embodiment are illustrated in the experimental part which follows, in Sections 3 and 4, under the terminology “dosage of rivaroxaban / apixaban / edoxaban universal measure ", respectively in sections 3.8 / 4.8 ( Figure 20 , rivaroxaban), 3.10 / 4.10 ( Figure 22 , apixaban), 3.12 / 4.12 ( Figure 24, edoxaban), The use of the expression "universal methodology" in this context reflects the fact that the kinetics used to recalculate the DOA concentration with another regression model, different from the regression model D generally previously used, are kinetics measured by a universal methodology as described here (in fact, such kinetics will generally have already been previously measured upstream during the successive stages of implementation of the method described here). In this context, the expression "universal methodology" may be substituted by "methodology improved on the basis of the universal methodology", if the context should not be sufficient to clarify a difference with the expressions "universal methodology" and "methodology optimized for DOAs" used above and in this description.The expression "improved methodology based on the universal methodology" refers more specifically to the entire chain of models successively implemented to arrive at a "recalculation" step such as the step annotated 7.1 on the . Figure 16 .

[0088] As illustrated in the experimental part, the interest of implementing this recalculation which is placed at a precise moment downstream of the chain of decision-making models as described here, is to obtain a more precise result for the measured concentrations, for the low concentrations of DOA chosen from rivaroxaban, apixaban and edoxaban. Therefore, entry into this particular embodiment is advantageously carried out if the concentration of DOA chosen from rivaroxaban, apixaban and edoxaban, measured with the first regression model, is less than, or less than or equal to, 200 ng / mL.

[0089] Thus, according to this particular embodiment, the method of the invention comprising an additional step of quantitative dosage of the characterized inhibitor (also called step of recalculation of the concentration of the characterized inhibitor) implementing a regression model D2, implemented at the end of the implementation of the regression model D described above if the concentration of inhibitor identified in the analyzed sample, determined by the model D, is less than or equal to 200 ng / mL, then providing as input to a regression model D2, in particular a supervised machine learning model, for example a model chosen from one of the following families: support vector machines, neural networks, decision trees, ensemble methods, and k nearest neighbors model, said model having been trained on a data set obtained under measurement conditions identical to those of step a.above (in the context of detection), of the kinetics(s) obtained in step a. above (in the context of detection), said regression model D2 making it possible to calculate (or rather, recalculate in relation to the calculation already carried out by implementing a regression model D) as output the concentration of inhibitor identified in the analyzed sample and optionally providing as output, for example by allocation in a variable, the concentration determined by the model D2.

[0090] "Allowing to calculate / recalculate the concentration of the identified inhibitor at the output" can also be written as "Allowing to determine the concentration of the identified inhibitor at the output" or "determining the concentration of the identified inhibitor at the output" or "returning the concentration of the identified inhibitor at the output". This data can replace the data already calculated and / or returned by the use of the regression model D, or come in addition, in order to illustrate for example a difference.

[0091] According to one aspect of this embodiment, if the concentration of DOA selected from rivaroxaban, apixaban and edoxaban, measured with the first regression model D, is greater than, or greater than or equal to, 200 ng / mL, then the implementation of a regression model D2 is not implemented, the final result of the chain of decisional or regression models ending with the implementation of the regression model D2.

[0092] Those skilled in the art will appreciate that the threshold value of 200 ng / mL was determined under the experimental conditions tested to optimize the results that can be obtained with the chain of decision or regression models described here.

[0093] A particular embodiment involving a D2 regression model is however only an additional, advantageous characteristic, but the implementation of which is not essential to achieve already relevant results (see the experimental section which describes implementations without this additional additional functionality): it is an additional characteristic at a particular point in a chain of models otherwise generally described here according to its various possible implementations.

[0094] The characteristics modified in the context of an experimental restart compared to those of a measurement of kinetics initially carried out for the realization of a “detection”, illustrated here by a reference to the kinetics obtained in step a. described above, are also recalled above in the present description, and illustrated in the experimental part by means of experiments and directives for the attention of the person skilled in the art, who will thus be able to distinguish the characteristics thereof.

[0095] By "D2 regression model", or "regression model type machine learning model", is meant the same as the above in relation to "Model A", "Model B" or "Model C" or "Model D", by analogy.

[0096] According to a particular embodiment, the model D2 is a neural network, in particular a multi-layer perceptron, in particular was obtained by training a neural network, in particular a multi-layer perceptron. A more particular embodiment applied to the search for a factor Xa inhibitor is as described in the experimental part.

[0097] In fact, the D2 model implementation step is a step necessarily implemented by computer, due to the complexity of the mathematical function empirically determined by learning. A D2 type model is trained on data obtained using a "universal methodology" type methodology, as described here.

[0098] Examples of implementation, in particular models for implementing a neural network, in particular a multi-layer perceptron, are known to those skilled in the art, if necessary by reference to the literature in the field (See for example, non-exhaustively, [Géron, 2017]). More specifically, the scikit-learn library (https: / / scikit-learn.org / ) can be used, for example. The parameters that the skilled person will consider setting are, for example, and independently or not of each other, the number of layers of neurons, the number of neurons per layer and the activation functions of the different neurons. But the user has a certain freedom of implementation with regard to the learning algorithm or the regularization parameters. The skilled person may in particular refer to Géron, 2017 or Bonaccorso, G.(2017), for the implementation of such models, including with regard to the methods for setting the aforementioned parameters, including where necessary the directives contained explicitly or implicitly in the present description. It goes without saying that the references “A”, “B”, “C”, “D” and “D2” for the models discussed in the present description have been indicated to facilitate reading, but that these letters can be omitted without the meaning of what they indicate being modified. The implementation of a D2 type model as described herein is understood in a step of recalculating a concentration when an inhibitor concentration of less than 200 ng / mL has previously been detected, as described herein. According to one embodiment, the measurement . in vitroof competitive kinetics by competitive enzymatic assay on a blood sample obtained from a subject, includes (whether in the context of an initial kinetics measurement or in the context of an experimental restart) the following steps: a. providing a diluted or undiluted blood sample, then b. adding to the blood sample a substrate specific for either factor Xa or factor IIa depending on the targeted enzyme and the desired inhibitor, in particular a substrate carrying a detectable marker, for example a visualizable one (for example a chromogenic, fluorescent or chemofluorescent marker) and in particular a chromogenic or fluorogenic substrate, c. incubation with raising of the temperature of the mixture obtained in b. to a temperature between (including the limits) 35 and 39°C, in particular 37°C, d. adding to the reaction mixture resulting from c. the targeted coagulation enzyme chosen from factor Xa and factor IIa, depending on the substrate added in step b., so as to trigger competition between an inhibition reaction and the enzymatic reaction caused, e.measurement by an instrument, over time, of the quantity of product resulting from the transformation of the substrate due to the action of the enzyme analyzed on the latter (factor Xa or factor Ila), where appropriate via the measurement of a marker associated with the substrate, released during said enzymatic reaction, and recording of the kinetics obtained.

[0099] In one embodiment, the instrument is a Stago STA-R device. This example is not, however, limiting. Any instrument incorporating a spectrophotometer can be used. Other examples include the devices known and marketed by Diagnostica Stago under the names Compact Max, STA-R or STA-R Max.

[0100] According to a particular embodiment, the competitive enzymatic assay is specific for factor Xa, and: a. the blood sample used is a plasma sample diluted 1 / 2 in Owren Koller buffer, b. in step b. the substrate is the MAPA-Gly-Arg-pNA reagent, c. in step c. the incubation time is 240 seconds, at 37°C, d. the factor Xa added to the mixture in step d. is bovine factor Xa (or alternatively according to another embodiment, human or recombinant factor Xa, or any alternative known to those skilled in the art), e. the measurement of the release of paranitroaniline (pNA) in step e. is carried out by colorimetry at 405 nm every two seconds for 156 seconds, on a suitable instrument, of the STA-R type marketed by Diagnostica Stago (see variants cited without limitation above). According to another particular embodiment, the competitive enzymatic assay is specific for factor Xa, and: a. the blood sample used is a plasma sample diluted 1 / 8 in Owren Koller buffer, b. in step b.the substrate is the MAPA-Gly-Arg-pNA reagent, c. in step c. the incubation time is 240 seconds, at 37°C, d. the factor Xa added to the mixture in step d. is bovine factor Xa (or alternatively according to another embodiment, human factor Xa, or recombinant, or any alternative known to those skilled in the art), e. the measurement of the release of paranitroaniline (pNA) in step e. is carried out by colorimetry at 405 nm every two seconds for 86 seconds, on a suitable instrument, of the STA-R type marketed by Diagnostica Stago (see variants cited without limitation above).

[0101] According to a particular embodiment, the competitive enzymatic assay is specific for factor IIa, and: a. the blood sample used is a plasma sample diluted 1 / 12 in TRIS EDTA buffer (in particular pH 8.4), b. in step b. the substrate is the EtM-SPro-Arg-pNA reagent, c. in step c. the incubation time is 240 seconds, in particular at 37°C, d. the factor IIa added to the mixture in step d. is bovine factor IIa (or alternatively according to another embodiment, human or recombinant factor IIa, or any alternative known to those skilled in the art), e. the measurement of the release of paranitroaniline (pNA) in step e. is carried out by colorimetry at 405 nm every two seconds for 156 seconds, on a suitable instrument, of the STA-R type marketed by Diagnostica Stago (see variants cited without limitation above).According to a variant applicable to all the embodiments described generically or specifically in the present application, the competitive enzymatic assay is carried out on a miniaturized device, for example a device using microfluidics, in a reaction volume between 1 and 20 µL. The method according to the invention, according to all its variants described here, can indeed advantageously, taking into account the problems that it makes it possible to solve, be implemented via a miniaturized device. According to one embodiment, applicable to any of the embodiments described here, if the substrate is a fluorometric substrate, then the measurement of the release of the fluorophore, for example AMC, is carried out by fluorometry.

[0102] According to a particular embodiment of the invention, the blood sample analyzed is a plasma sample.

[0103] According to another particular embodiment of the invention, the blood sample analyzed is a whole blood sample.

[0104] In a particular embodiment of the invention, the methods described here allow the quantitative dosage of a factor Xa inhibitor over the following ranges, defined relative to the final concentration [E] in molar in the factor Xa test: a concentration of HNF between approximately [E]*10.0 / 3.0 and approximately [E]*200.0 / 3.0. The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally dose this factor Xa inhibitor in the range [0.1, 2.0] IU / mL; a concentration of LMWH between approximately [E]*20.0 and approximately [E]*400.0. The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally dose this factor Xa inhibitor in the range [0.1, 2.0] IU anti-Xa / mL; a concentration of rivaroxaban between approximately [E] / 6.0 and approximately [E]*3.0. The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally dose this factor Xa inhibitor in the range [20, 600] ng / mL; an apixaban concentration between approximately [E] / 6.0 and approximately [E]*3.0.The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally dose this factor Xa inhibitor in the range [20, 600] ng / mL. an edoxaban concentration between approximately [E] / 8.0 and approximately [E]*3.0. The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally dose this factor Xa inhibitor in the range [20, 600] ng / mL. .

[0105] It should be noted that a person skilled in the art will be able, on the basis of the guidelines provided here, to easily determine the inhibitor concentrations that can be assayed by carrying out the test method defined in the examples. Furthermore, with respect to the indications given above, a person skilled in the art will be able to easily determine that a dilution of the sample leads to obtaining the assay of a target concentration in intervals of the same values, to the nearest dilution factor.

[0106] The selection of appropriate volumes of sample, buffer and reagents constitutes a classic development in the field of enzymatic assays, and of the invention.

[0107] As a particular example of the invention, obtained under the precise experimental conditions of the examples, the methods described here allow the quantitative dosage of a factor Xa inhibitor over the range: from 0.0 to 2.0 IU / mL for UFH; from 0.0 to 2.0 IU anti-Xa / mL for LMWH; from 0.0 to 180.0 ng / mL for rivaroxaban; from 0.0 to 190.0 ng / mL for apixaban; from 0.0 to 230.0 ng / mL for edoxaban.

[0108] According to another particular example, obtained under the precise experimental conditions of the examples, involving an experimental restart as described here, the quantitative dosage of a factor Xa inhibitor is permitted over the range: from 0.0 to 2.0 IU / mL for UFH; from 0.0 to 2.0 anti-Xa IU / mL for LMWH; from 0.0 to 720.0 ng / mL for rivaroxaban; from 0.0 to 760.0 ng / mL for apixaban; from 0.0 to 920.0 ng / mL for edoxaban.

[0109] For a molar conversion of the above values, the following conversion rule will be considered: inhibitor level I in nM = (level of I in ng / mL) / (molecular weight of I in kg / mol) * (sample volume) / (total volume).

[0110] In a particular embodiment of the invention the kinetics of the sample tested are recorded for a period of time between 10 and 10000, or between 10 and 9000, or between 10 and 8000, or between 10 and 7000, or between 10 and 6000, or between 10 and 5000, or between 10 and 4000, or between 10 and 3000, or between 10 and 2000, or between 10 and 1000, or between 10 and 900, or between 10 and 800, or between 10 and 700, or between 10 and 600, or between 10 and 500, or between 10 and 400, or between 10 and 300, or between 10 and 200, or between 20 and 200, or between 30 and 200, or between 40 and 200, or between 50 and 200, or between 60 and 200, or between 70 and 200, or between 80 and 150, seconds. Particular examples are given in the experimental part, which can be applied to any implementation as described here.It follows from the explanations given here that the person skilled in the art is able to adjust the measurement time depending on the verification that kinetics that can be used by him can be obtained (if necessary by following the test guidelines provided in the experimental section).

[0111] Decision-making classification models of type A, B, C or regression models D, including a D2 regression model, are given in the experimental part, as an example: the hyper-parameters that were obtained following optimization on validation data are indicated. However, it goes without saying that the absolute values that can be assigned to these hyperparameters may depend on the data sets used. They cannot therefore be fixed. However, a person skilled in the art can easily define, train and evaluate a machine learning model using the indications provided here as a guide, and the reference works at his disposal (for example, but not limited to Géron, A. (2017) or Bonaccorso, G. (2017), cited here).

[0112] For example: a. a classification decision model A may result from training a support vector machine, or an artificial neural network, in particular a multi-layer perceptron, with a data set comprising kinetics obtained under conditions identical to those implemented when performing the detection step a. described above, in a semi-supervised manner, with, where appropriate, resulting hyper-parameters which are as indicated in the experimental part, and / or b. a classification decision model B may result from training a k-nearest neighbor model, or an artificial neural network, in particular a multi-layer perceptron, with a data set comprising kinetics obtained under conditions identical to those implemented when performing step a.of detection described above, with where appropriate resulting hyper-parameters which are, for example, a value k equal to 5, and the metric being the Euclidean distance, and / or c. a classification decision model C may result from training an artificial neural network, in particular a multi-layer perceptron with a data set comprising kinetics obtained under conditions identical to those implemented when carrying out step a. of detection described above, or obtained under conditions identical to those implemented when carrying out an experimental restart., where appropriate, with for example, resulting hyper-parameters which are as indicated in the experimental part, and / or d.a D, or D2, regression model may result from training an artificial neural network, in particular a multi-layer perceptron with a data set comprising kinetics obtained under conditions identical to those implemented when carrying out the detection step a. described above, or obtained under conditions identical to those implemented when carrying out an experimental restart (in the case of a D regression model), where appropriate, with, for example, resulting hyper-parameters which are as indicated in the experimental part.

[0113] In fact, a learning algorithm according to the invention advantageously implements one or more decision-making models which make it possible to achieve, when tested on test data (making it possible to evaluate the overall and real performance of the trained and optimized model) a precision in the accuracy of the result which it (they) returns, greater than or equal to 70%, or 75%, or 80%, or 85%, or 90%, or 95%, applied to the accumulation of the different models used and / or to each model used.

[0114] Reference is made to the experimental part which indicates, as an example, a particular method of evaluating this performance.

[0115] Accuracy is assessed by comparing the result predicted by said trained model to the actual values. Illustrative examples are provided in Tables 1 to 6 of this description (experimental part). According to another aspect, a regression model, when used, can achieve, when tested on test data (making it possible to evaluate the overall and actual performance of the trained, or even optimized, model) an output result which is characterized by a linear regression slope of between 0.9 and 1.1, and a coefficient of determination R2 greater than or equal to 0.70, or 0.80, or 0.90, or 0.95 (according to the CLSI EP9-A2 criteria). Reference is made to the experimental part which indicates, by way of example, a particular method for evaluating this performance.The performance indicated here being achieved, it will be considered that the result achieved by the method according to the invention is at least qualitatively equivalent to a result obtained via a so-called conventional approach (i.e. a dosage method conventionally used to date, particularly in clinical practice).

[0116] According to one embodiment, the training and validation data used for the supervised machine learning of at least one of the models described here, with regard to the kinetics provided to the models, are of the type that can be used for experiments making it possible to determine pre-calibration or calibration curves of kinetics already conventionally carried out in the field, in particular when marketing kits that can be used to carry out such kinetics. The number of data and the choice of patients that made it possible to obtain them advantageously make it possible to ensure that the data provided will allow effective learning. The experimental part provides quantitative examples of the type of data that can be provided, with a convincing result in terms of reliability. A person skilled in the art is thus, or generally speaking, able to determine the necessary starting samples.

[0117] According to one embodiment, providing experimentally obtained kinetics as input to a model consists of providing the pairs of values constituted by each measured value for each discrete measurement point carried out during the measurement time.

[0118] The invention also relates to a data processing system or device for detecting, in a biological sample, the presence or identification of a factor Xa (FXa) or factor IIa (FIIa) inhibitor, said data processing system or device as defined in claim 13. Such a data processing system or device comprises means for implementing at least step b. of the "detection" method described herein, including the implementation of a model A of any one of the methods described herein, and means for providing as input one or more competition kinetics obtained via the performance of a competitive assay against either a factor Xa inhibitor or a factor IIa inhibitor, on a blood sample previously obtained from a subject, in particular via a measuring device, and means for providing as output the variables generated during this step.Where appropriate, the data processing system or device also comprises means for implementing the step implementing model B of any of the methods described here, and optionally also comprises means for providing as input and / or output the variables generated during these steps, and optionally also comprises means for implementing the steps involving the classification or regression decision models C, D, and where appropriate D2, described here, according to all their variants, in particular means for making use of a matrix resulting from the learning which made it possible to fix the parameters of these models, to return the classification or regression result taking into account the variables supplied to the model(s).

[0119] According to one embodiment, such a data processing system or such a device includes a kinetics measuring apparatus as described herein, in particular a measuring apparatus necessary to carry out a step a. of measuring one or more competition kinetics by carrying out a competitive assay on a blood sample previously obtained from a subject, as described herein. According to another embodiment, such a data processing system or such a device uses such a measuring apparatus, in particular a remote one. In fact, a measuring apparatus could also carry out said measuring step a., under control, in particular from a data processing system or a device, automatically or semi-automatically.

[0120] It should be noted that a so-called "manual" measurement, or at least semi-automatic if a semi-automatic device is used, is not excluded for the implementation of a method according to the invention, such a measurement nevertheless requiring, for its realization, the agents conventionally used in the field of realization of measurement of one or more competition kinetics via the realization of a competitive dosage on a blood sample previously obtained from a subject (see the contents of a kit described here, for an example), in addition to a measuring device.

[0121] The invention also relates to such a data processing system or device, further comprising a processor adapted to implement the steps indicated above and in the present description.

[0122] In particular, and in a non-limiting manner, the invention can be implemented via a computer station (possibly dedicated) connected to a network or via a wired connection, or within an embedded (dedicated) system.

[0123] In fact, it is understood that the steps from step b. of the “detection” method described here, and the subsequent steps since they use models of type A, B, C, D or D2 as described here, require a computer for their implementation. Step a. of measuring one or more competition kinetics can be carried out separately, manually, semi-automatically or automatically, or means can be integrated, in particular into a data processing system or a device described here, to carry out in a controlled manner, from said data processing system or said device, the carrying out of kinetic measurement(s).

[0124] The invention also relates, as defined in claim 14, to a computer program comprising instructions which, when the program is executed by a computer, cause the latter to implement at least step b. of the “detection” method described here. Where appropriate, these instructions also cause the latter to implement the step implementing model B of any one of the methods described here, and optionally the program comprises instructions making it possible to retrieve as input and / or provide as output the variables generated during these steps. Optionally the program comprises instructions causing the latter to implement the steps involving the classification or regression decision models C, D, and where appropriate D2, described here, adapted to any embodiment as described.

[0125] According to a particular embodiment, such a computer program comprises instructions which lead a data processing system or a device, in particular a device including a kinetics measuring apparatus as defined and described in the present application, or using such a measuring apparatus, in particular a remote one, to execute steps at least step b. of the “detection” method described here, and where appropriate also the step implementing model B of any of the methods described here, and optionally lead to executing the steps involving the classification or regression decision models C, D, and where appropriate D2, described here.

[0126] According to a particular embodiment, a computer program further comprises instructions which, when the program is executed by a computer, cause the latter to also implement, in a manner controlled by the instructions of the program, a step a. of measuring one or more competition kinetics via the performance of a competitive assay on a blood sample previously obtained from a subject, as described here.

[0127] Alternatively, a method in vitrofor detecting in a biological sample the presence of an inhibitor of a blood clotting enzyme, according to the invention in all its embodiments set out here, can see its step a. of measuring one or more competition kinetics, carried out separately from a control system centralizing the control of the means necessary for the subsequent steps. Conversely, according to another embodiment, means for carrying out said step a. in a centralized manner, at least in part taking into account the specificity of the measurement carried out, can also be centralized by the same control member in the form of a program executed by a computer as described here.

[0128] The invention also relates, as defined in claim 15, to a computer-readable recording medium comprising instructions which, when executed by a computer, cause the latter to implement at least step b. of the "detection" method described herein. Where appropriate, the instructions also cause the latter to also implement the step implementing the model B of any one of the methods described herein, and optionally also make it possible to recover as input and / or provide as output the variables generated during these steps, and optionally cause the latter to implement the steps involving the classification or regression decision models C, D, and where appropriate D2, described herein.

[0129] According to a particular embodiment, a computer-readable recording medium further comprises instructions which, when executed by a computer, cause the latter to also implement a step a. of measuring one or more competition kinetics via the performance of a competitive assay on a blood sample previously obtained from a subject, as described here.

[0130] The invention also relates, as defined in claim 16, to a computer-readable data carrier on which the computer program described herein is recorded, or a signal from a data carrier carrying the computer program described herein.

[0131] The invention finally relates, as defined in claim 17, to a kit, in particular adapted for the implementation of a method as described here in any embodiment, comprising: a. A substrate specific for FXa or FIIa, e.g., the substrate MAPA-Gly-Arg-pNA for factor Xa, e.g., the substrate EtM-SPro-Arg-pNA for factor IIa, and b. Optionally, FXa and / or FIIa, e.g., bovine factor Xa, e.g., human factor Xa, e.g., bovine factor IIa, e.g., human factor IIa, and c. Optionally, one or more suitable buffers, e.g., Owren Koller buffer, e.g., Tris EDTA buffer, and d. Optionally, instructions for carrying out measurement(s) of one or more competition kinetics by performing a competitive enzyme assay employing the substrate, and e. a system and / or device and / or computer program and / or computer-readable data carrier described herein, f. and optionally instructions for implementing the method according to the invention, where appropriate in connection with the instructions of point d.above, relating to the characteristics of the competitive enzymatic assay to be implemented, g. and optionally instructions relating to the use of a signal from a data carrier described herein, for the implementation of a method according to any one of the embodiments described herein.

[0132] Other characteristics and advantages of the invention appear in the examples which follow as well as in the Figures which illustrate the realization of different particular modes of the invention. BRIEF DESCRIPTION OF THE FIGURES

[0133] Figure 1 . Principle of competitive dosage in vitro of blood clotting enzyme inhibitors: kinetic diagrams. Figure 2 . Principle of measurement.(1) a blood sample (diluted or not) supposed to contain an inhibitor I (and if necessary) the molecule A) of a coagulation enzyme E is placed in a consumable. (2) a substrate S specific to the enzyme E is added to the reaction mixture. (3) an incubation step (generally several tens of seconds) brings the reaction mixture to temperature (typically 37°C). (4) The enzyme E is added to the reaction mixture, thus triggering competition between the inhibition reaction and the enzymatic reaction. (5) the actual measurement is then carried out (generally over a period of several tens of seconds); the enzyme E transforms the substrate S into product P while being inhibited in parallel by the inhibitor I (and if necessary the molecule A) if it(they) is(are) present in the sample. The cleavage of the substrate S into product P induces the release of a label that can be measured by an instrument.(6) the measurement results in the recording of kinetics which is impacted by the presence, mode of action and concentration of inhibitor I (and if necessary of molecule A). Legend: (A) Sample assumed to contain an inhibitor, (B) Enzyme. Figure 3 . Cascade of machine learning models. Diagram legend: 1. Start, 2. Presence of an enzyme inhibitor?, 3. Inhibitor category, 4. Direct Reversible, 5. Indirect Irreversible, 6. Inhibitor identification, 7. Inhibitor quantification, 8. End. Figure 4 Principle of the invention: detection, identification and quantification of inhibitors of a blood coagulation enzyme. Diagram legend: 1. Measurement, 2. Recording of kinetics, 3. Post-processing by machine learning models, 4. Biological result, 5. Detection, identification and quantification of an inhibitor of a blood coagulation enzyme. (1) a blood sample supposed to contain an inhibitor I of a coagulation enzyme E is brought into contact with the enzyme E and a substrate S specific for the enzyme E within the same consumable in vitro.The cleavage of substrate S by enzyme E into a product P causes the release of a marker measurable by a laboratory instrument. (2) The experimental measurement results in the recording of a curve proportional to the concentration of product P over time. This kinetics is impacted by the presence, mode of action and concentration of inhibitor. (3) The kinetics is analyzed and interpreted by a cascade of machine learning models that renders the biological result. (4) The biological result returned indicates whether an inhibitor of enzyme E is present in the measured sample, and if so which one and at what concentration. Figure 5 . Universal anti-Xa methodology : sensitivity to HNF.The figure illustrates the sensitivity of this method to various HNF concentrations: it is observed that it allows to measure HNF concentrations between approximately [E]*10.0 / 3.0 and [E]*200.0 / 3.0 ([E] being the final concentration in the enzyme test). The curves were measured using the particular universal methodology described here ([E]=10 nM, [HNF] (nM)=33,67,133,200,267,333,400,467,533,600 and 670 or [HNF] (IU / mL)=0.1,0.2,0.4,0.6,0.8,1.0,1.2,1.4,1.6,1.8 and 2.0). Figure 6 . Universal anti-Xa methodology : sensitivity to LMWH.The figure illustrates the sensitivity of this method to various LMWH concentrations: it is observed that it can measure LMWH concentrations between approximately [E]*20.0 and [E]*400.0 ([E] being the final concentration in the enzyme assay). The curves were measured using the particular universal methodology described here ([E]=10 nM, [LMWH] (nM)=200,400,800,1200,1600,2000,2200,2400,2600,2800 and 4000 or [LMWH] (anti-Xa IU / mL)=0.1,0.2,0.4,0.6,0.8,1.0,1.2,1.4,1.6,1.8 and 2.0). Figure 7 . Universal anti-Xa methodology : sensitivity to rivaroxaban.The figure illustrates the sensitivity of this method to various rivaroxaban concentrations: it is observed that it allows to measure rivaroxaban concentrations between approximately [E] / 6.0 and [E]*3.0 ([E] being the final concentration in the enzyme assay). The curves were measured using the particular universal methodology described here ([E]=10 nM, [rivaroxaban] (nM)=1.65,3.3,6.6,9.9,13.2,16.5,19.8,23.1,26.4,29.7 and 33 or [rivaroxaban] (ng / mL)=10,20,40,60,80,100,120,140,160,180 and 200). Figure 8 . Universal anti-Xa methodology : sensitivity to apixaban.The figure illustrates the sensitivity of this method to various apixaban concentrations: it is observed that it allows to measure apixaban concentrations between approximately [E] / 6.0 and [E]*3.0 ([E] being the final concentration in the enzyme assay). The curves were measured using the particular universal methodology described here ([E]=10 nM, [apixaban] (nM)=1.56,3.1,6.2,9.3,12.4,15.5,18.6,21.7,24.8,27.9 and 31 or [apixaban] (ng / mL)=10,20,40,60,80,100,120,140,160,180 and 200). Figure 9 . Universal anti-Xa methodology : sensitivity to edoxaban.The figure illustrates the sensitivity of this method to various edoxaban concentrations: it is observed that it allows to measure edoxaban concentrations between approximately [E] / 8.0 and [E]*3.0 ([E] being the final concentration in the enzyme assay). The curves were measured using the particular universal methodology described here ([E]=10 nM, [edoxaban] (nM)=1.3, 2.6, 5.2, 7.8, 10.4, 13.0, 15.6, 18.2, 20.8, 23.4 and 26 or [edoxaban] (ng / mL)=10, 20, 40, 60, 80, 100, 120, 140, 160, 180 and 200). Figure 10 . Cascade of machine learning models for the detection, identification and quantification of synthetic factor Xa inhibitors. Diagram key: 1. Start, 2. Presence of a factor Xa inhibitor?, 3. Category of anti-Xa, 4. Heparin, 5. DOAC, 5.1. Experimental restart, 6. Identification, 6.1. UFH, 6.2. LMWH, 6.3. Rivaroxaban, 6.4. Apixaban, 6.5. Edoxaban, 7. Quantification, 8. Level, 9. End. Figure 11 . Dosage of HNF.A. Simplicity analysis: The results of comparing the HNF levels measured using the approach described here (predicted [HNF]) to the HNF levels measured using the standard approach (measured [HNF]) on the test set data give a straight line with equation y=1.018x-0.0005 and a coefficient of determination R2=0.9844. B. Triplicate analysis: The results of comparing the HNF levels measured using the approach described here (predicted [HNF]) to the HNF levels measured using the standard approach (measured [HNF]) on the test set data give a straight line with equation y=1.002x+0.002 and a coefficient of determination R2=0.9925. Figure 12 . Dosage of LMWH.A. Simplicity analysis: The results of comparing the LMWH levels measured using the approach described here (predicted [LMWH]) to the LMWH levels measured using the standard approach (measured [LMWH]) on the test set data give a straight line with equation y=0.995x+0.006 and a coefficient of determination R2=0.9962. B. Triplicate analysis: The results of comparing the LMWH levels measured using the approach described here (predicted [LMWH]) to the LMWH levels measured using the standard approach (measured [LMWH]) on the test set data give a straight line with equation y=0.9977x+0.004 and a coefficient of determination R2=0.997. Figure 13 . Rivaroxaban dosage.A. Simplicity analysis: The results of comparing rivaroxaban levels measured using the approach described here ([rivaroxaban] predicted) to rivaroxaban levels measured using the standard approach ([rivaroxaban] measured) on the test set data give a straight line with equation y=1.05x+19.65 and a coefficient of determination R2=0.991. B. Triplicate analysis: The results of comparing rivaroxaban levels measured using the approach described here ([rivaroxaban] predicted) to rivaroxaban levels measured using the standard approach ([rivaroxaban] measured) on the test set data give a straight line with equation y=1.04x+20.3 and a coefficient of determination R2=0.9934. Figure 14 . Dosage of apixaban.A. Simplicity analysis: The results of comparing the dosages of apixaban levels measured using the approach described here ([apixaban] predicted) to the apixaban levels measured using the standard approach ([apixaban] measured) on the test set data give a straight line with equation y=1.14x-6.73 and a coefficient of determination R2=0.9945. B. Triplicate analysis: The results of comparing the dosages of apixaban levels measured using the approach described here ([apixaban] predicted) to the apixaban levels measured using the standard approach ([apixaban] measured) on the test set data give a straight line with equation y=1.13x-5.46 and a coefficient of determination R2=0.9958. Figure 15 . Dosage of edoxaban.A. Simplicity analysis: The results of comparing the dosages of edoxaban levels measured using the approach described here ([edoxaban] predicted) to the edoxaban levels measured using the standard approach ([edoxaban] measured) on the data in the test set give a straight line with equation y=0.905x+12.35 and a coefficient of determination R2=0.9853. B. Triplicate analysis: The results of comparing the dosages of edoxaban levels measured using the approach described here ([edoxaban] predicted) to the edoxaban levels measured using the standard approach ([edoxaban] measured) on the data in the test set give a straight line with equation y=0.93x+8.18 and a coefficient of determination R2=0.9881. Figure 16 . Cascade of machine learning models for the detection, identification and quantification of synthetic factor Xa inhibitors. Diagram legend: 1. Onset, 2. Presence of a factor Xa inhibitor?, 3. Category of anti-Xa, 4. Heparin, 5. DOAC, 5.1. Experimental relaunch, 6. Identification, 6.1. UFH, 6.2. LMWH, 6.3. Rivaroxaban, 6.4. Apixaban, 6.5. Edoxaban, 7. Quantification, 7.1. Recalculation of the concentration using the kinetics measured by the universal methodology, 8. Rate, 9. End. Figure 17 . Dosage of HNF. HAS. Simplified analysis : the results of comparing the dosages of HNF levels measured using the approach described in this document ([HNF] predicted) to the HNF levels measured using the standard approach ([HNF] measured) on the data from the test set give a straight line with the equation y =1.043x-0.03 and a coefficient of determination R 2< =0.9851. B. Triplicate analysis : the results of comparing the dosages of HNF levels measured using the approach described in this document ([HNF] predicted) to the HNF levels measured using the standard approach ([HNF] measured) on the data from the test set give a straight line with the equation y =1.04x-0.03 and a coefficient of determination R 2< =0.9856. Figure 18 . Dosage of LMWH. HAS. Simplified analysis: the results of comparing the dosages of LMWH levels measured using the approach described in this document ([LMWH] predicted) to the LMWH levels measured using the standard approach ([LMWH] measured) on the data from the test set give a straight line with the equation y =1.02x-0.02 and a coefficient of determination R 2< =0.996. B. Triplicate analysis : the results of comparing the dosages of LMWH levels measured using the approach described in this document ([LMWH] predicted) to the LMWH levels measured using the standard approach ([LMWH] measured) on the data from the test set give a straight line with the equation y =1.027x-0.03 and a coefficient of determination R 2< =0.9971. Figure 19 . Dosage of rivaroxaban (optimized AODs methodology). HAS. Simplified analysis: the results of comparing the dosages of rivaroxaban levels measured using the approach described in this document ([rivaroxaban] predicted) to the rivaroxaban levels measured using the standard approach ([rivaroxaban] measured) on the data from the test set give a straight line with the equation y =1.05x+19.65 and a coefficient of determination R 2< =0.991. B. Triplicate analysis : the results of comparing the dosages of rivaroxaban levels measured using the approach described in this document ([rivaroxaban] predicted) to the rivaroxaban levels measured using the standard approach ([rivaroxaban] measured) on the data from the test set give a straight line with the equation y =1.04x+20.3 and a coefficient of determination R 2< =0.9934. Figure 20 . Rivaroxaban dosage (universal methodology, or “improved methodology based on the universal methodology”). A. Simplified analysis: the results of comparing the dosages of rivaroxaban levels measured using the approach described in this document ([rivaroxaban] predicted) to the rivaroxaban levels measured using the standard approach ([rivaroxaban] measured) on the data from the test set give a straight line with the equation y =1.04x+1.74 and a coefficient of determination R 2< =0.985. B. Triplicate analysis : the results of comparing the dosages of rivaroxaban levels measured using the approach described in this document ([rivaroxaban] predicted) to the rivaroxaban levels measured using the standard approach ([rivaroxaban] measured) on the data from the test set give a straight line with the equation y =1.032x+1.99 and a coefficient of determination R 2< =0.989. Figure 21 . Dosage of apixaban (optimized AODs methodology). HAS. Simplified analysis: the results of comparing the dosages of apixaban levels measured using the approach described in this document ([apixaban] predicted) to the apixaban levels measured using the standard approach ([apixaban] measured) on the data from the test set give a straight line with the equation y =1.14x-6.73 and a coefficient of determination R 2< =0.9945. B. Triplicate analysis : the results of comparing the dosages of apixaban levels measured using the approach described in this document ([apixaban] predicted) to the apixaban levels measured using the standard approach ([apixaban] measured) on the data from the test set give a straight line with the equation y =1.13 x -5.46 and a coefficient of determination R 2< =0.9958. Figure 22 . Dosage of apixaban (universal methodology, or “improved methodology based on the universal methodology”). A. Simplified analysis: the results of comparing the dosages of apixaban levels measured using the approach described in this document ([apixaban] predicted) to the apixaban levels measured using the standard approach ([apixaban] measured) on the data from the test set give a straight line with the equation y =1.041 x -2.66 and a coefficient of determination R 2< =0.9792. B. Triplicate analysis : the results of comparing the dosages of apixaban levels measured using the approach described in this document ([apixaban] predicted) to the apixaban levels measured using the standard approach ([apixaban] measured) on the data from the test set give a straight line with the equation y =1.041 x -2.41 and a coefficient of determination R 2< =0.9877. Figure 23 . Dosage of edoxaban (optimized AODs methodology). HAS. Simplified analysis: the results of comparing the dosages of edoxaban levels measured using the approach described in this document ([edoxaban] predicted) to the edoxaban levels measured using the standard approach ([edoxaban] measured) on the data from the test set give a straight line with the equation y =0.905 x +12.35 and a coefficient of determination R 2< =0.9853. B. Triplicate analysis : the results of comparing the dosages of edoxaban levels measured using the approach described in this document ([edoxaban] predicted) to the edoxaban levels measured using the standard approach ([edoxaban] measured) on the data from the test set give a straight line with the equation y =0.93 x +8.18 and a coefficient of determination R 2< =0.9881. Figure 24 . Dosage of edoxaban (universal methodology, or “improved methodology based on the universal methodology”). A. Simplified analysis: the results of comparing the dosages of edoxaban levels measured using the approach described in this document ([edoxaban] predicted) to the edoxaban levels measured using the standard approach ([edoxaban] measured) on the data from the test set give a straight line with the equation y =0.9431 x +4.51 and a coefficient of determination R 2< =0.9968. B. Triplicate analysis : the results of comparing the dosages of edoxaban levels measured using the approach described in this document ([edoxaban] predicted) to the edoxaban levels measured using the standard approach ([edoxaban] measured) on the data from the test set give a straight line with the equation y =0.9442 x +4.26 and a coefficient of determination R 2< =0.9962. GENERAL INFORMATION - EXPERIMENTAL METHODOLOGY - PRINCIPLE OF THE METHOD 1. Statement of the principle underlying the invention

[0134] The present invention provides a method for detecting, identifying and quantifying blood clotting enzyme inhibitors in a blood sample. For this purpose, a blood clotting enzyme, a specific substrate for this enzyme and a blood sample are brought together in the same consumable, which can advantageously be a conventional consumable, as conventionally used for enzymatic assays in the field. The enzyme then cleaves the substrate into a product, thus leading to the release of a marker (the marker can for example be chromogenic, fluorescent, etc.); the appearance of this marker induces an observable change in the physical properties of the blood sample: the measurement then consists of recording the evolution of the physical properties of the sample over time by an instrument (The instrument can be a spectrophotometer if the marker is chromogenic, it can be a fluorimeter if the marker is fluorescent, etc.), i.e. recording kinetics. If an inhibitor of the enzyme is present in the blood sample, this will reduce the activity of the enzyme on its substrate and consequently impact the measured kinetics. A post-processing algorithm interprets this kinetics and produces the expected biological result: detection, identification and quantification of the inhibitor. 1.1 Experimental methodology

[0135] Blood clotting enzyme inhibitors, whether natural or synthetic, fall into two distinct families: direct inhibitors and indirect inhibitors. A direct inhibitor binds directly to the enzyme to exert its inhibitory action. An indirect inhibitor first binds to a second molecule to form a complex that can then exert an inhibitory action on the enzyme. In addition, these inhibitors also act according to two different reaction mechanisms: reversible kinetic patterns and irreversible kinetic patterns. A reversible inhibitor binds to the enzyme to form a dissociable complex, unlike an irreversible inhibitor, which binds to the enzyme to form an indissociable complex. Blood clotting enzyme inhibitors are therefore classified into four categories: reversible direct inhibitors, irreversible direct inhibitors, reversible indirect inhibitors, irreversible indirect inhibitors.

[0136] Classically, the enzymatic dosage of these inhibitors involves two biochemical reactions: An inhibition reaction if the inhibitor is direct reversible, the reaction scheme is E + I ⇄ E ⋅ I where inhibitor I binds directly to enzyme E to form the dissociable inactive complex E·I. If the inhibitor is direct irreversible, the reaction scheme is E + I → E ⋅ I where inhibitor I binds directly to enzyme E to form the inactive, indissociable complex E·I. If the inhibitor is indirect, reversible, the reaction scheme is E + A ⋅ I ⇄ E ⋅ A ⋅ I ↑ ↓ A + I where inhibitor I associates with a second molecule A to form the complex A·I capable of binding to the enzyme E to form the dissociable inactive complex E·A·I. If the inhibitor is indirect irreversible, the reaction scheme is E + A ⋅ I ⇄ E ⋅ A ⋅ I ↑ ↓ A + I where inhibitor I combines with a second molecule A to form the A·I complex capable of binding to enzyme E to form the inactive, inseparable E·A·I complex. An enzymatic reaction E + S ⇄ E ⋅ S → E + P where E denotes the blood clotting enzyme targeted by the inhibitor I assumed to be present in the sample; S represents a specific substrate of the enzyme, generally composed of a peptide sequence specific to the active site of the enzyme and a marker which can be chromogenic, fluorescent, electrochemical, etc. E·S denotes the unstable enzyme·substrate complex; P represents the product resulting from the catalysis of the substrate by the enzyme: in our case, the catalysis results in the release of the marker (chromogenic, fluorescent, electrochemical, etc.).

[0137] Note that reversible inhibition reactions follow second-order kinetics for the formation of the enzyme-inhibitor complex and first-order kinetics for its dissociation. Irreversible inhibition reactions follow second-order kinetics for the formation of the enzyme-inhibitor complex. The kinetics of the enzymatic reaction obeys the Henri-Michaelis-Menten formalism [Segel, 1993].

[0138] In order to carry out the enzymatic assay of these different inhibitors, the two biochemical reactions can be successive or in competition. In both cases, the assay is divided into two consecutive steps: incubation and measurement. When the two biochemical reactions are successive, the incubation step consists of bringing the blood sample supposed to contain an inhibitor I (and if necessary molecule A) into contact with the enzyme E present in excess. Inhibitor I (or the A·I complex) then reversibly or irreversibly inhibits enzyme E; the duration of the incubation must be long enough to allow the inhibition reaction to reach its biochemical equilibrium. Thus, the residual concentration of enzyme E is inversely proportional to the initial concentration of inhibitor I and thus reflects its inhibitory activity.The measurement step then consists of adding a substrate S specific to the enzyme E to the reaction mixture; the enzyme E, then present in residual quantity, transforms the substrate S into a product P which induces the release of a marker over time allowing the recording of kinetics, the latter typically being a straight line.

[0139] When the two biochemical reactions are in competition, the incubation step consists of bringing the blood sample supposed to contain an inhibitor I (and if necessary molecule A) into contact with the substrate S: the aim of this step is only to raise the temperature of the reaction mixture to 37°C. The measurement step then consists of adding the enzyme E to the reaction mixture in order to trigger the inhibition reaction and the enzymatic reaction in parallel: we speak of competitive reactions because the inhibitor I and the substrate S are in competition for the enzyme E. Thus, the enzyme E cleaves the substrate S into product P while being inhibited in parallel directly or indirectly by the inhibitor I.The transformation of substrate S into product P induces the release of a marker over time which allows the recording of kinetics: the reaction of transformation of the substrate into product being impacted by the action of the inhibitor on the enzyme, the recorded kinetics is also impacted. Therefore, the concentration of the inhibitor and its mode of action on the enzyme modify the kinetics recorded during the measurement step when the enzymatic assay puts the two biochemical reactions in competition.

[0140] The invention described herein uses the competitive approach; figure 1 synthesizes the kinetic diagrams associated with the dosage in vitro of various blood clotting enzyme inhibitors. The measurement in vitro for the detection, identification and quantification of blood clotting enzyme inhibitors is therefore carried out according to the following principle, illustrated in the figure 2 : a blood sample (diluted or not) supposed to contain an inhibitor I (and if necessary molecule A) of a coagulation enzyme E is placed in a consumable; a substrate S specific to the enzyme E is added to the reaction mixture; an incubation step (generally several tens of seconds) raises the reaction mixture to temperature (typically 37 °C); the enzyme E is added to the reaction mixture, thus triggering competition between the inhibition reaction and the enzymatic reaction; the actual measurement is then carried out (generally over a period of several tens of seconds): the enzyme E transforms the substrate S into product P while being inhibited in parallel by the inhibitor I (and if necessary molecule A) if it (they) is (are) present in the sample. The cleavage of the substrate S into product P induces the release of a label that can be measured by an instrument.the measurement results in the recording of kinetics which is impacted by the presence, mode of action and concentration of inhibitor I (and if necessary of molecule A).

[0141] Given the scope of application of the method described here, which aims more particularly at the detection and where appropriate the identification or even the quantification of anticoagulant inhibitors of factor Xa and / or factor IIa which are either irreversible indirect inhibitors (also referred to as the category of heparins, as described here, in the present description) or direct reversible inhibitors (also referred to as the category of DOACs, as described here, in the present description), the enzyme E added to the reaction mixture triggering the competition is, depending on, factor Xa or factor IIa. Conventionally, this enzyme must be added just before the start of the measurement leading to the recording of kinetics.

[0142] Since the method described here allows, for the first time, a "blind" detection of an inhibitor of factor Xa and / or factor IIa assumed to be contained in the sample analyzed, the question arises as to the range(s) of inhibitor concentration values (of factor Xa and / or factor IIa) assumed to be contained in the sample analyzed, for which the method described here will be effective.

[0143] The determination of the inhibitor concentrations (of factor Xa and / or factor IIa) that the kinetic measurement method specifically used makes it possible to detect, and where appropriate to identify, or even to quantify, can be done according to the procedure described in the “Examples” section, applied for the purposes of the demonstration in the search for a factor Xa inhibitor. Those skilled in the art will understand that these concentration ranges where detection is possible may vary depending on the types of enzyme and substrate used, in a conventional manner in the field of tests for detecting the presence of enzyme, in particular in blood samples. That being said, the methodology set out below in the “Examples” section allows those skilled in the art to determine without difficulty whether the range of inhibitor concentrations that can be sought is suitable for their needs.Furthermore, the methodology presented also shows that it is possible, for example for a given enzyme-substrate pair, in particular known from the prior art, to modulate the sensitivity range of detection, identification or even quantification according to the target range. This can be done, as shown experimentally below on the basis of implementation examples, by modulating the starting dilution of the sample to be analyzed. Thus, in particular, the particular embodiment referred to as “methodology optimized for DOAs” was designed and developed as a variation aimed at more precisely defined needs. A person skilled in the art will thus be able to adapt the dilution rate of the starting sample so as to achieve the target sensitivity range for the inhibitors sought blindly.Illustrative examples are included herein for those particular embodiments that fall within the scope of the novel blind detection method generically disclosed herein.

[0144] Finally, one (or more) post-processing algorithm(s) interpret(s), in particular following a chronological sequence according to the desired objective, the kinetics which is, where appropriate, impacted by the presence, mode of action and concentration of an inhibitor present in the sample (and if necessary of molecule A referred to above), which produces the expected biological result: detection, and / or identification and / or quantification of the inhibitor. This is described in the following section. 1.2 Post-processing

[0145] The previous section described an experimental methodology for obtaining kinetics impacted by the presence, mode of action and concentration of a blood clotting enzyme inhibitor through a competitive enzymatic assay. This section presents a post-processing method, based on Artificial Intelligence (AI) algorithms, interpreting the kinetics to produce the expected biological result: detection, identification and quantification of the inhibitor.

[0146] Artificial Intelligence is a computer science discipline born in the 1950s. Very schematically, its goal is to develop algorithms capable of reproducing the cognitive functions of the human brain. Within these cognitive capacities, learning (set of mechanisms leading to the acquisition of know-how, knowledge or knowledge) is by far the most studied application field in the context of AI. Called machine learning, it is the subject of an innumerable quantity of scientific publications but also and above all, it now finds numerous applications in our daily lives. The present invention makes use of so-called supervised learning models. Unlike classic algorithms, machine learning models are not explicitly programmed for the tasks they have to perform but they are trained to do so.Indeed, a machine learning model establishes a numerical link between an input data and an output data through an empirical mathematical function. Thus, for a given input data, the model calculates the associated output. When the output is an integer, the machine learning model responds to a classification problem (for example, a patient is healthy or a patient is sick); when the output is a real number, the machine learning model responds to a regression problem (for example, an anticoagulant level). The parameters of the empirical mathematical function are calculated by training on a database composed of pairs (input data, output data): a learning algorithm adjusts the parameters so that for a given input data, the model recalculates the associated output data as accurately as possible.Among the most widely used machine learning models, we can cite, for example, neural networks, random forests and support vector machines [Géron, 2017]. The skilled person has at his disposal, following for example the existing literature (notably [Géron, 2017] or Bonaccorso, G. (2017)), conventional ways of optimizing hyper-parameters of a machine learning model, so as to adapt them to the intended objective.

[0147] The classic approach to defining, training and evaluating a machine learning model is carried out on a database that will have been previously divided into three distinct datasets: The training data which will allow the determination of the parameters of the empirical mathematical function linking the pairs (input data, output data); The validation data which will be used to optimize the hyper-parameters of the machine learning model, in accordance with the knowledge of the person skilled in the art - see for example [Géron, 2017] - of the machine learning model and to measure the generalization capacity of the machine learning model trained on the training data; The test data which will allow the evaluation of the overall and real performance of the machine learning model trained on the training data and optimized on the validation data.

[0148] It is important to note that setting up machine learning models necessarily involves having a database made up of pairs (input data, output data). The higher the quantity and quality of this data, the better the training of these models and the greater the generalization capacity.

[0149] The invention proposed here uses, according to a particular embodiment going to the end of the chain of possible and conceivable conclusions, a cascade of machine learning models to analyze and interpret the experimentally measured kinetics. This cascade aims to render the biological result: it is described on the figure 3 , it being understood that depending on the result and the analysis sought, all of the steps may not be implemented.

[0150] The most complete particular cascade described here is composed of four machine learning models; these four models take as input the (same) kinetics obtained by the experimental measurement and give as output a result: The first model (also called model A) is a classification model: depending on the kinetics given to it as input, it determines whether or not the analyzed sample contains an inhibitor of the enzyme; The second model (also called model B) is also a classification model: knowing that the analyzed sample contains an inhibitor of the enzyme and depending on the kinetics given to it as input, it recognizes the category of the inhibitor; The third model (also called model C) is still a classification model: knowing the category of the enzyme inhibitor and depending on the kinetics given to it as input, it identifies the inhibitor present in the analyzed sample;The fourth and last possible model (also called model D) is a regression model: knowing the inhibitor present in the analyzed sample and according to the shape of the kinetics presented to it as input, it calculates the concentration of this inhibitor.

[0151] Each of these models is a machine learning model, for example a neural network, a support vector machine or others, in particular as described below and detailed in the results section with regard to particular embodiments, which is trained by pairs (kinetics, output data) of data from a database divided into three distinct sets (training set, validation set and test set) which will have been generated beforehand. The choice of the machine learning model (neural networks, support vector machines or others) is made according to the performances obtained on the validation set. The experimental section shows a way to analyze the performances obtained on a validation set. The first model is trained by pairs (kinetics, presence = yes or presence = no).The second model is driven by pairs (kinetics, inhibitor category = direct reversible or inhibitor category = direct irreversible or inhibitor category = indirect reversible or inhibitor category = indirect irreversible). The third model is driven by pairs (kinetics, inhibitor name = inhibitor1 or inhibitor name = inhibitor2 or inhibitor name = ...). The fourth model is driven by pairs (kinetics, inhibitor concentration). Finally, cascading these different models gives the expected biological result, depending on the desired result: detection, identification and quantification of an inhibitor of a blood clotting enzyme. 1.3 Summary

[0152] The purpose of this application is the detection of the presence, the identification, or even the quantification in vitroof blood clotting enzyme inhibitors, as described herein. Generally speaking, blood clotting enzyme inhibitors, whether natural or synthetic, fall into four categories: reversible direct inhibitors, irreversible direct inhibitors, reversible indirect inhibitors, and irreversible indirect inhibitors. Each category has its own biochemical reaction mechanism. Thus, in order to perform the assay, a blood sample assumed to contain an inhibitor I of a blood clotting enzyme E is brought into contact within the same consumable with the enzyme E and a substrate S specific to the enzyme E. The experimental measurement consists of putting into competition an inhibition reaction (between E and I) and an enzymatic reaction (between E and S).The cleavage of substrate S by enzyme E into product P causes the release of a label that will induce an observable change in the physical properties of the sample which is recorded by a measuring instrument over time. The resulting kinetics is impacted by the presence, the biochemical reaction mechanism and the concentration of inhibitor. These kinetics are then analyzed and interpreted by a cascade of machine learning models that renders the expected biological result. The . figure 4 summarizes the principle underlying the present invention. EXAMPLES 2. Application example: detection, identification and quantification of synthetic factor Xa inhibitors 2.1 Introduction

[0153] Factor Xa is a blood clotting enzyme which, when combined with factor Va on a phospholipid membrane and in the presence of calcium, forms the prothrombinase enzyme complex responsible for the activation of prothrombin into thrombin (or factor Ha). Prothrombin activated into thrombin is then capable of transforming soluble fibrinogen into an insoluble fibrin clot, the final step in the blood clotting cascade called fibrinoformation. Thus, factor Xa is an enzyme with a key role in the blood clotting process: promoting its activity promotes and amplifies blood clotting, while restricting its activity decreases and slows blood clotting. Many therapies therefore target factor Xa to reduce its activity and thus prevent the occurrence or recurrence of thromboembolic events such as phlebitis or pulmonary embolism.These include heparins and direct oral anticoagulants (anti-Xa).

[0154] Heparin is an anticoagulant drug administered subcutaneously or intravenously. To inhibit factor Xa, heparin combines with a plasma protein: antithrombin. Antithrombin is a natural inhibitor of blood clotting enzymes such as factor Ha, factor Xa, factor IXa and, to a lesser extent, factor VIIa (with or without tissue factor), factor XIa and factor Xlla. To exert its inhibitory action, antithrombin binds irreversibly to the active sites of these different enzymes; when combined with heparin, its inhibitory activity is increased by a factor of 1000. Heparin is therefore, among other things, an irreversible indirect inhibitor of factor Xa. There are two main families of heparins: unfractionated heparins (UFH) and low molecular weight heparins (LMWH). There is a third family of heparins: pentasaccharides such as fondaparinux or idraparinux.HNF mainly accentuates the action of antithrombin on thrombin while LMWH potentiates the action of antithrombin mainly on factor Xa.

[0155] Direct oral anticoagulants (DOACs) are anticoagulant drugs administered orally. They are divided into two classes: anti-Xa direct inhibitors of factor Xa, and anti-Ila direct inhibitors of thrombin. In the context of this proof of concept, a focus was made solely on the family of anti-Xa whose molecules available on the market are: rivaroxaban or Xarelto ®< marketed by Bayer / Janssen Pharmaceutical, apixaban or Eliquis ®< marketed by Bristol-Myers Squibb / Pfizer and edoxaban or Lixiana ®< / Savaysa ®< marketed by Daiichi Sankyo. These three molecules inhibit factor Xa by binding directly and reversibly to the active site of the enzyme: these three molecules are therefore reversible direct inhibitors of factor Xa.

[0156] As a proof of concept, this section describes the application of the method described in section 1 to the detection, identification and quantification of synthetic factor Xa inhibitors. The elements described here can be easily transposed to the detection, identification and quantification of synthetic factor IIa inhibitors. On this point, it is also noted that heparins are irreversible indirect inhibitors of both factors Xa and IIa. 2.2 Principle 2.2.1 Experimental methodology 2.2.1.1 Sensitivity of the method according to the invention to different concentrations of inhibitor(s) sought, and adaptation techniques, if necessary

[0157] The commercial methods currently available for the measurement of anti-Xa anticoagulants such as heparins and DOACs all operate on the same principle: an enzymatic measurement with a dedicated experimental methodology, a dedicated calibration and controls and calibrators dedicated to each molecule (UFH, LMWH, rivaroxaban, apixaban and edoxaban). We also know the Stago STA ®< - Multi-Hep Calibrator kit which allows the measurement of both UFH and LMWH using a common methodology and a hybrid calibration. However, this kit does not allow the detection of DOACs.

[0158] The approach proposed here allows the implementation of a "universal methodology" sensitive to the presence of the majority of anti-Xa anticoagulants, i.e. heparins (UFH and LMWH) and DOACs (rivaroxaban, apixaban and edoxaban). To do this, a volume of plasma from the sample to be assayed is diluted in a buffer. A specific substrate for factor Xa is added to the reaction mixture and the whole is incubated for elevation to 37°C. Finally, the addition of factor Xa triggers the reaction and the measurement is carried out over several seconds.

[0159] The invention concerns a necessarily competitive assay: also, the enzyme must be the triggering reagent and therefore be added last before initiating the measurement.

[0160] For a final molar concentration [E] in the factor Xa test, it has been observed and experimentally demonstrated that the methodology herein entitled “universal methodology” allows the measurement of HNF concentrations between approximately [E]*10.0 / 3.0 and approximately [E]*200.0 / 3.0, as illustrated in the Figure 5 The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally dose the interval [0.1, 2.0] IU / mL.

[0161] For a final molar concentration [E] in the factor Xa test, it has been observed and experimentally demonstrated that the methodology herein entitled “universal methodology” allows the measurement of LMWH concentrations between approximately [E]*20.0 and approximately [E]*400.0, as illustrated in the figure 6 The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally measure the interval [0.1, 2.0] IU anti-Xa / mL.

[0162] For a final molar concentration [E] in the factor Xa test, it has been observed and experimentally demonstrated that the methodology herein entitled “universal methodology” allows the measurement of rivaroxaban concentrations between approximately [E] / 6.0 and approximately [E]*3.0, as illustrated in the figure 7 The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally measure the interval [20, 600] ng / mL.

[0163] For a final molar concentration [E] in the factor Xa test, it has been observed and experimentally demonstrated that the methodology herein entitled “universal methodology” allows the measurement of apixaban concentrations between approximately [E] / 6.0 and approximately [E]*3.0, as illustrated in the figure 8The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally measure the interval [20, 600] ng / mL.

[0164] For a final molar concentration [E] in the factor Xa test, it has been observed and experimentally demonstrated that the methodology herein entitled “universal methodology” allows the measurement of edoxaban concentrations between approximately [E] / 8.0 and approximately [E]*3.0, as illustrated in the figure 9 The different volumes of sample, buffer and reagents are chosen and adjusted in order to ideally measure the interval [20, 600] ng / mL.

[0165] Since the measurement is competitive, the substrate must of course, in the context of the particular embodiment described here, be specific for factor Xa, but have little affinity for the latter so as not to interfere with the reaction between the enzyme and its inhibitor; for this, the substrate must have a high Michaelis constant KM, approximately between 10 µM and 1000 µM. Since the affinity between the enzyme and the substrate is deliberately chosen to be low, the catalytic constant kcat of the enzyme for the substrate must be sufficiently high, for example greater than approximately 10 s-1, to generate the product P (i.e. the label) and allow a signal to be measured. A high catalytic constant also allows the concentration of the enzyme · substrate complex to be minimized. Finally, the initial substrate concentration [S]0 must be sufficient to allow the generation of the label throughout the duration d of the measurement, for example respecting the following inequality S 0 > v max S 0 K M + S 0 d

[0166] However, it should not be too high to interfere minimally with the reaction between the enzyme and its inhibitor: its value can for example be less than KM*10.

[0167] The duration of the measurement is chosen to be long enough to allow the anticoagulant, if present, to exert its inhibitory action on the enzyme and for this to be observable via the measurement; for example a duration of between 10 and 1000 seconds. 2.2.1.2 Specific application examples

[0168] Synthetic factor Xa inhibitors are heparins, which are irreversible indirect inhibitors, and anti-Xa DOACs, which are reversible direct inhibitors.

[0169] Finally, a particular example of application under the methodology called "universal methodology" was implemented with the commercial kit Stago STA ®< - Liquid Anti-Xa. The person skilled in the art will be able to easily adapt the proof of concept provided here, to the use for the initial dosage of any kit having the same aim as that used here, or to the composition of such a kit, in particular to adapt to the nature of a given enzyme-substrate pair, the proof of concept provided here not being limiting in itself for the realization of the method according to the invention. Where appropriate, elements allowing, for greater ease of implementation, the choice of an appropriate kit are reported above. These indications can be: the substrate used must be specific for factor Xa but have little affinity for the latter so as not to interfere with the reaction between the enzyme and its inhibitor. For example, the substrate used must have a Michaelis constant KM considered high, in particular between approximately 10 µM and 1000 µM; The affinity between the enzyme and the substrate being deliberately chosen to be low, the catalytic constant kcat of the enzyme for the substrate must be sufficiently high, for example greater than approximately 10 s-1, or greater than 10 s-1; The initial concentration of substrate [S]0 must be sufficient to allow the generation of the marker throughout the duration d of the measurement, without being too high to interfere at least with the reaction between the enzyme and its inhibitor: its value may for example be less than KM*10;The duration of the measurement, which is not in itself part of a commercially available kit, is preferably chosen to be long enough to allow the anticoagulant, if present, to exert its inhibitory action on the enzyme and for this to be observable via the measurement; for example, a duration of between 10 and 1000 seconds. Specific examples are described below. ;

[0170] According to this particular embodiment, the measurement in vitro for the detection, identification, and quantification of synthetic factor Xa inhibitors was carried out according to the following principle: 25 µl of plasma sample presumed to contain heparin or anti-Xa DOAC are diluted in 25 µl of Owren Koller (TOK) buffer, then placed in a consumable; 150 µl of the reagent "Substrate" (MAPA-Gly-Arg-pNA) (reagent from the commercial kit STA ®< - Liquid Anti-Xa) specific for factor Xa are added to the reaction mixture; a 240-second incubation step raises the reaction mixture to 37°C; 150 µl of the reagent "F. Xa" (bovine factor Xa) (reagent from the commercial kit STA ®< - Liquid Anti-Xa) are added to the reaction mixture, thus triggering competition between the inhibition reaction and the enzymatic reaction; the actual measurement is then carried out over 156 seconds: factor Xa transforms the substrate into product while being inhibited in parallel by heparin or DOA if they are present in the sample.The cleavage of the substrate into product induces the release of paranitroaniline (pNA) which is measured by colorimetry at 405 nm every two seconds on a STA-R ®< type instrument. The measurement results in the recording of kinetics which is impacted by the presence, mode of action and concentration of these synthetic inhibitors of factor Xa.

[0171] Furthermore, according to another particular embodiment, the methodology described above has been optimized to allow both the detection of the presence of a synthetic factor Xa inhibitor, the recognition of the category of the synthetic factor Xa inhibitor, the identification of heparin as well as the dosage of UFH and LMWH. The first methodology, referred to as "universal" above, does not in fact allow the dosage of DOAs over the entire desired concentration range (this range can, naturally, vary according to the objectives), A second methodology has been optimized for the identification of DOAs as well as for the dosage of rivaroxaban, apixaban and edoxaban. It is carried out according to the following principle: 6.25 µl of plasma sample containing an anti-Xa DOA are diluted in 43.75 µl of Owren Koller (TOK) buffer, then placed in a consumable; 150 µl of the reagent "Substrate" (MAPA-Gly-Arg-pNA, reagent of the commercial kit STA ®< - Liquid Anti-Xa) specific for factor Xa are added to the reaction mixture; a 240-second incubation step raises the reaction mixture to 37°C; 150 µl of the reagent "F. Xa" (bovine factor Xa, reagent of the commercial kit STA ®< - Liquid Anti-Xa) are added to the reaction mixture, thus triggering competition between the inhibition reaction and the enzymatic reaction; The actual measurement is then carried out over 86 seconds: factor Xa transforms the substrate into product while being inhibited in parallel by the AOD present in the sample. The cleavage of the substrate into product induces the release of paranitroaniline (pNA) which is measured by colorimetry at 405 nm every two seconds on a STA-R ®< type instrument.the measurement results in the recording of kinetics which is impacted by the presence, mode of action and concentration of the anti-Xa AOD.

[0172] Note that the second methodology is identical to the first methodology except for the sample dilution factor. The measurement duration also differs: the optical density is measured every two seconds until t = 86 s. In this description, the characteristics of the first methodology may be referred to as corresponding to a particular embodiment of a “universal methodology” and the characteristics of the second methodology may be referred to as corresponding to a particular embodiment of an “AODs optimized methodology”.

[0173] In fact, with regard to the detection sensitivity of the two particular examples of methodology reported here, according to the implementation methods described here (in particular enzyme-substrate pair, sample dilution and measurement duration), the verifications carried out and shown in Figures 5 to 9 , commented above, allow us to show the following.

[0174] For the universal methodology, as applied above to “factor Xa”, the final concentrations in the test are: in enzyme: [E]≈ 10 nM in substrate: [S]≈ 482 µM if the inhibitor is an UFH: 0 ≤ [I] ≤ 670 nM or 2 IU / mL; an LMWH: 0 ≤ [I] ≤ 4000 nM or 2 anti-Xa IU / mL; rivaroxaban: 0 ≤ [I] ≤ 30 nM or 180 ng / mL; apixaban: 0 ≤ [I] ≤ 30 nM or 190 ng / mL; edoxaban: 0 ≤ [I] ≤ 30 nM or 230 ng / mL;

[0175] In addition, the catalytic constant kcat and the Michaelis constant KM associated with the enzyme-substrate reaction are approximately 400 s-1 and 500 µM, respectively.

[0176] For the optimized AODs methodology the final concentrations in the test are: in enzyme: same as above; in substrate: same as above; if the inhibitor is rivaroxaban: 0 ≤ [I] ≤ 30 nM or 720 ng / mL; of apixaban: 0 ≤ [I] ≤ 30 nM or 760 ng / mL; of edoxaban: 0 ≤ [I] ≤ 30 nM or 920 ng / mL;

[0177] The catalytic constant kcat and the Michaelis constant KM associated with the enzyme-substrate reaction have the same values as before.

[0178] In summary, the approach described here allows, using the universal methodology, to dose UFH, LMWH, rivaroxaban, apixaban and edoxaban over the range: from 0.0 to 2.0 IU / mL for UFH; from 0.0 to 2.0 IU anti-Xa / mL for LMWH; from 0.0 to 180.0 ng / mL for rivaroxaban; from 0.0 to 190.0 ng / mL for apixaban; from 0.0 to 230.0 ng / mL for edoxaban.

[0179] It allows, using the optimized AODs methodology as applied above to “factor Xa”, to dose rivaroxaban, apixaban and edoxaban over the range (experimental results not illustrated here, but obtained following manipulations identical to those which made it possible to obtain the results of the figures 5 to 9 , with the adjustments made in the methodology here called AOD optimized methodology): from 0.0 to 720.0 ng / mL for rivaroxaban; from 0.0 to 760.0 ng / mL for apixaban; from 0.0 to 920.0 ng / mL for edoxaban.

[0180] Thus, through a single method and without calibration, the approach proposed and described in the present application makes it possible to detect the presence and to measure five anti-Xa molecules in contrast to current commercial methods which require and impose a dedicated methodology and a calibration per molecule. It also makes it possible to identify the molecule, something impossible today. An assay more particularly applied to the search for a factor IIa inhibitor can comprise the following steps and / or modalities: Sample = 175µL Dilution to 1 / 12 with reagent R2 of the STA-Stachrom Heparin kit ∘ Composition: TRIS EDTA ph 8.4 ∘ Preparation: 15 mL bottle QSP 150mL Incubation 240 sec (in particular at a temperature as indicated in the other examples, or the present application) Ra = 75µL = STA-Stachrom ATIII Substrate ∘ Composition: CBS 61.50 chromogenic substrate, approximately 1.4µmol of EtM-SPro-Arg-pNA, AcOH per mL of reconstituted reagent. ∘ Reconstitution with 6 mL of distilled water Rd = 50µL = STA-Stachrom ATIII Thrombin ∘ Composition: Bovine thrombin, approximately 11.3 nKat per mL after reconstitution ∘ Reconstitution: with 6 mL of previously diluted R2 STA-Stachrom Heparin. 2.2.2 Post-processing

[0181] There figure 10details the cascade of machine learning models, representative of the particular embodiment implemented here, which analyzes and interprets the kinetics obtained by the experimental measurement in order to produce the expected biological result.

[0182] The cascade is composed of nine machine learning models; these nine models take as input the kinetics obtained by the experimental measurement and give as output a result: The first model (Model A) is a classification model: based on the kinetics given as input, it determines whether or not the analyzed sample contains a synthetic factor Xa inhibitor; The second model (Model B) is also a classification model: knowing that the analyzed sample contains a synthetic factor Xa inhibitor and based on the kinetics given as input, it recognizes the category of the synthetic anti-Xa inhibitor: heparin or DOAC; If the inhibitor is a heparin, and based on the kinetics given as input, a classification model (Model C) identifies whether it is an UFH or an LMWH; If the inhibitor is an UFH, and based on the kinetics given as input, a regression model calculates the UFH concentration;If the inhibitor is an LMWH, and according to the shape of the kinetics given to it as input, a regression model calculates the LMWH concentration; According to a particular embodiment, if the inhibitor is a DOAC, an experimental measurement is relaunched using the optimized DOAC methodology detailed in this description and new kinetics are recorded and then presented as input to a classification model (Model C) which identifies whether the inhibitor is rivaroxaban, apixaban or edoxaban; If the inhibitor is rivaroxaban, and according to the shape of the kinetics presented to it as input, a regression model calculates the rivaroxaban concentration; If the inhibitor is apixaban, and according to the shape of the kinetics presented to it as input, a regression model calculates the apixaban concentration;If the inhibitor is edoxaban, and according to the shape of the kinetics presented to it as input, a regression model calculates the edoxaban concentration.

[0183] Finally, a regression model D can be used to perform a dosage.

[0184] The following sections list respectively for each of these machine learning models, the training and validation datasets which were used for their training, as well as the associated algorithms, so as to illustrate a proof of concept at the basis of the present invention.

[0185] Note that for each data set, the concentrations of synthetic factor Xa inhibitors (UFH, LMWH, rivaroxaban, apixaban and edoxaban) were measured on a STA-R ®< automaton using the STA-R ®< -Liquid Anti-Xa commercial kit, as well as the associated commercial calibrators and methodologies. Naturally, the invention which is the subject of the present application can also be carried out, if necessary in accordance with the manufacturers' guidelines, using the contents of different kits for its implementation. This experimental part details methods for ensuring correct transposition, in particular with regard to the ranges of values of the inhibitors which can be detected.Finally, it will be noted that, logically, the conditions in which the kinetic measurements were carried out must coincide between the samples to be analyzed by the method of the invention, and those which were used to produce the learning and validation data, for each learning model considered.

[0186] Detection of the presence or absence of an anti-Xa anticoagulant

[0187] Datasets Training data 3 kinetics measured on the STA-R ®< AUT05450 on a plasma spiked with 0.0 IU / ml of calcium UFH (Calciparine ®< ); 3 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.0 IU / ml of sodium UFH (Heparin Choay ®< ); 3 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.0 IU anti-Xa / ml of LMWH (Fragmine ®< ); 3 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.0 IU anti-Xa / ml of LMWH (Lovenox ®< ); 3 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.0 IU anti-Xa / ml of LMWH (Innohep ®< ); 3 kinetics measured on the STA-R ®< AUT05016 on a plasma spiked with 0.0 IU / ml of calcium UFH (Calciparine ®< ); 3 kinetics measured on the STA-R ®< AUT00603 on a plasma spiked with 0.0 IU / ml of sodium UFH (Heparin Choay ®< ); 3 kinetics measured on the STA-R ®< AUT05016 on a plasma spiked with 0.0 IU anti-Xa / ml of LMWH (Fragmine ®< ); 3 kinetics measured on the STA-R ®< AUT00603 on a plasma spiked with 0.0 IU anti-Xa / ml of LMWH (Lovenox ®< ); 3 kinetics measured on the STA-R ®< AUT00603 on a plasma spiked with 0.0 IU anti-Xa / ml of LMWH (Innohep ®< ); 3 kinetics measured on the STA-R ®< AUT06399 on a plasma spiked with 0.0 IU / ml of calcium UFH (Calciparine ®< ); 3 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 0.0 IU / ml of sodium HNF (Heparin Choay ®< ); 3 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 0.0 IU anti-Xa / ml of LMWH (Fragmine ®< ); 3 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 0.0 IU anti-Xa / ml of LMWH (Lovenox ®< ); 3 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 0.0 IU anti-Xa / ml of LMWH (Innohep ®< ); 3 kinetics measured on the STA-R ®< AUT06399 on a plasma overloaded by 0.0 ng / ml of rivaroxaban (Xarelto ®< ); 3 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 0.0 ng / ml of apixaban (Eliquis ®< ); 3 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 0.0 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); 3 kinetics measured on the STA-R ®< AUT00603 on plasma spiked with 0.0 ng / ml of rivaroxaban (Xarelto ®< ); 3 kinetics measured on the STA-R ®< AUT00603 on plasma spiked with 0.0 ng / ml apixaban (Eliquis ®< ); 3 kinetics measured on the STA-R ®< AUT00603 on plasma spiked with 0.0 ng / ml edoxaban (Lixiana ®< / Savaysa ®< ); 3 kinetics measured on the STA-R ®< AUT06366 on plasma spiked with 0.0 ng / ml rivaroxaban (Xarelto ®< ); 3 kinetics measured on the STA-R ®< AUT06366 on plasma spiked with 0.0 ng / ml apixaban (Eliquis ®< ); 3 kinetics measured on the STA-R ®< AUT06366 on a plasma overloaded by 0.0 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); 3 kinetics measured on the STA-R ®< AUT05450 on plasma spiked with 0.0 ng / ml of rivaroxaban (Xarelto ®< ); 3 kinetics measured on the STA-R ®< AUT05450 on plasma spiked with 0.0 ng / ml of apixaban (Eliquis ®< ); 3 kinetics measured on the STA-R ®< AUT05450 on plasma spiked with 0.0 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); Measurements carried out in triplicate for each spike; Universal methodology.

[0188] Machine learning model: A support vector machine (One Class SVM) was trained with the previously described dataset (plasmas not overloaded with factor Xa inhibitors) in a semi-supervised manner. Hyper-parameter optimization was performed using the leave-one-out cross-validation technique. The model hyper-parameters are: Kernel function: Radial Basis Function y≈3.77 10-5 v≈0.0131 Identification of the anti-Xa anticoagulant category

[0189] Datasets Training data 60 kinetics measured on the STA-R ®< AUT05450 on a plasma spiked with 0.12, 0.225, 0.33, 0.43, 0.55, 0.66, 0.75, 0.9, 0.875, 0.995, 1.13, 1.24, 1.315, 1.44, 1.53, 1.63, 1.7, 1.88, 1.955 and 2.005 IU / ml of calcium HNF (Calciparine ®< ); 60 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.11, 0.21, 0.32, 0.43, 0.545, 0.625, 0.635, 0.76, 0.87, 0.965, 1.14, 1.17, 1.295, 1.425, 1.52, 1.605, 1.695, 1.835, 1.84 and 1.99 IU / ml of sodium HNF (Heparin Choay ®< ); 60 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.145, 0.245, 0.35, 0.44, 0.535, 0.64, 0.735, 0.85, 0.935, 1.07, 1.14, 1.24, 1.265, 1.44, 1.495, 1.55, 1.66, 1.745, 1.785 and 2.255 IU anti-Xa / ml of LMWH (Fragmine ®< );60 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.105, 0.19, 0.3, 0.4, 0.485, 0.57, 0.685, 0.78, 0.865, 0.945, 1.035, 1.145, 1.23, 1.32, 1.42, 1.495, 1.58, 1.69, 1.755 and 1.805 IU anti-Xa / ml of LMWH (Lovenox ®< ); 60 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.11, 0.2, 0.295, 0.395, 0.465, 0.535, 0.63, 0.795, 0.885, 0.955, 1.055, 1.21, 1.27, 1.37, 1.455, 1.515, 1.64, 1.81, 1.83 and 1.97 IU anti-Xa / ml of LMWH (Innohep ®< ); 60 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 10, 20, 30, 38, 46, 64, 72, 82, 90, 101, 110, 115, 130, 140, 148, 152, 166, 164 and 192 ng / ml of rivaroxaban (Xarelto ®< ); 60 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 10, 20, 32, 41, 46, 57, 65, 76, 79, 96, 105, 116, 125, 135, 138, 152, 161, 169, 180 and 186 ng / ml of apixaban (Eliquis ®< );60 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 10, 21, 29, 39, 48, 60, 73, 81, 92, 107, 115, 129, 136, 129, 127, 149, 163, 181, 195 and 200 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); 60 kinetics measured on the STA-R ®< AUT00603 on plasma spiked with 10, 20, 35, 41, 49, 56, 66, 73, 84, 93, 105, 110, 116, 134, 143, 155, 160, 175, 181 and 198 ng / ml of rivaroxaban (Xarelto ®< ); 60 kinetics measured on the STA-R ®< AUT00603 on plasma spiked with 10, 20, 20, 26, 36, 43, 53, 60, 70, 75, 93, 103, 110, 121, 135, 136, 152, 160, 175, 185 and 191 ng / ml of apixaban (Eliquis ®< ); 60 kinetics measured on the STA-R ®< AUT00603 on plasma spiked with 10, 21, 27, 40, 49, 64, 73, 90, 99, 108, 122, 132, 144, 130, 136, 153, 169, 192, 194 and 216 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< );60 kinetics measured on the STA-R ®< AUT06366 on plasma spiked with 10, 20, 25, 34, 43, 50, 62, 74, 84, 93, 104, 108, 118, 133, 143, 159, 158, 176, 173 and 202 ng / ml of rivaroxaban (Xarelto ®< ); 60 kinetics measured on the STA-R ®< AUT06366 on plasma spiked with 10, 20, 25, 34, 44, 50, 61, 70, 74, 94, 103, 110, 122, 135, 138, 147, 162, 173, 182 and 186 ng / ml of apixaban (Eliquis ®< ); 60 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 10, 20, 28, 40, 49, 64, 74, 95, 103, 110, 122, 132, 143, 131, 132, 157, 171, 195, 191 and 201 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); Measurements carried out in triplicate for each spike; Universal methodology. Validation data 60 kinetics measured on the STA-R ®< AUT05016 on plasma spiked with 0.095, 0.205, 0.315, 0.44, 0.555, 0.655, 0.77, 0.84, 0.855, 0.98, 1.15, 1.275, 1.335, 1.43, 1.555, 1.6, 1.75, 1.845, 1.92 and 1.96 IU / ml of calcium HNF (Calciparine ®< );60 kinetics measured on the STA-R ®< AUT00603 on a plasma spiked with 0.115, 0.23, 0.32, 0.455, 0.535, 0.655, 0.66, 0.78, 0.865, 0.995, 1.175, 1.2, 1.355, 1.47, 1.555, 1.64, 1.735, 1.8, 1.88 and 2.025 IU / ml of sodium HNF (Heparin Choay ®< ); 60 kinetics measured on the STA-R ®< AUT05016 on a plasma spiked with 0.165, 0.255, 0.345, 0.445, 0.54, 0.63, 0.73, 0.82, 0.92, 1.045, 1.125, 1.16, 1.3, 1.355, 1.475, 1.57, 1.685, 1.75, 1.835 and 2.16 IU anti-Xa / ml of LMWH (Fragmine ®< ); 60 kinetics measured on the STA-R ®< AUT00603 on a plasma spiked with 0.1, 0.195, 0.31, 0.4, 0.515, 0.625, 0.71, 0.83, 0.925, 1.015, 1.085, 1.205, 1.28, 1.36, 1.49, 1.575, 1.65, 1.81, 1.85 and 1.965 IU anti-Xa / ml of LMWH (Lovenox ®< ); 60 kinetics measured on the STA-R ®< AUT00603 on a plasma spiked with 0.14, 0.22, 0.305, 0.42, 0.5, 0.56, 0.66, 0.83, 0.905, 0.995, 1.07, 1.25, 1.275, 1.365, 1.47, 1.615, 1.66, 1.805, 1.9 and 1.975 IU anti-Xa / ml of LMWH (Innohep ®< );60 kinetics measured on the STA-R ®< AUT05450 on plasma spiked with 10, 20, 35, 41, 48, 59, 65, 77, 85, 95, 103, 113, 121, 134, 145, 160, 161, 171, 169 and 202 ng / ml of rivaroxaban (Xarelto ®< ); 60 kinetics measured on the STA-R ®< AUT05450 on plasma spiked with 10, 20, 27, 39, 41, 52, 61, 67, 76, 93, 96, 105, 123, 127, 140, 145, 155, 167, 184 and 182 ng / ml of apixaban (Eliquis ®< ); 60 kinetics measured on the STA-R ®< AUT05450 on a plasma spiked with 10, 21, 28, 41, 49, 62, 74, 92, 101, 112, 120, 132, 142, 129, 133, 155, 168, 188, 200 and 206 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); Measurements carried out in triplicate for each spike; Universal methodology. ; Machine learning model

[0190] A k-nearest neighbor model was trained with the dataset described above. Its hyperparameters are: k = 5; Metric: Euclidean distance. Identification of heparins

[0191] Datasets Training data 60 kinetics measured on the STA-R ®< AUT05450 on plasma spiked with 0.12, 0.225, 0.33, 0.43, 0.55, 0.66, 0.75, 0.9, 0.875, 0.995, 1.13, 1.24, 1.315, 1.44, 1.53, 1.63, 1.7, 1.88, 1.955 and 2.005 IU / ml of calcium UFH (Calciparine ®< ); 60 kinetics measured on the STA-R ®< AUT05016 on a plasma overloaded with 0.095, 0.205, 0.315, 0.44, 0.555, 0.655, 0.77, 0.84, 0.855, 0.98, 1.15, 1.275, 1.335, 1.43, 1.555, 1.6, 1.75, 1.845, 1.92 and 1.96 IU / ml of calcium HNF (Calciparine ®< ); 60 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.11, 0.21, 0.32, 0.43, 0.545, 0.625, 0.635, 0.76, 0.87, 0.965, 1.14, 1.17, 1.295, 1.425, 1.52, 1.605, 1.695, 1.835, 1.84 and 1.99 IU / ml of sodium HNF (Heparin Choay ®< ); 60 kinetics measured on the STA-R ®< AUT00603 on a plasma overloaded by 0.115, 0.23, 0.32, 0.455, 0.535, 0.655, 0.66, 0.78, 0.865, 0.995, 1.175, 1.2, 1.355, 1.47, 1.555, 1.64, 1.735, 1.8, 1.88 and 2.025 IU / ml of sodium HNF (Heparin Choay ®< ); 60 kinetics measured on the STA-R ®< AUT06366 on a plasma overloaded with 0.145, 0.245, 0.35, 0.44, 0.535, 0.64, 0.735, 0.85, 0.935, 1.07, 1.14, 1.24, 1.265, 1.44, 1.495, 1.55, 1.66, 1.745, 1.785 and 2.255 IU anti-Xa / ml of LMWH (Fragmine ®< ); 60 kinetics measured on the STA-R ®< AUT05016 on a plasma spiked with 0.165, 0.255, 0.345, 0.445, 0.54, 0.63, 0.73, 0.82, 0.92, 1.045, 1.125, 1.16, 1.3, 1.355, 1.475, 1.57, 1.685, 1.75, 1.835 and 2.16 IU anti-Xa / ml of LMWH (Fragmine ®< ); 60 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.105, 0.19, 0.3, 0.4, 0.485, 0.57, 0.685, 0.78, 0.865, 0.945, 1.035, 1.145, 1.23, 1.32, 1.42, 1.495, 1.58, 1.69, 1.755 and 1.805 IU anti-Xa / ml of LMWH (Lovenox ®< ); 60 kinetics measured on the STA-R ®< AUT00603 on a plasma overloaded by 0.1, 0.195, 0.31, 0.4, 0.515, 0.625, 0.71, 0.83, 0.925, 1.015, 1.085, 1.205, 1.28, 1.36, 1.49, 1.575, 1.65, 1.81, 1.85 and 1.965 anti-Xa IU / ml of LMWH (Lovenox ®< ); 60 kinetics measured on the STA-R ®< AUT06366 on plasma spiked with 0.11, 0.2, 0.295, 0.395, 0.465, 0.535, 0.63, 0.795, 0.885, 0.955, 1.055, 1.21, 1.27, 1.37, 1.455, 1.515, 1.64, 1.81, 1.83 and 1.97 anti-Xa IU / ml of LMWH (Innohep ®< ); 60 kinetics measured on the STA-R ®< AUT00603 on a plasma spiked with 0.14, 0.22, 0.305, 0.42, 0.5, 0.56, 0.66, 0.83, 0.905, 0.995, 1.07, 1.25, 1.275, 1.365, 1.47, 1.615, 1.66, 1.805, 1.9 and 1.975 IU anti-Xa / ml of LMWH (Innohep ®< ); Measurements carried out in triplicate for each spike; Universal methodology. Validation data 24 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75 and 2.0 IU / ml of calcium UFH (Calciparine ®< ); 24 kinetics measured on the STA-R ®< AUT06399 on plasma spiked with 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75 and 2.0 IU / ml of sodium HNF (Heparin Choay ®< ); 24 kinetics measured on the STA-R ®< AUT06399 on a plasma spiked with 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75 and 2.0 IU anti-Xa / ml of LMWH (Fragmine ®< ); 24 kinetics measured on the STA-R ®< AUT06399 on a plasma spiked with 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75 and 2.0 IU anti-Xa / ml of LMWH (Lovenox ®< ); 24 kinetics measured on the STA-R ®< AUT06399 on a plasma spiked with 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75 and 2.0 IU anti-Xa / ml of LMWH (Innohep ®< ); Measurements carried out in triplicate for each spike; Universal methodology. Machine learning model

[0192] A multi-layer perceptron (neural network) was trained with the dataset described above. Its hyper-parameters are: 77 neurons in the input layer; 27 neurons in the hidden layer; 3 neurons in the output layer; Activation function for each neuron in the different hidden layers: Rectified Linear Unit (ReLU); Activation function for each neuron in the output layer: softmax. Unfractionated heparin dosages

[0193] Datasets Training data 63 kinetics measured on the STA-R ®< AUT05450 on plasma spiked with 0.0, 0.12, 0.225, 0.33, 0.43, 0.55, 0.66, 0.75, 0.9, 0.875, 0.995, 1.13, 1.24, 1.315, 1.44, 1.53, 1.63, 1.7, 1.88, 1.955 and 2.005 IU / ml of calcium UFH (Calciparine ®< ); 63 kinetics measured on the STA-R ®< AUT06366 on plasma spiked with 0.0, 0.11, 0.21, 0.32, 0.43, 0.545, 0.625, 0.635, 0.76, 0.87, 0.965, 1.14, 1.17, 1.295, 1.425, 1.52, 1.605, 1.695, 1.835, 1.84 and 1.99 IU / ml of sodium HNF (Heparin Choay ®< ); Measurements carried out in triplicate for each spike; Universal methodology. Validation data 63 kinetics measured on the STA-R ®< AUT05016 on plasma spiked with 0.0, 0.095, 0.205, 0.315, 0.44, 0.555, 0.655, 0.77, 0.84, 0.855, 0.98, 1.15, 1.275, 1.335, 1.43, 1.555, 1.6, 1.75, 1.845, 1.92 and 1.96 IU / ml of calcium UFH (Calciparine ®< ); 63 kinetics measured on the STA-R ®< AUT00603 on a plasma overloaded by 0.0, 0.115, 0.23, 0.32, 0.455, 0.535, 0.655, 0.66, 0.78, 0.865, 0.995, 1.175, 1.2, 1.355, 1.47, 1.555, 1.64, 1.735, 1.8, 1.88 and 2.025 IU / ml of sodium UFH (Heparin Choay ®< ); Measurements carried out in triplicate for each overload; Universal methodology. . Machine learning model

[0194] A multi-layer perceptron (neural network) was trained with the dataset described above. Its hyper-parameters are: 77 neurons in the input layer; 27 neurons in the hidden layer; 1 neuron in the output layer; Activation function for each neuron in the different hidden layers: Rectified Linear Unit (ReLU). Low molecular weight heparin assays

[0195] Datasets Training data 63 kinetics measured on the STA-R ®< AUT06366 on plasma spiked with 0.0, 0.145, 0.245, 0.35, 0.44, 0.535, 0.64, 0.735, 0.85, 0.935, 1.07, 1.14, 1.24, 1.265, 1.44, 1.495, 1.55, 1.66, 1.745, 1.785 and 2.255 anti-Xa IU / ml of LMWH (Fragmine ®< ); 63 kinetics measured on the STA-R ®< AUT06366 on plasma spiked with 0.0, 0.105, 0.19, 0.3, 0.4, 0.485, 0.57, 0.685, 0.78, 0.865, 0.945, 1.035, 1.145, 1.23, 1.32, 1.42, 1.495, 1.58, 1.69, 1.755 and 1.805 IU anti-Xa / ml of LMWH (Lovenox ®< ); 63 kinetics measured on the STA-R ®< AUT06366 on a plasma spiked with 0.0, 0.11, 0.2, 0.295, 0.395, 0.465, 0.535, 0.63, 0.795, 0.885, 0.955, 1.055, 1.21, 1.27, 1.37, 1.455, 1.515, 1.64, 1.81, 1.83 and 1.97 IU anti-Xa / ml of LMWH (Innohep ®< ); Measurements carried out in triplicate for each spike; Universal methodology. Validation data 63 kinetics measured on the STA-R ®< AUT05016 on a plasma overloaded by 0.0, 0.165, 0.255, 0.345, 0.445, 0.54, 0.63, 0.73, 0.82, 0.92, 1.045, 1.125, 1.16, 1.3, 1.355, 1.475, 1.57, 1.685, 1.75, 1.835 and 2.16 anti-Xa IU / ml of LMWH (Fragmine ®<); 63 kinetics measured on the STA-R ®< AUT00603 on a plasma overloaded by 0.0, 0.1, 0.195, 0.31, 0.4, 0.515, 0.625, 0.71, 0.83, 0.925, 1.015, 1.085, 1.205, 1.28, 1.36, 1.49, 1.575, 1.65, 1.81, 1.85 and 1.965. UI anti-Xa / ml of LMWH (Lovenox ®< ); 63 kinetics measured on the STA-R ®< AUT00603 on plasma spiked with 0.0, 0.14, 0.22, 0.305, 0.42, 0.5, 0.56, 0.66, 0.83, 0.905, 0.995, 1.07, 1.25, 1.275, 1.365, 1.47, 1.615, 1.66, 1.805, 1.9 and 1.975 UI anti-Xa / ml of LMWH (Innohep ®< ); Measurements carried out in triplicate for each overload; Universal methodology. Machine learning model

[0196] A multi-layer perceptron (neural network) was trained with the dataset described above. Its hyper-parameters are: 77 neurons in the input layer; 27 neurons in the hidden layer; 1 neuron in the output layer; Activation function for each neuron in the different hidden layers: Rectified Linear Unit (ReLU). Identification of anti-Xa DOAC

[0197] Datasets Training data 60 kinetics measured on the STA-R ®< AUT00460 on plasma spiked with 10, 20, 29, 37, 44, 54, 62, 89, 115, 143, 160, 192, 226, 252, 287, 319, 331, 361, 391 and 407 ng / ml of rivaroxaban (Xarelto ®< ); 69 kinetics measured on the STA-R ®< AUT00460 on plasma spiked with 10, 20, 31, 42, 48, 72, 92, 116, 143, 163, 192, 217, 247, 276, 298, 321, 348, 370, 397, 416, 439, 459 and 473 ng / ml of apixaban (Eliquis ®< ); 69 kinetics measured on the STA-R ®< AUT00460 on a plasma spiked with 16, 21, 30, 39, 51, 76, 100, 128, 142, 155, 187, 215, 245, 269, 284, 322, 347, 359, 380, 393, 410, 426 and 436 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); Measurements carried out in triplicate for each spike; Optimized methodology AODs.Validation data 60 kinetics measured on the STA-R ®< AUT00460 on plasma spiked with 10, 20, 29, 37, 44, 54, 62, 89, 115, 143, 160, 192, 226, 252, 287, 319, 331, 361, 391 and 407 ng / ml of rivaroxaban (Xarelto ®< ); 69 kinetics measured on the STA-R ®< AUT00460 on plasma spiked with 10, 20, 31, 42, 48, 72, 92, 116, 143, 163, 192, 217, 247, 276, 298, 321, 348, 370, 397, 416, 439, 459 and 473 ng / ml of apixaban (Eliquis ®< ); 69 kinetics measured on the STA-R ®< AUT00460 on a plasma overloaded with 16, 21, 30, 39, 51, 76, 100, 128, 142, 155, 187, 215, 245, 269, 284, 322, 347, 359, 380, 393, 410, 426 and 436 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); Measurements carried out in triplicate for each overload; Optimized methodology AODs. . Machine learning model

[0198] A multi-layer perceptron (neural network) was trained with the dataset described above. Its hyper-parameters are: 42 neurons in the input layer; 29 neurons in the first hidden layer; 16 neurons in the second hidden layer; 3 neurons in the output layer; Activation function for each neuron in the different hidden layers: Rectified Linear Unit (ReLU); Activation function for each neuron in the output layer: softmax. Rivaroxaban dosage

[0199] Datasets Training data 63 kinetics measured on the STA-R ®< AUT00460 on plasma spiked with 0, 10, 20, 29, 37, 44, 54, 62, 89, 115, 143, 160, 192, 226, 252, 287, 319, 331, 361, 391 and 407 ng / ml of rivaroxaban (Xarelto ®< ); Measurements performed in triplicate for each spike; Optimized methodology AODs. Validation data 63 kinetics measured on the STA-R ®< AUT00460 on plasma spiked with 0, 10, 20, 29, 37, 44, 54, 62, 89, 115, 143, 160, 192, 226, 252, 287, 319, 331, 361, 391 and 407 ng / ml of rivaroxaban (Xarelto ®< ); Measurements carried out in triplicate for each spike; Optimized methodology AODs. Machine learning model

[0200] A multi-layer perceptron (neural network) was trained with the dataset described above. Its hyper-parameters are: 42 neurons in the input layer; 29 neurons in the first hidden layer; 16 neurons in the second hidden layer; 1 neuron in the output layer; Activation function for each neuron in the different hidden layers: Rectified Linear Unit (ReLU). Apixaban dosage

[0201] Datasets Training data 72 kinetics measured on the STA-R ®< AUT00460 on plasma spiked with 0, 10, 20, 31, 42, 48, 72, 92, 116, 143, 163, 192, 217, 247, 276, 298, 321, 348, 370, 397, 416, 439, 459 and 473 ng / ml of apixaban (Eliquis ®< ); Measurements performed in triplicate for each spike; Optimized methodology AODs. Validation data 72 kinetics measured on the STA-R ®< AUT00460 on plasma spiked with 0, 10, 20, 31, 42, 48, 72, 92, 116, 143, 163, 192, 217, 247, 276, 298, 321, 348, 370, 397, 416, 439, 459 and 473 ng / ml of apixaban (Eliquis ®< ); Measurements carried out in triplicate for each spike; Optimized methodology AODs. Machine learning model

[0202] A multi-layer perceptron (neural network) was trained with the dataset described above. Its hyper-parameters are: 42 neurons in the input layer; 29 neurons in the first hidden layer; 16 neurons in the second hidden layer; 1 neuron in the output layer; Activation function for each neuron in the different hidden layers: Rectified Linear Unit (ReLU). Edoxaban dosage

[0203] Datasets Training data 72 kinetics measured on the STA-R ®< AUT00460 on a plasma spiked with 0, 16, 21, 30, 39, 51, 76, 100, 128, 142, 155, 187, 215, 245, 269, 284, 322, 347, 359, 380, 393, 410, 426 and 436 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); Measurements carried out in triplicate for each spike; Optimized methodology AODs. Validation data 72 kinetics measured on the STA-R ®< AUT00460 on plasma spiked with 0, 16, 21, 30, 39, 51, 76, 100, 128, 142, 155, 187, 215, 245, 269, 284, 322, 347, 359, 380, 393, 410, 426 and 436 ng / ml of edoxaban (Lixiana ®< / Savaysa ®< ); Measurements carried out in triplicate for each spike; Optimized methodology AODs. Machine learning model

[0204] A multi-layer perceptron (neural network) was trained with the dataset described above. Its hyper-parameters are: 42 neurons in the input layer; 29 neurons in the first hidden layer; 16 neurons in the second hidden layer; 1 neuron in the output layer; Activation function for each neuron in the different hidden layers: Rectified Linear Unit (ReLU). 2.3 Results

[0205] In this section, the performance results obtained by the different machine learning models of the particular cascade of the figure 10 on data measured on real samples using the experimental methodology described in Section 2.2.1. Each subsection details the measured data as well as the performance obtained. One measurement in simplicate is sufficient to report a result; however, as demonstrated in this section, a measurement in triplicate can improve performance. 2.3.1 Detection of the presence or absence of an anti-Xa anticoagulant

[0206] In this section, the results obtained for the detection of the presence or absence of an anti-Xa anticoagulant using the invention are given. Test data

[0207] Simplicity analysis: 298 kinetics measured on 298 true samples (39 normal plasmas, 259 plasmas from patients under anti-Xa anticoagulants); Triplicate analysis: 894 kinetics measured on 298 true samples (39 normal plasmas, 259 plasmas from patients under anti-Xa anticoagulants); The instrument is the STA-R ®< AUT06399; Universal methodology. [Table 1]

[0208] Table 1: Presence or absence of an anti-Xa anticoagulant: confusion matrix. Simplicity analysis. Predicted outcome Presence of anti-Xa Absence of anti-Xa Real value Presence of anti-Xa 257 2 Absence of anti-Xa 1 38

[0209] The results give an accuracy of 99.61% for detecting the presence of an anti-Xa anticoagulant. The results give an accuracy of 97.44% for detecting the absence of an anti-Xa anticoagulant. [Table 2]

[0210] Table 2: Presence or absence of an anti-Xa anticoagulant: confusion matrix. Triplicate analysis. Predicted outcome Presence of anti-Xa Absence of anti-Xa Real value Presence of anti-Xa 257 2 Absence of anti-Xa 1 38

[0211] The results give an accuracy of 99.61% for detecting the presence of an anti-Xa anticoagulant. The results give an accuracy of 97.44% for detecting the absence of an anti-Xa anticoagulant. Results

[0212] Tables 1 and 2 give respectively the confusion matrices associated with the detection of the presence or absence of an anti-Xa anticoagulant when the analysis is carried out in simplicat and when the analysis is carried out in triplicate on the data of the test set. The results for the detection of the presence of an anti-Xa anticoagulant give an accuracy of 99.61% when the analysis is carried out in simplicat and an accuracy of 99.61% when the analysis is carried out in triplicate. The results for the detection of the absence of an anti-Xa anticoagulant give an accuracy of 97.44% when the analysis is carried out in simplicat and an accuracy of 97.44% when the analysis is carried out in triplicate. 2.3.2 Identification of the anti-Xa anticoagulant category

[0213] In this section, the results obtained for the identification of the category of the anti-Xa anticoagulant using the particular embodiment of the invention described here are given.

[0214] In this section, the results obtained for the identification of the category of the anti-Xa anticoagulant using the particular embodiment of the invention described here are given. Test data

[0215] Simplicity analysis: 259 kinetics measured on 259 true samples (91 plasmas from patients under heparin, 168 plasmas from patients under anti-Xa DOACs); Triplicate analysis: 777 kinetics measured on 259 true samples (91 plasmas from patients under heparin, 168 plasmas from patients under anti-Xa DOACs); The instrument is the STA-R ®< AUT06399; Universal methodology. [Table 3]

[0216] Table 3: Identification of the anti-Xa anticoagulant category: confusion matrix. Simplicity analysis. The results give an accuracy of 98.07% for the identification of the anti-Xa anticoagulant category. Predicted outcome Heparin AOD Real value Heparin 86 5 AOD 0 168 [Table 4]

[0217] Table 4: Identification of anti-Xa anticoagulant category: confusion matrix. Triplicate analysis. The results give an accuracy of 97.68% for the identification of the category of anti-Xa anticoagulant. Predicted outcome Heparin AOD Real value Heparin 86 5 AOD 1 167 Results

[0218] Tables 3 and 4 give respectively the confusion matrices associated with the identification of the anti-Xa anticoagulant category when the analysis is carried out in simplicat and when the analysis is carried out in triplicate on the data of the test set. The results for the identification of the anti-Xa anticoagulant category give an accuracy of 98.07% when the analysis is carried out in simplicat and an accuracy of 97.68% when the analysis is carried out in triplicate. 2.3.3 Identification of heparins

[0219] In this section, the results obtained for the identification of heparins using the particular embodiment of the invention described here are given. Test data

[0220] Simplicity analysis: 91 kinetics measured on 91 true samples (29 true samples from patients under UFH, 62 true samples from patients under LMWH); Triplicate analysis: 273 kinetics measured on 91 true samples (29 true samples from patients under UFH, 62 true samples from patients under LMWH); The instrument is the STA-R ®< AUT06399; Universal methodology. [Table 5]

[0221] Table 5: Heparin identification: confusion matrix. Simplicity analysis. The results give an accuracy of 92.31% for heparin identification. Predicted outcome HNF HBPM Real value HNF 25 4 HBPM 3 59 [Table 6]

[0222] Table 6: Heparin identification: confusion matrix. Triplicate analysis. The results give an accuracy of 93.41% for heparin identification. Predicted outcome HNF HBPM Real value HNF 26 3 HBPM 3 59 Results

[0223] Tables 5 and 6 give the confusion matrices associated with the identification of heparins when the analysis is carried out in simplicat and when the analysis is carried out in triplicate on the data of the test set, respectively. The results for the identification of heparins give an accuracy of 92.31% when the analysis is carried out in simplicat and an accuracy of 93.41% when the analysis is carried out in triplicate. 2.3.4 Dosages of unfractionated heparins

[0224] In this section, the results of the HNF level assays on patient samples obtained using the particular embodiment of the invention described here are given, in comparison with the levels measured using the standard approach (STA ®< - Liquid Anti-Xa commercial kit). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R2 is greater than or equal to 0.95 (CLSI EP9-A2 criteria). Test data

[0225] Simplicity analysis: 24 kinetics measured on 24 real samples from patients under HNF; Triplicate analysis: 72 kinetics measured on 24 real samples from patients under HNF; The instrument is the STA-R ®< AUT06399; Universal methodology. Results

[0226] There figure 11gives the results of comparing the HNF levels measured using the approach described here to the HNF levels measured using the standard approach on the test set data. These comparisons give a straight line with equation y=1.018x-0.0005 and a coefficient of determination R2=0.9844 when the analysis is carried out in simplicity; they give a straight line with equation y=1.002x+0.002 and a coefficient of determination R2=0.9925 when the analysis is carried out in triplicate. 2.3.5 Assays of low molecular weight heparins

[0227] In this section, the results of the LMWH level assays on patient samples obtained using the particular embodiment of the invention described here are given in comparison with the levels measured using the standard approach (STA ®< - Liquid Anti-Xa commercial kit). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R2 is greater than or equal to 0.95 (CLSI EP9-A2 criteria). Test data

[0228] Simplicity analysis: 62 kinetics measured on 62 real samples from patients under LMWH; Triplicate analysis: 186 kinetics measured on 62 real samples from patients under LMWH; The instrument is the STA-R ®< AUT06399; Universal methodology. Results

[0229] There figure 12gives the results of comparing the dosages of LMWH levels measured using the approach described here to the LMWH levels measured using the standard approach on the data from the test set. These comparisons give a straight line with equation y=0.995x+0.006 and a coefficient of determination R2=0.9962 when the analysis is carried out in simplicity; they give a straight line with equation y=0.9977x+0.004 and a coefficient of determination R2=0.997 when the analysis is carried out in triplicate. 2.3.6 Identification of anti-Xa DOAC

[0230] In this section, the results obtained for the identification of anti-Xa AODs using the particular embodiment of the invention described here are given. Test data

[0231] Simplicity analysis: 168 kinetics measured on 168 true samples (65 true samples from patients under rivaroxaban, 45 true samples from patients under apixaban, 58 true samples from patients under edoxaban); Triplicate analysis: 504 kinetics measured on 168 true samples (65 true samples from patients under rivaroxaban, 45 true samples from patients under apixaban, 58 true samples from patients under edoxaban); The instrument is the STA-R ®< AUT06399; Optimized methodology AODs. Results

[0232] Tables 7 and 8 give respectively the confusion matrices associated with the identification of anti-Xa DOAs when the analysis is carried out in simplification and when the analysis is carried out in triplicate on the data of the test set. The results for the identification of anti-Xa DOAs give an accuracy of 91.67% when the analysis is carried out in simplification and an accuracy of 96.43% when the analysis is carried out in triplicate. [Table 7]

[0233] Table 7: Identification of anti-Xa DOAs: confusion matrix. Simplicity analysis. The results give an accuracy of 91.67% for the identification of anti-Xa DOAs. Predicted outcome Rivaroxaban Apixaban edoxaban Real value Rivaroxaban 61 1 3 Apixaban 0 45 0 edoxaban 10 0 48 [Table 8]

[0234] Table 8: Identification of anti-Xa DOAs: confusion matrix. Triplicate analysis. The results give an accuracy of 96.43% for the identification of anti-Xa DOAs. Predicted outcome -1 Rivaroxaban Apixaban edoxaban Real value Rivaroxaban 1 64 0 0 Apixaban 0 0 45 0 edoxaban 0 5 0 53 2.3.7 Rivaroxaban dosage

[0235] In this section, the results of the rivaroxaban level assays on patient samples obtained using the particular embodiment of the invention described here are given, in comparison with the levels measured using the standard approach (STA ®< - Liquid Anti-Xa commercial kit). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R2 is greater than or equal to 0.95 (CLSI EP9-A2 criteria). Test data

[0236] Simplicity analysis: 62 kinetics measured on 62 real samples of patients under rivaroxaban; Triplicate analysis: 186 kinetics measured on 62 real samples of patients under rivaroxaban; The instrument is the STA-R ®< AUT06399; Optimized methodology AODs. Results

[0237] There figure 13gives the results of comparison of the dosages of rivaroxaban levels measured using the approach of the particular embodiment of the invention described here with the rivaroxaban levels measured using the standard approach on the data of the test set. These comparisons give a straight line of equation y=1.05x+19.65 and a coefficient of determination R2=0.991 when the analysis is carried out in simplicity; they give a straight line of equation y=1.04x+20.3 and a coefficient of determination R2=0.9934 when the analysis is carried out in triplicate. 2.3.8 Dosage of apixaban

[0238] In this section, the results of the assays of apixaban levels on patient samples obtained using the particular embodiment of the invention described here are given, in comparison with the levels measured using the standard approach (commercial kit STA ®< - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). Test data

[0239] Simplicity analysis: 45 kinetics measured on 45 real samples of patients under apixaban; Triplicate analysis: 135 kinetics measured on 45 real samples of patients under apixaban; The instrument is the STA-R ®< AUT06399; Optimized methodology AODs. Results

[0240] There figure 14gives the results of comparison of the dosages of the apixaban levels measured using the approach of the particular embodiment of the invention described here with the apixaban levels measured using the standard approach on the data of the test set. These comparisons give a straight line of equation y=1.14x-6.73 and a coefficient of determination R 2< =0.9945 when the analysis is carried out in simplicity; they give a straight line of equation y=1.13x-5.46 and a coefficient of determination R 2< =0.9958 when the analysis is carried out in triplicate. 2.3.9 Dosage of edoxaban

[0241] In this section, the results of the edoxaban level assays on patient samples obtained using the particular embodiment of the invention described here are given, in comparison with the levels measured using the standard approach (STA ®< commercial kit - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). Test data

[0242] Simplicity analysis: 56 kinetics measured on 56 real samples of patients under edoxaban; Triplicate analysis: 168 kinetics measured on 56 real samples of patients under edoxaban; The instrument is the STA-R ®< AUT06399; Optimized methodology AODs. Results

[0243] There figure 15gives the results of comparison of the dosages of edoxaban levels measured using the approach of the particular embodiment of the invention described here with the edoxaban levels measured using the standard approach on the data of the test set. These comparisons give a straight line of equation y=0.905x+12.35 and a coefficient of determination R 2< =0.9853 when the analysis is carried out in simplicity; they give a straight line of equation y=0.93x+8.18 and a coefficient of determination R 2< =0.9881 when the analysis is carried out in triplicate. 3. Second post-processing study of the detection, identification and quantification of synthetic factor Xa inhibitors

[0244] The figure Error! Reference source not found. 6 details the cascade (followed in this second study) of machine learning models that analyze and interpret the kinetics obtained by the experimental measurement in order to produce the expected biological result.

[0245] The cascade is composed of twelve machine learning models; these twelve models take as input the kinetics obtained by the experimental measurement and give as output a result: The first model is a classification model: based on the kinetics given as input, it determines whether or not the analyzed sample contains a synthetic factor Xa inhibitor; The second model is also a classification model: knowing that the analyzed sample contains a synthetic factor Xa inhibitor and based on the kinetics given as input, it recognizes the category of the synthetic anti-Xa inhibitor: heparin or DOAC; If the inhibitor is a heparin, and based on the kinetics given as input, a classification model identifies whether it is an UFH or an LMWH; If the inhibitor is an UFH, and based on the kinetics given as input, a regression model calculates the UFH concentration; If the inhibitor is an LMWH, and based on the kinetics given as input,a regression model calculates the LMWH concentration; If the inhibitor is a DOAC, an experimental measurement is re-run using the optimized DOAC methodology and a new kinetic is recorded and then presented as input to a classification model that identifies whether the inhibitor is rivaroxaban, apixaban or edoxaban: If the inhibitor is rivaroxaban, and according to the shape of the kinetics (measured by the optimized DOAC methodology) presented to it as input, a regression model calculates the rivaroxaban concentration. If this concentration is less than 200 ng / mL,a second regression model recalculates the rivaroxaban concentration, this time using as input the kinetics measured by the universal methodology: the latter allows for more precise rendering of results for low rivaroxaban concentrations. Otherwise, the result is rendered directly. (this embodiment corresponds to the methodology described as the “improved methodology based on the universal methodology” in the present description). If the inhibitor is apixaban, and according to the shape of the kinetics (measured by the optimized AODs methodology) presented to it as input, a regression model calculates the apixaban concentration. If this concentration is less than 200 ng / mL,a second regression model recalculates the apixaban concentration, this time using as input the kinetics measured by the universal methodology: the latter allows for more precise rendering of results for low apixaban concentrations. Otherwise, the result is rendered directly. (this embodiment corresponds to the methodology described as the “improved methodology based on the universal methodology” in the present description). If the inhibitor is edoxaban, and according to the shape of the kinetics (measured by the optimized AODs methodology) presented to it as input, a regression model calculates the edoxaban concentration. If this concentration is less than 200 ng / mL,a second regression model recalculates the edoxaban concentration, this time using as input the kinetics measured by the universal methodology: the latter allows for more precise rendering of results for low edoxaban concentrations. Otherwise the result is rendered directly. (this embodiment corresponds to the methodology described as the “improved methodology based on the universal methodology” in this description).

[0246] The following sections list, respectively, for each of these machine learning models, the training and validation datasets used for their training, as well as the associated algorithms.

[0247] Note that for each data set, the concentrations of synthetic factor Xa inhibitors (UFH, LMWH, rivaroxaban, apixaban and edoxaban) were measured on a STA-R ®< automated system using the commercial STA ®< - Liquid Anti-Xa kit, as well as the associated commercial calibrators and methodologies. 3.1 Detection of the presence or absence of an anti-Xa anticoagulant 3.1.1 Datasets

[0248] The samples without anticoagulant were prepared for different tests. This is a plasma matrix diluted in the same proportions as the samples spiked on the day of the test: 3 kinetics measured on 24.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) + 3 kinetics measured on 24.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07). a normal pooled plasma sample batch 03.2017 spiked with 0.00 IU / mL of sodium HNF (Heparin Choay ®< ), prepared extemporaneously and tested simultaneously on both machines. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 18.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) a normal pooled plasma sample lot 03.2017 spiked with a theoretical rate of 0.00 IU / mL of sodium HNF (Heparin Choay ®< ), prepared extemporaneously. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 23.10.2019 on the STA-R ®< AUT05016 (soft version 3.04.07) + 3 kinetics measured on 23.10.2019 on the STA-R ®< AUT05450 (soft version 3.04.07) a normal pool sample of plasmas lot 03.2017 overloaded by 0.00 IU / mL of calcium HNF (Calciparine ®< ), prepared extemporaneously and tested simultaneously on the two machines. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 17.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) a normal pool sample of plasmas lot 03.2017 spiked with a theoretical rate of 0.00 IU / mL of calcium HNF (Calciparine ®< ), prepared extemporaneously. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 19.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) + 3 kinetics measured on 19.10.2017 on the STA-R ®< AUT05016 (soft version 3.04.07) a normal pooled plasma sample lot 03.2017 spiked with 0.00 IU / mL of LMWH dalteparin sodium (Fragmine ®< ), prepared extemporaneously and tested simultaneously on both machines. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 17.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) a normal pool sample of plasmas lot 03.2017 overloaded with a theoretical rate of 0.00 IU / mL of LMWH dalteparin sodium (Fragmine ®< ), prepared extemporaneously. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 26.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) + 3 kinetics measured on 26.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) a normal pooled plasma sample lot 03.2017 spiked with 0.00 IU / mL of LMWH enoxaparin sodium (Lovenox ®< ), prepared extemporaneously and tested simultaneously on both machines. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 19.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) a normal pooled plasma sample lot 03.2017 spiked with a theoretical rate of 0.00 IU / mL of LMWH enoxaparin sodium (Lovenox ®< ), prepared extemporaneously. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 25.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) + 3 kinetics measured on 25.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) a normal pool sample of plasmas batch 03.2017 spiked with 0.00 IU / mL of LMWH tinzaparin sodium (InnoHep ®< ), prepared extemporaneously and tested simultaneously on both machines. STA ®< - Liquid Anti-Xa batch 251187. 3 kinetics measured on 18.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) a normal pool sample of plasmas batch 03.2017 spiked with a theoretical level of 0.00 IU / mL of LMWH tinzaparin sodium (InnoHep ®< ), prepared extemporaneously. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 18.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) a normal pool sample of plasmas lot 03.2017 spiked with a theoretical rate of 0.00 IU / mL of Fondaparinux (Arixtra ®< ), prepared extemporaneously. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 10.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) + 3 kinetics measured on 17.10.2017 on the STA-R ®< AUT05450 (soft version 3.04.07) + 3 kinetics measured on 17.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) + 3 kinetics measured on 10.10.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) a normal pooled plasma sample lot 03.2017 spiked with 0 ng / mL of Xarelto ®< (rivaroxaban), prepared and stored at -80°C. STA ®< - Liquid Anti-Xa lot 251738. 3 kinetics measured on 05.02.2018 on the STA-R AUT06399 (soft version 3.04.07) a normal pooled plasma sample lot 03.2017 spiked with 0 ng / mL of Xarelto ®< (rivaroxaban), prepared and stored at -80°C. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 17.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) + 3 kinetics measured on 17.10.2017 on the STA-R ®< AUT05450 (soft version 3.04.07) + 3 kinetics measured on 21.09.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) + 3 kinetics measured on 20.09.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) a normal pooled plasma sample lot 03.2017 spiked with 0 ng / mL of Lixiana ®< (edoxaban), prepared and stored at -80°C. STA ®< - Liquid Anti-Xa lot 251738. 3 kinetics measured on 05.02.2018 on the STA-R AUT06399 (soft version 3.04.07) a normal pooled plasma sample lot 03.2017 spiked with 0 ng / mL of Lixiana ®< (edoxaban), prepared and stored at -80°C. STA ®< - Liquid Anti-Xa lot 251187. 3 kinetics measured on 16.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) + 3 kinetics measured on 05.10.2017 on the STA-R ®< AUT06360 (soft version 3.04.07) + 3 kinetics measured on 16.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) + 3 kinetics measured on 05.10.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) a normal pooled plasma sample lot 03.2017 spiked with 0 ng / mL of Eliquis ®< (apixaban), prepared and stored at -80°C. STA ®< - Liquid Anti-Xa lot 251738. 3 kinetics measured on 05.02.2018 on the STA-R AUT06399 (soft version 3.04.07) a normal pooled sample of plasma lot 03.2017 spiked with 0 ng / mL Eliquis ®< (apixaban), prepared and stored at -80°C. STA ®< - Liquid Anti-Xa lot 251187. .

[0249] Regarding sodium HNF (Heparin Choay ®< ), the data generated are: 60 kinetics measured on 24.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.12, 0.23, 0.32, 0.46, 0.54, 0.66, 0.66, 0.78, 0.86, 1.00, 1.18, 1.20, 1.36, 1.47, 1.56, 1.64, 1.74, 1.80, 1.88 and 2.03 IU / mL; each sample level is tested in n=3 ; the samples were prepared extemporaneously and tested simultaneously on both automatons. 60 kinetics measured on 24.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.11, 0.21, 0.32, 0.43, 0.55, 0.63, 0.64, 0.76, 0.87, 0.97, 1.14, 1.17, 1.30, 1.43, 1.52, 1.61, 1.70, 1.84, 1.84 and 1.99 IU / mL; each sample level is tested in n=3 ;The samples were prepared extemporaneously and tested simultaneously on the two machines.

[0250] Regarding calcium HNF (Calciparine ®< ), the data generated are: 60 kinetics measured on 23.10.2019 on the STA-R ®< AUT05016 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.10, 0.21, 0.32, 0.44, 0.56, 0.66, 0.77, 0.84, 0.86, 0.98, 1.15, 1.28, 1.34, 1.43, 1.56, 1.60, 1.75, 1.85, 1.92 and 1.96 IU / mL; each level of samples is tested in n=3 ; the samples were prepared extemporaneously and tested simultaneously on both automatons. 60 kinetics measured on 23.10.2019 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.12, 0.23, 0.33, 0.43, 0.55, 0.66, 0.75, 0.90, 0.88, 1.00, 1.13, 1.24, 1.32, 1.44, 1.53, 1.63, 1.70, 1.88, 1.96 and 2.01 IU / mL; each sample level is tested in n=3 ;The samples were prepared extemporaneously and tested simultaneously on the two machines.

[0251] Regarding the LMWH dalteparin sodium (Fragmine ®< ), the data generated are: 60 kinetics measured on 19.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.15, 0.25, 0.35, 0.44, 0.54, 0.64, 0.74, 0.85, 0.94, 1.07, 1.14, 1.24, 1.27, 1.44, 1.50, 1.55, 1.66, 1.75, 1.79, 2.26 IU Anti-Xa / mL; each sample level is tested in n= 3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 19.10.2017 on the STA-R ®< AUT05016 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.15, 0.25, 0.35, 0.44, 0.54, 0.64, 0.74, 0.85, 0.94, 1.07, 1.14, 1.24, 1.27, 1.44, 1.50, 1.55, 1.66, 1.75, 1.79, 2.26 IU Anti-Xa / mL; each sample level is tested in n=3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0252] Regarding the LMWH enoxaparin sodium (Lovenox ®< ), the data generated are: 60 kinetics measured on 26.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.10, 0.20, 0.31, 0.40, 0.52, 0.63, 0.71, 0.83, 0.93, 1.02, 1.09, 1.21, 1.28, 1.36, 1.49, 1.58, 1.65, 1.81, 1.85, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 26.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.10, 0.20, 0.31, 0.40, 0.52, 0.63, 0.71, 0.83, 0.93, 1.02, 1.09, 1.21, 1.28, 1.36, 1.49, 1.58, 1.65, 1.81, 1.85, 1.97 IU Anti-Xa / mL; each sample level is tested in n=3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0253] Regarding the LMWH tinzaparin sodium (InnoHep ®< ), the data generated are: 60 kinetics measured on 25.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.11, 0.20, 0.30, 0.40, 0.46, 0.54, 0.63, 0.80, 0.89, 0.96, 1.06, 1.21, 1.27, 1.37, 1.46, 1.52, 1.64, 1.81, 1.83, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 25.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.11, 0.20, 0.30, 0.40, 0.46, 0.54, 0.63, 0.80, 0.89, 0.96, 1.06, 1.21, 1.27, 1.37, 1.46, 1.52, 1.64, 1.81, 1.83, 1.97 IU Anti-Xa / mL; each sample level is tested in n=3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0254] Batch number of STA ®< - Liquid Anti-Xa: 251187, batch used for the generation of kinetics in universal methodology and for the commercial dosage of heparin overload.

[0255] Regarding Xarelto ®< (rivaroxaban), the data generated are: 60 kinetics measured on 10.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 10, 20, 35, 41, 49, 56, 66, 74, 84, 93, 105, 110, 116, 134, 143, 155, 160, 175, 181 and 198 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 60 kinetics measured on 17.10.2017 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 35, 41, 48, 59, 65, 77, 85, 95, 103, 113, 121, 134, 145, 160, 161, 172, 169, 202 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 60 kinetics measured on 17.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 25, 34, 43, 50, 62, 74, 84, 93, 104, 109, 118, 133, 143, 159, 158, 176, 173 and 203 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 60 kinetics measured on 10.10.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 30, 38, 46, 57, 64, 72, 82, 90, 101, 110, 115, 130, 140, 148, 152, 166, 164 and 192 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C.

[0256] Regarding Eliquis ®< (apixaban), the data generated are: 60 kinetics measured on 16.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 10, 20, 27, 39, 41, 52, 61, 67, 76, 93, 96, 105, 123, 127, 140, 145, 155, 167, 184 and 182 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 60 kinetics measured on 05.10.2017 on the STA-R ®< AUT06360 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 26, 36, 43, 53, 60, 70, 75, 93, 103, 109, 121, 135, 136, 153, 160, 175, 185 and 192 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 60 kinetics measured on 16.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 26, 34, 44, 50, 61, 70, 74, 94, 103, 109, 122, 135, 138, 147, 162, 173, 182 and 186 ng / mL; each sample level is tested in n=3. Samples were prepared and stored at -80°C. 60 kinetics measured on 05.10.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 33, 41, 46, 57, 65, 76, 79, 96, 105, 116, 125, 135, 138, 153, 161, 169, 180 and 186 ng / mL; each sample level is tested in n =3. Samples were prepared and stored at -80°C.

[0257] Regarding Lixiana ®< (edoxaban), the data generated are: 60 kinetics measured on 17.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 10, 20, 28, 40, 49, 64, 74, 95, 103, 109, 122, 132, 144, 131, 132, 157, 171, 195, 191 and 201 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 60 kinetics measured on 17.10.2017 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 21, 28, 41, 49, 62, 74, 92, 102, 112, 120, 133, 142, 129, 133, 155, 168, 188, 200 and 206 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 60 kinetics measured on 21.09.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 21, 27, 40, 49, 64, 73, 90, 99, 108, 122, 132, 144, 130, 136, 153, 169, 192, 194 and 216 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 60 kinetics measured on 20.09.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 21, 29, 39, 48, 60, 73, 81, 92, 107, 115, 129, 136, 129, 127, 149, 163, 181, 195 and 199 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C.

[0258] Batch number of STA ®< - Liquid Anti-Xa: 251738, batch used for the generation of kinetics in universal methodology and for the commercial dosage of AOD overloads. 3.1.2 Machine learning model Data organization

[0259] Data from samples without anticoagulant molecule were mixed and split into two datasets named here ABSENCE-0 and ABSENCE-1.

[0260] The machine learning model is trained by cross-validation divided into two subsets as follows:Subset 1: Training data: ABSENCE-0, sodium UFH data (Choay Heparin ®< ) generated on the STA-R ®< AUT06366, calcium UFH data (Calciparin ®< ) generated on the STA-R ®< AUT05450, dalteparin sodium LMWH data (Fragmine ®< ) generated on the STA-R ®< AUT06366, enoxaparin sodium LMWH data (Lovenox ®< ) generated on the STA-R ®< AUT06366, tinzaparin sodium LMWH data (InnoHep ®< ) generated on the STA-R ®< AUT06366, rivaroxaban data (Xarelto ®< ) generated on the STA-R ®< AUT06366, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05450, edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT06366, edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT05450, apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT06366, apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT06360.Validation data: ABSENCE-1, UFH sodium data (Heparin Choay ®< ) generated on the STA-R ®< AUT00603, UFH calcium data (Calciparin ®< ) generated on the STA-R ®< AUT05016, LMWH dalteparin sodium data (Fragmine ®< ) generated on the STA-R ®< AUT05016, LMWH enoxaparin sodium data (Lovenox ®< ) generated on the STA-R ®< AUT00603, LMWH tinzaparin sodium data (InnoHep ®< ) generated on the STA-R ®< AUT00603, rivaroxaban data (Xarelto ®< ) generated on the STA-R ®< AUT00603, rivaroxaban data (Xarelto ®< ) generated on STA-R ®< AUT06399, edoxaban data (Lixiana ®< ) generated on STA-R ®< AUT00603, edoxaban data (Lixiana ®< ) generated on STA-R ®< AUT06399, apixaban data (Eliquis ®< ) generated on STA-R ®< AUT00603, apixaban data (Eliquis ®< ) generated on STA-R ®< AUT06399.Subset 2: Training data: ABSENCE-1, sodium UFH data (Choay Heparin ®< ) generated on the STA-R ®< AUT00603, calcium UFH data (Calciparin ®< ) generated on the STA-R ®< AUT05016, dalteparin sodium LMWH data (Fragmine ®< ) generated on the STA-R ®< AUT05016, enoxaparin sodium LMWH data (Lovenox ®< ) generated on the STA-R ®< AUT00603, tinzaparin sodium LMWH data (InnoHep ®< ) generated on the STA-R ®< AUT00603, rivaroxaban data (Xarelto ®< ) generated on the STA-R ®< AUT00603, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06399, edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT00603, edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT06399, apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT00603, apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT06399.Validation data: ABSENCE-0, UFH sodium data (Heparin Choay ®< ) generated on STA-R ®< AUT06366, UFH calcium data (Calciparin ®< ) generated on STA-R ®< AUT05450, LMWH dalteparin sodium data (Fragmine ®< ) generated on STA-R ®< AUT06366, LMWH enoxaparin sodium data (Lovenox ®< ) generated on STA-R ®< AUT06366, LMWH tinzaparin sodium data (InnoHep ®< ) generated on STA-R ®< AUT06366, rivaroxaban data (Xarelto ®< ) generated on STA-R ®< AUT06366, rivaroxaban data (Xarelto ®< ) generated on STA-R ®< AUT05450, edoxaban data (Lixiana ®< ) generated on STA-R ®< AUT06366, edoxaban data (Lixiana ®< ) generated on STA-R ®< AUT05450, apixaban data (Eliquis ®< ) generated on STA-R ®< AUT06366, apixaban data (Eliquis ®< ) generated on STA-R ®< AUT06360. .

[0261] The final learning is done on the entire data set. Machine learning model description

[0262] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 77 neurons Activation functions: Identity Hidden layer 40 neurons Activation functions: ReLU Output layer 2 neurons Activation function: Softmax Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.00001 Learning strategy: seed search Cost function: cross entropy 3.2 Identification of the anti-Xa anticoagulant category 3.2.1 Datasets

[0263] Regarding sodium HNF (Heparin Choay ®< ), the data generated are: 60 kinetics measured on 24.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.12, 0.23, 0.32, 0.46, 0.54, 0.66, 0.66, 0.78, 0.86, 1.00, 1.18, 1.20, 1.36, 1.47, 1.56, 1.64, 1.74, 1.80, 1.88 and 2.03 IU / mL; each sample level is tested in n=3 ; the samples were prepared extemporaneously and tested simultaneously on both automatons. 60 kinetics measured on 24.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.11, 0.21, 0.32, 0.43, 0.55, 0.63, 0.64, 0.76, 0.87, 0.97, 1.14, 1.17, 1.30, 1.43, 1.52, 1.61, 1.70, 1.84, 1.84 and 1.99 IU / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0264] Regarding calcium HNF (Calciparine ®< ), the data generated are: 60 kinetics measured on 23.10.2019 on the STA-R ®< AUT05016 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.10, 0.21, 0.32, 0.44, 0.56, 0.66, 0.77, 0.84, 0.86, 0.98, 1.15, 1.28, 1.34, 1.43, 1.56, 1.60, 1.75, 1.85, 1.92 and 1.96 IU / mL; each level of samples is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 23.10.2019 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.12, 0.23, 0.33, 0.43, 0.55, 0.66, 0.75, 0.90, 0.88, 1.00, 1.13, 1.24, 1.32, 1.44, 1.53, 1.63, 1.70, 1.88, 1.96 and 2.01 IU / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0265] Regarding the LMWH dalteparin sodium (Fragmine ®< ), the data generated are: 60 kinetics measured on 19.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.15, 0.25, 0.35, 0.44, 0.54, 0.64, 0.74, 0.85, 0.94, 1.07, 1.14, 1.24, 1.27, 1.44, 1.50, 1.55, 1.66, 1.75, 1.79, 2.26 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 19.10.2017 on the STA-R ®< AUT05016 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.15, 0.25, 0.35, 0.44, 0.54, 0.64, 0.74, 0.85, 0.94, 1.07, 1.14, 1.24, 1.27, 1.44, 1.50, 1.55, 1.66, 1.75, 1.79, 2.26 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0266] Regarding the LMWH enoxaparin sodium (Lovenox ®< ), the data generated are: 60 kinetics measured on 26.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.10, 0.20, 0.31, 0.40, 0.52, 0.63, 0.71, 0.83, 0.93, 1.02, 1.09, 1.21, 1.28, 1.36, 1.49, 1.58, 1.65, 1.81, 1.85, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 26.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.10, 0.20, 0.31, 0.40, 0.52, 0.63, 0.71, 0.83, 0.93, 1.02, 1.09, 1.21, 1.28, 1.36, 1.49, 1.58, 1.65, 1.81, 1.85, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0267] Regarding the LMWH tinzaparin sodium (InnoHep ®< ), the data generated are: 60 kinetics measured on 25.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.11, 0.20, 0.30, 0.40, 0.46, 0.54, 0.63, 0.80, 0.89, 0.96, 1.06, 1.21, 1.27, 1.37, 1.46, 1.52, 1.64, 1.81, 1.83, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 25.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas batch 03.2017 overloaded with 0.11, 0.20, 0.30, 0.40, 0.46, 0.54, 0.63, 0.80, 0.89, 0.96, 1.06, 1.21, 1.27, 1.37, 1.46, 1.52, 1.64, 1.81, 1.83, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0268] Batch number of STA ®< - Liquid Anti-Xa: 251187, batch used for the generation of kinetics in universal methodology and for the commercial dosage of heparin overload.

[0269] Regarding Xarelto ®< (rivaroxaban), the data generated are: 60 kinetics measured on 10.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 10, 20, 35, 41, 49, 56, 66, 74, 84, 93, 105, 110, 116, 134, 143, 155, 160, 175, 181 and 198 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 60 kinetics measured on 17.10.2017 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 35, 41, 48, 59, 65, 77, 85, 95, 103, 113, 121, 134, 145, 160, 161, 172, 169, 202 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 60 kinetics measured on 17.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 25, 34, 43, 50, 62, 74, 84, 93, 104, 109, 118, 133, 143, 159, 158, 176, 173 and 203 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 60 kinetics measured on 10.10.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 30, 38, 46, 57, 64, 72, 82, 90, 101, 110, 115, 130, 140, 148, 152, 166, 164 and 192 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C.

[0270] Regarding Eliquis ®< (apixaban), the data generated are: 60 kinetics measured on 16.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 10, 20, 27, 39, 41, 52, 61, 67, 76, 93, 96, 105, 123, 127, 140, 145, 155, 167, 184 and 182 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 60 kinetics measured on 05.10.2017 on the STA-R ®< AUT06360 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 26, 36, 43, 53, 60, 70, 75, 93, 103, 109, 121, 135, 136, 153, 160, 175, 185 and 192 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 60 kinetics measured on 16.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 26, 34, 44, 50, 61, 70, 74, 94, 103, 109, 122, 135, 138, 147, 162, 173, 182 and 186 ng / mL; each sample level is tested in n=3. Samples were prepared and stored at -80°C. 60 kinetics measured on 05.10.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 20, 33, 41, 46, 57, 65, 76, 79, 96, 105, 116, 125, 135, 138, 153, 161, 169, 180 and 186 ng / mL; each sample level is tested in n =3. Samples were prepared and stored at -80°C.

[0271] Regarding Lixiana ®< (edoxaban), the data generated are: 60 kinetics measured on 17.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 10, 20, 28, 40, 49, 64, 74, 95, 103, 109, 122, 132, 144, 131, 132, 157, 171, 195, 191 and 201 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 60 kinetics measured on 17.10.2017 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 21, 28, 41, 49, 62, 74, 92, 102, 112, 120, 133, 142, 129, 133, 155, 168, 188, 200 and 206 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 60 kinetics measured on 21.09.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 21, 27, 40, 49, 64, 73, 90, 99, 108, 122, 132, 144, 130, 136, 153, 169, 192, 194 and 216 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 60 kinetics measured on 20.09.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 10, 21, 29, 39, 48, 60, 73, 81, 92, 107, 115, 129, 136, 129, 127, 149, 163, 181, 195 and 199 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C.

[0272] Batch number of STA ®< - Liquid Anti-Xa: 251738, batch used for the generation of kinetics in universal methodology and for the commercial dosage of AOD overloads. 3.2.2 Machine learning model Data organization

[0273] The machine learning model is trained by cross-validation divided into two subsets as follows: Subset 1: Training data: sodium UFH data (Choay Heparin ®< ) generated on the STA-R ®< AUT06366, calcium UFH data (Calciparin ®< ) generated on the STA-R ®< AUT05450, dalteparin sodium LMWH data (Fragmine ®< ) generated on the STA-R ®< AUT06366, enoxaparin sodium LMWH data (Lovenox ®< ) generated on the STA-R ®< AUT06366, tinzaparin sodium LMWH data (InnoHep ®< ) generated on the STA-R ®< AUT06366, rivaroxaban data (Xarelto ®< ) generated on the STA-R ®< AUT06366, data rivaroxaban (Xarelto ®< ) data generated on STA-R ®< AUT05450, edoxaban (Lixiana ®< ) data generated on STA-R ®< AUT06366, edoxaban (Lixiana ®< ) data generated on STA-R ®< AUT05450, apixaban (Eliquis ®< ) data generated on STA-R ®< AUT06366, apixaban (Eliquis ®< ) data generated on STA-R ®< AUT06360.Validation data: UFH sodium data (Heparin Choay ®< ) generated on the STA-R ®< AUT00603, UFH calcium data (Calciparin ®< ) generated on the STA-R ®< AUT05016, LMWH dalteparin sodium data (Fragmine ®< ) generated on the STA-R ®< AUT05016, LMWH enoxaparin sodium data (Lovenox ®< ) generated on the STA-R ®< AUT00603, LMWH tinzaparin sodium data (InnoHep ®< ) generated on the STA-R ®< AUT00603, rivaroxaban data (Xarelto ®< ) generated on the STA-R ®< AUT00603, rivaroxaban data (Xarelto ®< ) generated on the STA-R ®< AUT06399, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00603, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06399, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT00603, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06399.Subset 2: Training data: sodium UFH data (Choay Heparin ®< ) generated on the STA-R ®< AUT00603, calcium UFH data (Calciparin ®< ) generated on the STA-R ®< AUT05016, dalteparin sodium LMWH data (Fragmine ®< ) generated on the STA-R ®< AUT05016, enoxaparin sodium LMWH data (Lovenox ®< ) generated on the STA-R ®< AUT00603, tinzaparin sodium LMWH data (InnoHep ®< ) generated on the STA-R ®< AUT00603, rivaroxaban data (Xarelto ®< ) generated on the STA-R ®< AUT00603, data rivaroxaban (Xarelto ®< ) generated on STA-R ®< AUT06399, edoxaban (Lixiana ®< ) data generated on STA-R ®< AUT00603, edoxaban (Lixiana ®< ) data generated on STA-R ®< AUT06399, apixaban (Eliquis ®< ) data generated on STA-R ®< AUT00603, apixaban (Eliquis ®< ) data generated on STA-R ®< AUT06399.Validation data: UFH sodium data (Heparin Choay ®< ) generated on the STA-R ®< AUT06366, UFH calcium data (Calciparin ®< ) generated on the STA-R ®< AUT05450, LMWH dalteparin sodium data (Fragmine ®< ) generated on the STA-R ®< AUT06366, LMWH enoxaparin sodium data (Lovenox ®< ) generated on the STA-R ®< AUT06366, LMWH tinzaparin sodium data (InnoHep ®< ) generated on the STA-R ®< AUT06366, rivaroxaban data (Xarelto ®< ) generated on the STA-R ®< AUT06366, rivaroxaban data (Xarelto ®< ) generated on the STA-R ®< AUT05450, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05450, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06360. .

[0274] The final learning is done on the entire data set. Machine learning model description

[0275] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 77 neurons Activation functions: Identity Hidden layer 18 neurons Activation functions: ReLU Output layer 2 neurons Activation function: Softmax Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.001 Learning strategy: seed search Cost function: cross entropy 3.3 Identification of heparins 3.3.1 Datasets

[0276] Regarding sodium HNF (Heparin Choay ®< ), the data generated are: 60 kinetics measured on 24.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.12, 0.23, 0.32, 0.46, 0.54, 0.66, 0.66, 0.78, 0.86, 1.00, 1.18, 1.20, 1.36, 1.47, 1.56, 1.64, 1.74, 1.80, 1.88 and 2.03 IU / mL; each sample level is tested in n=3 ; the samples were prepared extemporaneously and tested simultaneously on both automatons. 60 kinetics measured on 24.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.11, 0.21, 0.32, 0.43, 0.55, 0.63, 0.64, 0.76, 0.87, 0.97, 1.14, 1.17, 1.30, 1.43, 1.52, 1.61, 1.70, 1.84, 1.84 and 1.99 IU / mL; each sample level is tested in n=3 ;the samples were prepared extemporaneously and tested simultaneously on both machines. 24 kinetics measured on 18.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with a theoretical rate of 0.25, 0.50, 0.75, 1.00, 1.25, 1.50, 1.75 and 2.00 IU / mL; each sample level is tested in n=3 ; the samples were prepared extemporaneously.

[0277] Regarding calcium HNF (Calciparine ®< ), the data generated are: 60 kinetics measured on 23.10.2019 on the STA-R ®< AUT05016 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.10, 0.21, 0.32, 0.44, 0.56, 0.66, 0.77, 0.84, 0.86, 0.98, 1.15, 1.28, 1.34, 1.43, 1.56, 1.60, 1.75, 1.85, 1.92 and 1.96 IU / mL; each level of samples is tested in n=3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 23.10.2019 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.12, 0.23, 0.33, 0.43, 0.55, 0.66, 0.75, 0.90, 0.88, 1.00, 1.13, 1.24, 1.32, 1.44, 1.53, 1.63, 1.70, 1.88, 1.96 and 2.01 IU / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 24 kinetics measured on 17.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with a theoretical rate of 0.25, 0.50, 0.75, 1.00, 1.25, 1.50, 1.75 and 2.00 IU / mL; each sample level is tested in n =3 ; the samples were prepared extemporaneously.

[0278] Regarding the LMWH dalteparin sodium (Fragmine ®< ), the data generated are: 60 kinetics measured on 19.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.15, 0.25, 0.35, 0.44, 0.54, 0.64, 0.74, 0.85, 0.94, 1.07, 1.14, 1.24, 1.27, 1.44, 1.50, 1.55, 1.66, 1.75, 1.79, 2.26 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 19.10.2017 on the STA-R ®< AUT05016 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.15, 0.25, 0.35, 0.44, 0.54, 0.64, 0.74, 0.85, 0.94, 1.07, 1.14, 1.24, 1.27, 1.44, 1.50, 1.55, 1.66, 1.75, 1.79, 2.26 IU Anti-Xa / mL; each sample level is tested in n=3; the samples were prepared extemporaneously and tested simultaneously on both machines. 24 kinetics measured on 17.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas batch 03.2017 overloaded with a theoretical rate of 0.25, 0.50, 0.75, 1.00, 1.25, 1.50, 1.75 and 2.00 IU / mL; each sample level is tested in n =3; the samples were prepared extemporaneously.

[0279] Regarding the LMWH enoxaparin sodium (Lovenox ®< ), the data generated are: 60 kinetics measured on 26.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.10, 0.20, 0.31, 0.40, 0.52, 0.63, 0.71, 0.83, 0.93, 1.02, 1.09, 1.21, 1.28, 1.36, 1.49, 1.58, 1.65, 1.81, 1.85, 1.97 IU Anti-Xa / mL; each sample level is tested in n=3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 26.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.10, 0.20, 0.31, 0.40, 0.52, 0.63, 0.71, 0.83, 0.93, 1.02, 1.09, 1.21, 1.28, 1.36, 1.49, 1.58, 1.65, 1.81, 1.85, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 24 kinetics measured on 19.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with a theoretical rate of 0.25, 0.50, 0.75, 1.00, 1.25, 1.50, 1.75 and 2.00 IU / mL; each sample level is tested in n =3; the samples were prepared extemporaneously.

[0280] Regarding the LMWH tinzaparin sodium (InnoHep ®< ), the data generated are: 60 kinetics measured on 25.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.11, 0.20, 0.30, 0.40, 0.46, 0.54, 0.63, 0.80, 0.89, 0.96, 1.06, 1.21, 1.27, 1.37, 1.46, 1.52, 1.64, 1.81, 1.83, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 60 kinetics measured on 25.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.11, 0.20, 0.30, 0.40, 0.46, 0.54, 0.63, 0.80, 0.89, 0.96, 1.06, 1.21, 1.27, 1.37, 1.46, 1.52, 1.64, 1.81, 1.83, 1.97 IU Anti-Xa / mL; each sample level is tested in n=3; the samples were prepared extemporaneously and tested simultaneously on both machines. 24 kinetics measured on 18.07.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with a theoretical rate of 0.25, 0.50, 0.75, 1.00, 1.25, 1.50, 1.75 and 2.00 IU / mL; each sample level is tested in n =3; the samples were prepared extemporaneously.

[0281] Batch number of STA ®< - Liquid Anti-Xa: 251187, batch used for the generation of kinetics in universal methodology and for the commercial dosage of heparin overload (when applicable). 3.3.2 Machine learning model Data organization

[0282] The machine learning model is trained through cross-validation organized as follows: Training data: UFH sodium data (Heparin Choay ®< ) generated on the STA-R ®< AUT06366, UFH sodium data (Heparin Choay ®< ) generated on the STA-R ®< AUT00603, UFH calcium data (Calciparin ®< ) generated on the STA-R ®< AUT05450, UFH calcium data (Calciparin ®< ) generated on the STA-R ®< AUT05016, LMWH dalteparin sodium data (Fragmine ®< ) generated on the STA-R ®< AUT 06366, LMWH dalteparin sodium data (Fragmine ®< ) generated on the STA-R ®< AUT 05016, LMWH enoxaparin sodium data (Lovenox ®< ) generated on STA-R ®< AUT 06366, LMWH enoxaparin sodium (Lovenox ®< ) data generated on STA-R ®< AUT 00603, LMWH tinzaparin sodium (InnoHep ®< ) data generated on STA-R ®< AUT 06366, LMWH tinzaparin sodium (InnoHep ®< ) data generated on STA-R ®< AUT 00603.Validation data: Sodium UFH data (Choay Heparin ®< ) generated on STA-R ®< AUT06399, Calcium UFH data (Calciparin ®< ) generated on STA-R ®< AUT06399, Dalteparin sodium LMWH data (Fragmine ®< ) generated on STA-R ®< AUT 06399, Enoxaparin sodium LMWH data (Lovenox ®< ) generated on STA-R ®< AUT 06399, Tinzaparin sodium LMWH data (InnoHep ®< ) generated on STA-R ®< AUT 06399. Machine learning model description

[0283] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 77 neurons Activation functions: Identity Hidden layer 40 neurons Activation functions: ReLU Output layer 2 neurons Activation function: Softmax Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.9 Learning strategy: seed search Cost function: cross entropy 3.4 Assays of unfractionated heparins 3.4.1 Datasets

[0284] Regarding sodium HNF (Heparin Choay ®< ), the data generated are: 63 kinetics measured on 24.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.00, 0.12, 0.23, 0.32, 0.46, 0.54, 0.66, 0.66, 0.78, 0.86, 1.00, 1.18, 1.20, 1.36, 1.47, 1.56, 1.64, 1.74, 1.80, 1.88 and 2.03 IU / mL; each sample level is tested in n=3 ; the samples were prepared extemporaneously and tested simultaneously on both automatons. 63 kinetics measured on 24.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.00, 0.11, 0.21, 0.32, 0.43, 0.55, 0.63, 0.64, 0.76, 0.87, 0.97, 1.14, 1.17, 1.30, 1.43, 1.52, 1.61, 1.70, 1.84, 1.84 and 1.99 IU / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0285] Regarding calcium HNF (Calciparine ®< ), the data generated are: 63 kinetics measured on 23.10.2019 on the STA-R ®< AUT05016 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded by 0.00, 0.10, 0.21, 0.32, 0.44, 0.56, 0.66, 0.77, 0.84, 0.86, 0.98, 1.15, 1.28, 1.34, 1.43, 1.56, 1.60, 1.75, 1.85, 1.92 and 1.96 IU / mL; each level of samples is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 63 kinetics measured on 23.10.2019 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas batch 03.2017 overloaded with 0.00, 0.12, 0.23, 0.33, 0.43, 0.55, 0.66, 0.75, 0.90, 0.88, 1.00, 1.13, 1.24, 1.32, 1.44, 1.53, 1.63, 1.70, 1.88, 1.96 and 2.01 IU / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0286] Batch number of STA ®< - Liquid Anti-Xa: 251187, batch used for the generation of kinetics in universal methodology and for the commercial dosage of heparin overload. 3.4.2 Machine learning model Data organization

[0287] The machine learning model is trained by cross-validation divided into four subsets as follows: Subset 1: Training data: Sodium UFH data (Choay Heparin ®< ) generated on the STA-R ®< AUT06366, Calcium UFH data (Calciparin ®< ) generated on the STA-R ®< AUT05450. Validation data: Sodium UFH data (Choay Heparin ®< ) generated on the STA-R ®< AUT00603, Calcium UFH data (Calciparin ®< ) generated on the STA-R ®< AUT05016. Subset 2: Training data: Sodium UFH data (Choay Heparin ®< ) generated on the STA-R ®< AUT06366, Calcium UFH data (Calciparin ®< ) generated on the STA-R ®< AUT05016. Validation data: Sodium HNF data (Choay Heparin ®< ) generated on the STA-R ®< AUT00603, Calcium HNF data (Calciparin ®< ) generated on the STA-R ®< AUT05450. Subset 3: Training data: Sodium HNF data (Choay Heparin ®< ) generated on the STA-R ®< AUT00603, Calcium HNF data (Calciparin ®< ) generated on the STA-R ®< AUT05450.Validation data: Sodium HNF data (Choay Heparin ®< ) generated on the STA-R ®< AUT06366, Calcium HNF data (Calciparin ®< ) generated on the STA-R ®< AUT05016. Subset 4: Training data: Sodium HNF data (Choay Heparin ®< ) generated on the STA-R ®< AUT00603, Calcium HNF data (Calciparin ®< ) generated on the STA-R ®< AUT05016. Validation data: Sodium HNF data (Choay Heparin ®< ) generated on the STA-R ®< AUT06366, Calcium HNF data (Calciparin ®< ) generated on the STA-R ®< AUT05450.

[0288] The final learning is done on the entire data set. Machine learning model description

[0289] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 77 neurons Activation functions: Identity Hidden layer 40 neurons Activation functions: ReLU Output layer 1 neuron Activation function: Identity Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.01 Learning strategy: seed search Cost function: mean square error 3.5 Assays of low molecular weight heparins 3.5.1 Datasets

[0290] Regarding the LMWH dalteparin sodium (Fragmine ®< ), the data generated are: 63 kinetics measured on 19.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.00, 0.15, 0.25, 0.35, 0.44, 0.54, 0.64, 0.74, 0.85, 0.94, 1.07, 1.14, 1.24, 1.27, 1.44, 1.50, 1.55, 1.66, 1.75, 1.79, 2.26 IU Anti-Xa / mL; each sample level is tested in n=3 ; the samples were prepared extemporaneously and tested simultaneously on both machines. 63 kinetics measured on 19.10.2017 on the STA-R ®< AUT05016 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.00, 0.15, 0.25, 0.35, 0.44, 0.54, 0.64, 0.74, 0.85, 0.94, 1.07, 1.14, 1.24, 1.27, 1.44, 1.50, 1.55, 1.66, 1.75, 1.79, 2.26 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0291] Regarding the LMWH enoxaparin sodium (Lovenox ®< ), the data generated are: 63 kinetics measured on 26.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.00, 0.10, 0.20, 0.31, 0.40, 0.52, 0.63, 0.71, 0.83, 0.93, 1.02, 1.09, 1.21, 1.28, 1.36, 1.49, 1.58, 1.65, 1.81, 1.85, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 63 kinetics measured on 26.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.00, 0.10, 0.20, 0.31, 0.40, 0.52, 0.63, 0.71, 0.83, 0.93, 1.02, 1.09, 1.21, 1.28, 1.36, 1.49, 1.58, 1.65, 1.81, 1.85, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0292] Regarding the LMWH tinzaparin sodium (InnoHep ®< ), the data generated are: 63 kinetics measured on 25.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0.00, 0.11, 0.20, 0.30, 0.40, 0.46, 0.54, 0.63, 0.80, 0.89, 0.96, 1.06, 1.21, 1.27, 1.37, 1.46, 1.52, 1.64, 1.81, 1.83, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on both machines. 63 kinetics measured on 25.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 overloaded with 0.00, 0.11, 0.20, 0.30, 0.40, 0.46, 0.54, 0.63, 0.80, 0.89, 0.96, 1.06, 1.21, 1.27, 1.37, 1.46, 1.52, 1.64, 1.81, 1.83, 1.97 IU Anti-Xa / mL; each sample level is tested in n =3; the samples were prepared extemporaneously and tested simultaneously on the two automatons.

[0293] Batch number of STA ®< - Liquid Anti-Xa: 251187, batch used for the generation of kinetics in universal methodology and for the commercial dosage of heparin overload. 3.5.2 Machine learning model Data organization

[0294] The machine learning model is trained by cross-validation divided into two subsets as follows: Subset 1: Training data: LMWH dalteparin sodium data (Fragmine ®< ) generated on the STA-R ®< AUT05016, LMWH enoxaparin sodium data (Lovenox ®< ) generated on the STA-R ®< AUT00603, LMWH tinzaparin sodium data (InnoHep ®< ) generated on the STA-R ®< AUT00603. Validation data: LMWH dalteparin sodium (Fragmine ®< ) data generated on STA-R ®< AUT06366, LMWH enoxaparin sodium (Lovenox ®< ) data generated on STA-R ®< AUT06366, LMWH tinzaparin sodium (InnoHep ®< ) data generated on STA-R ®< AUT06366. Subset 2: Training data: LMWH dalteparin sodium data (Fragmine ®< ) generated on the STA-R ®< AUT06366, LMWH enoxaparin sodium data (Lovenox ®< ) generated on the STA-R ®< AUT06366, LMWH tinzaparin sodium data (InnoHep ®< ) generated on the STA-R ®< AUT06366.Validation data: LMWH dalteparin sodium (Fragmine ®< ) data generated on the STA-R ®< AUT05016 LMWH enoxaparin sodium (Lovenox ®< ) data generated on the STA-R ®< AUT00603, LMWH tinzaparin sodium (InnoHep ®< ) data generated on the STA-R ®< AUT00603. .

[0295] The final learning is done on the entire data set. Machine learning model description

[0296] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 77 neurons Activation functions: Identity Hidden layer 40 neurons Activation functions: ReLU Output layer 1 neuron Activation function: Identity Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.1 Learning strategy: seed search Cost function: mean square error 3.6 Identification of anti-Xa AOD (optimized AODs methodology) 3.6.1 Datasets

[0297] The data generated for Xarelto ® (rivaroxaban) are: 60 kinetics measured in January 2016 on the STA-R ®< AUT00460 normal pool of plasmas spiked with 10, 20, 29, 37, 44, 54, 62, 89, 115, 143, 160, 192, 226, 252, 287, 319, 331, 361, 391 and 407 ng / mL; each sample level is tested inn=3 ; samples were prepared and stored at -80°C. 60 kinetics measured in June 2016 on the STA-R ®< AUT00722 normal plasma pool spiked with 10, 20, 29, 37, 44, 54, 62, 89, 115, 143, 160, 192, 226, 252, 287, 319, 331, 361, 391 and 407 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C.

[0298] The Eliquis ® (apixaban) data generated are: 69 kinetics measured in January 2016 on the STA-R ®< AUT00460 normal pool of plasmas spiked with 10, 20, 31, 42, 48, 72, 92, 116, 143, 163, 192, 217, 247, 276, 298, 321, 348, 370, 397, 416, 439, 459 and 473 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 69 kinetics measured in June 2016 on the STA-R ®< AUT00722 normal plasma pool spiked with 10, 20, 31, 42, 48, 72, 92, 116, 143, 163, 192, 217, 247, 276, 298, 321, 348, 370, 397, 416, 439, 459 and 473 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C.

[0299] The data generated for Lixiana ®< (edoxaban) are: 69 kinetics measured in January 2016 on the STA-R ®< AUT00460 normal pool of plasmas spiked with 16, 21, 30, 39, 51, 76, 100, 128, 142, 155, 187, 215, 245, 269, 284, 322, 347, 359, 380, 393, 410, 426 and 436 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 69 kinetics measured in June 2016 on the STA-R ®< AUT00722 normal plasma pool spiked with 16, 21, 30, 39, 51, 76, 100, 128, 142, 155, 187, 215, 245, 269, 284, 322, 347, 359, 380, 393, 410, 426 and 436 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 3.6.2 Machine learning model Data organization

[0300] The machine learning model is trained through cross-validation organized as follows: Training data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00460, apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT00460, edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT00460. Validation data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00722, apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT00722, edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT00722. Machine learning model description

[0301] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 42 neurons Activation functions: Identity Hidden layer 29 neurons Activation functions: ReLU Hidden layer 16 neurons Activation functions: ReLU Output layer 3 neurons Activation function: Softmax Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.01 Learning strategy: seed search Cost function: cross entropy 3.7 Rivaroxaban dosage (optimized AODs methodology) 3.7.1 Datasets

[0302] The data generated for Xarelto ® (rivaroxaban) are: 63 kinetics measured in January 2016 on the STA-R ®< AUT00460 normal pool of plasmas spiked with 0, 10, 20, 29, 37, 44, 54, 62, 89, 115, 143, 160, 192, 226, 252, 287, 319, 331, 361, 391 and 407 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 63 kinetics measured in June 2016 on the STA-R ®< AUT00722 normal plasma pool spiked with 0, 10, 20, 29, 37, 44, 54, 62, 89, 115, 143, 160, 192, 226, 252, 287, 319, 331, 361, 391 and 407 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 3.7.2 Machine learning model Data organization

[0303] The machine learning model is trained through cross-validation organized as follows: Training data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00460. Validation data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00722. Machine learning model description

[0304] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 42 neurons Activation functions: Identity Hidden layer 29 neurons Activation functions: ReLU Hidden layer 16 neurons Activation functions: ReLU Output layer 1 neuron Activation function: Identity Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.01 Learning strategy: seed search Cost function: mean square error 3.8 Rivaroxaban dosages (universal methodology or "improved methodology based on the universal methodology") 3.8.1 Datasets

[0305] Data for Xarelto ® (rivaroxaban) were generated in two separate studies. For the first study, the data generated are: 63 kinetics measured on 10.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 20, 35, 41, 49, 56, 66, 74, 84, 93, 105, 110, 116, 134, 143, 155, 160, 175, 181 and 198 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 63 kinetics measured on 17.10.2017 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 20, 35, 41, 48, 59, 65, 77, 85, 95, 103, 113, 121, 134, 145, 160, 161, 172, 169, 202 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 63 kinetics measured on 17.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 20, 25, 34, 43, 50, 62, 74, 84, 93, 104, 109, 118, 133, 143, 159, 158, 176, 173 and 203 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 63 kinetics measured on 10.10.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 20, 30, 38, 46, 57, 64, 72, 82, 90, 101, 110, 115, 130, 140, 148, 152, 166, 164 and 192 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C.

[0306] For this first study, STA ®< - Liquid Anti-Xa lot 251738 was used for the generation of kinetics in universal methodology and for the commercial dosage of rivaroxaban overload. Regarding the second study, the data generated are: 63 kinetics measured on 05 / 14 / 2019 on the STA-R ®< AUT05676 (soft version 3.04.07) normal pool of plasmas lot 19059RD (code 22824) overloaded with 0, 11, 21, 28, 42, 50, 58, 68, 77, 93, 102, 113, 121, 129, 141, 153, 160, 174, 184, 201 and 204 ng / mL; each level of samples is tested inn =3; samples were prepared and stored at -80°C, then tested simultaneously on the three automatons. 63 kinetics measured on 14.05.2019 on the STA-R ®< AUT05980 (soft version 3.04.07) normal pool of plasmas lot 19059RD (code 22824) overloaded with 0, 15, 23, 32, 47, 52, 61, 74, 77, 92, 108, 112, 120, 129, 139, 151, 164, 174, 184, 191 and 206 ng / mL; each level of samples is tested in n =3; samples were prepared and stored at -80°C, then tested simultaneously on the three automatons. 63 kinetics measured on 14.05.2019 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 19059RD (code 22824) overloaded with 0, 13, 22, 29, 44, 51, 60, 72, 79, 96, 104, 118, 122, 132, 143, 155, 161, 175, 186, 198 and 201 ng / mL; each level of samples is tested in n =3; the samples were prepared and stored at -80°C, then tested simultaneously on the three machines.

[0307] For this second study, STA ®< - Liquid Anti-Xa lot 253225 was used for the generation of kinetics in universal methodology and for the commercial dosage of rivaroxaban overload. 3.8.2 Machine learning model Data organization

[0308] The machine learning model is trained by cross-validation divided into seven subsets as follows: Subset 1: Training data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00603, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05450, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05676, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05980, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366 (bis). Validation data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06399.Subset 2: Training data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00603, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05450, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05676, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05980, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366 (bis), rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06399. Validation data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366.Subset 3: Training data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00603, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05450, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05676, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05980, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06399. Validation data: rivaroxaban (Xarelto ®< ) data generated on STA-R ®< AUT06366 (bis).Subset 4: Training data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00603, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05450, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05676, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366 (bis), rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06399. Validation data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05980.Subset 5: Training data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00603, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05450, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05980, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366 (bis), rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06399. Validation data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05676.Subset 6: Training data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00603, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05676, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05980, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366 (bis), rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06399. Validation data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05450.Subset 7: Training data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05450, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05676, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT05980, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366, rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06366 (bis), rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT06399. Validation data: rivaroxaban (Xarelto ®< ) data generated on the STA-R ®< AUT00603. .

[0309] The final learning is done on the entire data set. Machine learning model description

[0310] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 77 neurons Activation functions: Identity Hidden layer 40 neurons Activation functions: ReLU Output layer 1 neuron Activation function: Identity Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.001 Learning strategy: seed search Cost function: mean square error 3.9 Apixaban dosage (optimized AODs methodology) 3.9.1 Datasets

[0311] The Eliquis ® (apixaban) data generated are: 72 kinetics measured in January 2016 on the STA-R ®< AUT00460 normal pool of plasmas spiked with 0, 10, 20, 31, 42, 48, 72, 92, 116, 143, 163, 192, 217, 247, 276, 298, 321, 348, 370, 397, 416, 439, 459 and 473 ng / mL; each sample level is tested in n=3 ; samples were prepared and stored at -80°C. 72 kinetics measured in June 2016 on the STA-R ®< AUT00722 normal plasma pool spiked with 0, 10, 20, 31, 42, 48, 72, 92, 116, 143, 163, 192, 217, 247, 276, 298, 321, 348, 370, 397, 416, 439, 459 and 473 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 3.9.2 Machine learning model Data organization

[0312] The machine learning model is trained through cross-validation organized as follows: Training data: apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT00460. Validation data: apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT00722. Machine learning model description

[0313] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 42 neurons Activation functions: Identity Hidden layer 29 neurons Activation functions: ReLU Hidden layer 16 neurons Activation functions: ReLU Output layer 1 neuron Activation function: Identity Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.01 Learning strategy: seed search Cost function: mean square error 3.10 Apixaban Dosages (Universal or "improved methodology based on the universal methodology") 3.10.1 Datasets

[0314] Data for Eliquis ® (apixaban) were generated in two separate studies. For the first study, the data generated are: 63 kinetics measured on 16.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 20, 27, 39, 41, 52, 61, 67, 76, 93, 96, 105, 123, 127, 140, 145, 155, 167, 184 and 182 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 63 kinetics measured on 05.10.2017 on the STA-R ®< AUT06360 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 20, 26, 36, 43, 53, 60, 70, 75, 93, 103, 109, 121, 135, 136, 153, 160, 175, 185 and 192 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 63 kinetics measured on 16.10.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 20, 26, 34, 44, 50, 61, 70, 74, 94, 103, 109, 122, 135, 138, 147, 162, 173, 182 and 186 ng / mL; each sample level is tested in n=3. Samples were prepared and stored at -80°C. 63 kinetics measured on 05.10.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 20, 33, 41, 46, 57, 65, 76, 79, 96, 105, 116, 125, 135, 138, 153, 161, 169, 180 and 186 ng / mL; each sample level is tested in n =3. Samples were prepared and stored at -80°C.

[0315] For this first study, STA ®< - Liquid Anti-Xa lot 251738 was used for the generation of kinetics in universal methodology and for the commercial dosage of apixaban overload. Regarding the second study, the data generated are: 63 kinetics measured on 07.05.2019 on the STA-R ®< AUT05676 (soft version 3.04.07) normal pool of plasmas lot 19059RD (code 22824) overloaded with 0, 11, 16, 32, 39, 52, 62, 70, 84, 95, 104, 118, 123, 132, 142, 156, 167, 174, 183, 190 and 198 ng / mL; each level of samples is tested inn =3; samples were prepared and stored at -80°C, then tested simultaneously on the three automatons. 63 kinetics measured on 07.05.2019 on the STA-R ®< AUT05980 (soft version 3.04.07) normal pool of plasmas lot 19059RD (code 22824) overloaded with 0, 16, 23, 37, 48, 60, 73, 80, 96, 102, 111, 125, 136, 146, 155, 166, 186, 191, 193, 205 and 211 ng / mL; each level of samples is tested in n =3; samples were prepared and stored at -80°C, then tested simultaneously on the three automatons. 63 kinetics measured on 07.05.2019 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 19059RD (code 22824) overloaded with 0, 9, 16, 29, 36, 50, 61, 67, 81, 94, 102, 115, 123, 126, 144, 156, 159, 175, 183, 191 and 192 ng / mL; each level of samples is tested in n =3; the samples were prepared and stored at -80°C, then tested simultaneously on the three machines.

[0316] For this second study, STA ®< - Liquid Anti-Xa lot 253225 was used for the generation of kinetics in universal methodology and for the commercial dosage of apixaban overload. 3.10.2 Machine learning model Data organization

[0317] The machine learning model is trained by cross-validation divided into seven subsets as follows: Subset 1: Training data: apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT00603, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05676, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05980, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06360, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366 (bis). Validation data: apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT06399. Subset 2: Training data: apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT00603, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05676, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05980, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06360, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366 (bis), apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06399.Validation data: apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT06366. Subset 3: Training data: apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT00603, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05676, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05980, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06360, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06399. Validation data: apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT06366 (bis).Subset 4: Training data: apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT00603, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05676, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05980, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366 (bis), apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06399. Validation data: apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT06360. Subset 5: Training data: apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT00603, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05676, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06360, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366 (bis), apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06399.Validation data: apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT05980. Subset 6: Training data: apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT00603, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05980, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06360, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366 (bis), apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06399. Validation data: apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT05676.Subset 7: Training data: apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05676, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT05980, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06360, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366, apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06366 (bis), apixaban data (Eliquis ®< ) generated on the STA-R ®< AUT06399. Validation data: apixaban (Eliquis ®< ) data generated on the STA-R ®< AUT00603. .

[0318] The final learning is done on the entire data set. Machine learning model description

[0319] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 77 neurons Activation functions: Identity Hidden layer 40 neurons Activation functions: ReLU Output layer 1 neuron Activation function: Identity Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.001 Learning strategy: seed search Cost function: mean square error 3.11 Dosage of edoxaban (optimized AODs methodology) 3.11.1 Datasets

[0320] The data generated for Lixiana ®< (edoxaban) are: 72 kinetics measured in January 2016 on the STA-R ®< AUT00460 normal pool of plasmas spiked with 0, 16, 21, 30, 39, 51, 76, 100, 128, 142, 155, 187, 215, 245, 269, 284, 322, 347, 359, 380, 393, 410, 426 and 436 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 72 kinetics measured in June 2016 on the STA-R ®< AUT00722 normal plasma pool spiked with 0, 16, 21, 30, 39, 51, 76, 100, 128, 142, 155, 187, 215, 245, 269, 284, 322, 347, 359, 380, 393, 410, 426 and 436 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 3.11.2 Machine learning model Data organization

[0321] The machine learning model is trained through cross-validation organized as follows: Training data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00460. Validation data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00722. Machine learning model description

[0322] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 42 neurons Activation functions: Identity Hidden layer 29 neurons Activation functions: ReLU Hidden layer 16 neurons Activation functions: ReLU Output layer 1 neuron Activation function: Identity Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.01 Learning strategy: seed search Cost function: mean square error 3.12 Dosages of edoxaban (universal methodology or "improved methodology based on the universal methodology") 3.12.1 Datasets

[0323] Data for Lixiana ® (edoxaban) were generated in two separate studies. For the first study, the data generated are: 63 kinetics measured on 17.10.2017 on the STA-R ®< AUT00603 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 20, 28, 40, 49, 64, 74, 95, 103, 109, 122, 132, 144, 131, 132, 157, 171, 195, 191 and 201 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 63 kinetics measured on 17.10.2017 on the STA-R ®< AUT05450 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 21, 28, 41, 49, 62, 74, 92, 102, 112, 120, 133, 142, 129, 133, 155, 168, 188, 200 and 206 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C. 63 kinetics measured on 21.09.2017 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 21, 27, 40, 49, 64, 73, 90, 99, 108, 122, 132, 144, 130, 136, 153, 169, 192, 194 and 216 ng / mL; each sample level is tested in n=3; samples were prepared and stored at -80°C. 63 kinetics measured on 20.09.2017 on the STA-R ®< AUT06399 (soft version 3.04.07) normal pool of plasmas lot 03.2017 spiked with 0, 10, 21, 29, 39, 48, 60, 73, 81, 92, 107, 115, 129, 136, 129, 127, 149, 163, 181, 195 and 199 ng / mL; each sample level is tested in n =3; samples were prepared and stored at -80°C.

[0324] For this first study, STA ®< - Liquid Anti-Xa lot 251738 was used for the generation of kinetics in universal methodology and for the commercial dosage of edoxaban overload. Regarding the second study, the data generated are: 63 kinetics measured on 06.05.2019 on the STA-R ®< AUT05676 (soft version 3.04.07) normal pool of plasmas lot 19059RD (code 22824) overloaded with 0, 12, 19, 27, 36, 50, 59, 70, 81, 89, 100, 112, 119, 136, 145, 142, 163, 176, 186, 198 and 207 ng / mL; each level of samples is tested inn =3; samples were prepared and stored at -80°C, then tested simultaneously on the three automatons. 63 kinetics measured on 06.05.2019 on the STA-R ®< AUT05980 (soft version 3.04.07) normal pool of plasmas lot 19059RD (code 22824) overloaded with 0, 13, 20, 29, 37, 50, 61, 71, 83, 88, 95, 108, 115, 134, 141, 137, 161, 169, 176, 188 and 198 ng / mL; each level of samples is tested in n =3; samples were prepared and stored at -80°C, then tested simultaneously on the three automatons. 63 kinetics measured on 06.05.2019 on the STA-R ®< AUT06366 (soft version 3.04.07) normal pool of plasmas lot 19059RD (code 22824) overloaded with 0, 14, 20, 29, 35, 52, 58, 70, 77, 85, 97, 107, 114, 131, 140, 136, 165, 170, 177, 189 and 200 ng / mL; each level of samples is tested in n =3; the samples were prepared and stored at -80°C, then tested simultaneously on the three machines.

[0325] For this second study, STA ®< - Liquid Anti-Xa lot 253225 was used for the generation of kinetics in universal methodology and for the commercial dosage of edoxaban overload. 3.12.2 Machine learning model Data organization

[0326] The machine learning model is trained by cross-validation divided into seven subsets as follows: Subset 1: Training data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00603, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05450, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05676, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05980, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366 (bis). Validation data: edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT06399. Subset 2: Training data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00603, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05450, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05676, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05980, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06399.Validation data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366 (bis). Subset 3: Training data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00603, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05450, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05676, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05980, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366 (bis), edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06399. Validation data: edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT06366.Subset 4: Training data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00603, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05450, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05676, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366 (bis), edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06399. Validation data: edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT05980. Subset 5: Training data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00603, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05450, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05980, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366 (bis), edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06399.Validation data: edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT05676. Subset 6: Training data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00603, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05676, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05980, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366 (bis), edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06399. Validation data: edoxaban (Lixiana ®< ) data generated on the STA-R ®< AUT05450.Subset 7: Training data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05450, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05676, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT05980, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366, edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06366 (bis), edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT06399. Validation data: edoxaban data (Lixiana ®< ) generated on the STA-R ®< AUT00603. .

[0327] The final learning is done on the entire data set. Machine learning model description

[0328] A multi-layer perceptron (neural network) was trained according to the strategy described in the previous section. It is defined by the following parameters and hyperparameters: Data preprocessing: normalization between 0 and 1 Machine learning model: Multi-layer perceptrons Hyper-parameters: Input layer 77 neurons Activation functions: Identity Hidden layer 40 neurons Activation functions: ReLU Output layer 1 neuron Activation function: Identity Initialization of the weight and bias matrix: Xavier Glorot Numerical method for weight and bias optimization: L-BFGS Regularization method: L2, alpha = 0.001 Learning strategy: seed search Cost function: mean square error 4 Results of the second study (data from section 3. above)

[0329] In this section, we list the performance results obtained by the different cascade machine learning models in Figure Error ! Reference source not found.6on data measured on real samples using the experimental methodology described here. Each subsection details the measured data as well as the performance obtained. A single measurement is sufficient to report a result; however, as demonstrated in this section, a triplicate measurement can improve performance. 4.1 Detection of the presence or absence of an anti-Xa anticoagulant

[0330] In this section, we give the results obtained for the detection of the presence or absence of an anti-Xa anticoagulant using the method described in section 3. above. 4.1.1 Test data

[0331] 39 samples identified as normal based on their PT, APTT and Fibrinogen results, 24 frozen samples from patients treated with unfractionated heparin (UFH), 62 frozen samples from patients treated with low molecular weight heparin (LMWH), 44 frozen samples from patients treated with Xarelto ®< (rivaroxaban), 37 frozen samples from patients treated with Eliquis ®< (apixaban) and 42 frozen samples from patients treated with Lixiana ®< (edoxaban) were tested.

[0332] Data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) where each sample is tested in n=3 with the universal methodology.

[0333] In the end, the simplicate analysis is done on 248 kinetics measured on 248 samples and the triplicate analysis is done on 744 kinetics measured on 248 samples. 4.1.2 Results

[0334] Tables 9 and 10 give respectively the confusion matrices associated with the detection of the presence or absence of an anti-Xa anticoagulant when the analysis is carried out in simplification and when the analysis is carried out in triplicate on the data of the test set. The results for the detection of the presence of an anti-Xa anticoagulant give an accuracy of 100% whether the analysis is carried out in simplification or in triplicate. The results for the detection of the absence of an anti-Xa anticoagulant give an accuracy of 97.44% whether the analysis is carried out in simplification or in triplicate. Presence or absence of an anti-Xa anticoagulant: confusion matrix. Simplicity analysis. TAB. 9: The results give an accuracy of 100% for detecting the presence of an anti-Xa anticoagulant. The results give an accuracy of 97.44% for detecting the absence of an anti-Xa anticoagulant. predicted outcome Presence of anti-Xa Absence of anti-Xa actual value Presence of anti-Xa 209 0 Absence of anti-Xa 1 38 Presence or absence of an anti-Xa anticoagulant: confusion matrix. Triplicate analysis. TAB. 10: The results give an accuracy of 100% for detecting the presence of an anti-Xa anticoagulant. The results give an accuracy of 97.44% for detecting the absence of an anti-Xa anticoagulant. predicted outcome Presence of anti-Xa Absence of anti-Xa actual value Presence of anti-Xa 209 0 Absence of anti-Xa 1 38 4.2 Identification of the anti-Xa anticoagulant category

[0335] In this section, we give the results obtained for the identification of the anti-Xa anticoagulant category using the method described in section 3. above. 4.2.1 Test data

[0336] 24 frozen samples from patients treated with unfractionated heparin (UFH), 62 frozen samples from patients treated with low molecular weight heparin (LMWH), 44 frozen samples from patients treated with Xarelto ®< (rivaroxaban), 37 frozen samples from patients treated with Eliquis ®< (apixaban) and 42 frozen samples from patients treated with Lixiana ®< (edoxaban) were tested.

[0337] Data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) where each sample is tested in n=3 with the universal methodology.

[0338] In the end, the simplicate analysis is done on 209 kinetics measured on 209 samples and the triplicate analysis is done on 627 kinetics measured on 209 samples. 4.2.2 Results

[0339] Tables 11 and 12 give the confusion matrices associated with the identification of the anti-Xa anticoagulant category when the analysis is carried out in simplification and when the analysis is carried out in triplicate on the test set data, respectively. The results for the identification of the anti-Xa anticoagulant category give an accuracy of 100% when the analysis is carried out in simplification and an accuracy of 100% when the analysis is carried out in triplicate. Identification of the anti-Xa anticoagulant category: confusion matrix. Simplicity analysis. TAB. 11: The results give 100% accuracy for identifying the anti-Xa anticoagulant category. predicted outcome Heparin AOD actual value Heparin 86 0 AOD 0 123 Identification of the anti-Xa anticoagulant category: confusion matrix. Triplicate analysis. TAB. 12: The results give 100% accuracy for identifying the anti-Xa anticoagulant category. predicted outcome Heparin AOD actual value Heparin 86 0 AOD 0 123 4.3 Identification of heparins

[0340] In this section we give the results obtained for the identification of heparins using the method described in section 3. above. 4.3.1 Test data

[0341] 24 frozen samples from patients treated with unfractionated heparin (UFH) and 62 frozen samples from patients treated with low molecular weight heparin (LMWH) were tested.

[0342] Data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) where each sample is tested in n=3 with the universal methodology.

[0343] In the end, the simplicate analysis is done on 86 kinetics measured on 86 samples and the triplicate analysis is done on 258 kinetics measured on 86 samples. 4.3.2 Results

[0344] Tables 13 and 14 give the confusion matrices associated with the identification of heparins when the analysis is carried out in simplicat and when the analysis is carried out in triplicate on the data of the test set, respectively. The results for the identification of heparins give an accuracy of 91.86% when the analysis is carried out in simplicat and an accuracy of 90.70% when the analysis is carried out in triplicate. Identification of heparins: confusion matrix. Simplicity analysis. TAB. 13: The results give an accuracy of 91.86% for the identification of heparins. predicted outcome HNF HBPM actual value HNF 21 3 HBPM 4 58 Identification of heparins: confusion matrix. Triplicate analysis. TAB. 14: The results give an accuracy of 90.70% for the identification of heparins. predicted outcome HNF HBPM actual value HNF 21 3 HBPM 5 57 4.4 Assays of unfractionated heparins

[0345] In this section, we give the results of the HNF level measurements on patient samples obtained using the method described in section 3. above, in comparison with the levels measured using the standard approach (commercial kit STA ®< - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). 4.4.1 Test data

[0346] 24 frozen samples from patients treated with unfractionated heparin (UFH) were tested.

[0347] The data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) under the following conditions: each sample is tested in n =3 with universal methodology; each sample is tested in n=2 with the reference method (commercial methodology of STA ®< - Liquid Anti-Xa). The reference rate used in the method comparison is therefore a rate verified after freezing the sample.

[0348] In the end, the simplicate analysis is done on 24 kinetics measured on 24 samples and the triplicate analysis is done on 72 kinetics measured on 24 samples. 4.4.2 Results

[0349] There figure 17 gives the results of comparing the HNF levels measured using the approach described in this document to the HNF levels measured using the standard approach on the test set data. These comparisons give a straight line with the equation y =1.04x-0.03 and a coefficient of determination R 2< =0.9851 when the analysis is carried out in simplicat; they give a straight line of equation y =1.04x-0.03 and a coefficient R2< =0.9856 when the analysis is performed in triplicate. 4.5 Assays of low molecular weight heparins

[0350] In this section, we give the results of the LMWH level assays on patient samples obtained using the invention described in this document in comparison with the levels measured using the standard approach (commercial kit STA ®< - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). 4.5.1 Test data

[0351] 62 frozen samples from patients treated with low molecular weight heparin (LMWH) were tested.

[0352] The data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) under the following conditions: each sample is tested in n=3 with the universal methodology; each sample is tested in n=2 with the reference method (commercial methodology of STA ®< - Liquid Anti-Xa). The reference rate used in the method comparison is therefore a rate verified after freezing the sample.

[0353] In the end, the simplicate analysis is done on 62 kinetics measured on 62 samples and the triplicate analysis is done on 186 kinetics measured on 62 samples. 4.5.2 Results

[0354] There figure 18 gives the results of comparing the dosages of LMWH levels measured using the approach described in this document to the LMWH levels measured using the standard approach on the test set data. These comparisons give a straight line with the equation y =1.02x-0.02 and a coefficient of determination R2< =0.996 when the analysis is carried out in simplicat; they give a straight line of equation y =1.027x-0.03 and a coefficient of determination R 2< =0.9971 when the analysis is performed in triplicate. 4.6 Identification of anti-Xa AOD (optimized AODs methodology)

[0355] In this section, we give the results obtained for the identification of anti-Xa AODs using the invention described in this document. 4.6.1 Test data

[0356] 62 frozen samples from patients treated with Xarelto ®< (rivaroxaban), 45 frozen samples from patients treated with Eliquis ®< (apixaban) and 56 frozen samples from patients treated with Lixiana ®< (edoxaban) were tested.

[0357] Data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) where each sample is tested in n=3 with the optimized AODs methodology. In the end, the simplicate analysis is done on 163 kinetics measured on 163 samples and the triplicate analysis is done on 489 kinetics measured on 163 samples. 4.6.2 Results

[0358] Tables 15 and 16 give respectively the confusion matrices associated with the identification of anti-Xa DOAs when the analysis is carried out in simplification and when the analysis is carried out in triplicate on the data of the test set. The results for the identification of anti-Xa DOAs give an accuracy of 92.64% when the analysis is carried out in simplification and an accuracy of 97.55% when the analysis is carried out in triplicate. Identification of anti-Xa AODs confusion matrix. Simplicity analysis. TAB. 15: : The results give an accuracy of 92.64% for the identification of anti-Xa DOAs. predicted outcome rivaroxaban apixaban edoxaban actual value rivaroxaban 59 0 3 apixaban 0 45 0 edoxaban 9 0 47 Identification of anti-Xa DOACs: confusion matrix. Triplicate analysis. TAB. 16: The results give an accuracy of 97.55% for the identification of anti-Xa DOAs. predicted outcome rivaroxaban apixaban edoxaban actual value rivaroxaban 62 0 0 apixaban 0 45 0 edoxaban 4 0 52 4.7 Rivaroxaban dosage (optimized AODs methodology)

[0359] In this section, we give the results of rivaroxaban level assays on patient samples obtained using the invention described in this document (optimized AODs methodology) in comparison with the levels measured using the standard approach (commercial kit STA ®< - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). 4.7.1 Test data

[0360] 62 frozen samples from patients treated with Xarelto ® (rivaroxaban) were tested.

[0361] The data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) under the following conditions: each sample is tested in n =3 with the optimized AODs methodology; each sample is tested in n=2 with the reference method (commercial methodology of STA ®< - Liquid Anti-Xa). The reference rate used in the method comparison is therefore a rate verified after freezing the sample.

[0362] In the end, the simplicate analysis is done on 62 kinetics measured on 62 samples and the triplicate analysis is done on 186 kinetics measured on 62 samples. 4.7.2 Results

[0363] There figure 19 gives the results of comparing the dosages of rivaroxaban levels measured using the approach described in this document to the rivaroxaban levels measured using the standard approach on the test set data. These comparisons give a straight line with the equation y =1.05x+19.65 and a coefficient of determination R 2< =0.991 when the analysis is carried out in simplicat; they give a straight line of equation y =1.04x+20.3 and a coefficient of determination R2< =0.9934 when the analysis is performed in triplicate. 4.8 Rivaroxaban dosage (universal methodology or "improved methodology based on the universal methodology")

[0364] In this section, we give the results of rivaroxaban level assays on patient samples obtained using the invention described in this document (universal methodology) in comparison with the levels measured using the standard approach (commercial kit STA ®< - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). 4.8.1 Test data

[0365] 44 frozen samples from patients treated with Xarelto ® (rivaroxaban) were tested.

[0366] The data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) under the following conditions: each sample is tested in n=3 with the universal methodology; each sample is tested in n=2 with the reference method (commercial methodology of STA ®< - Liquid Anti-Xa). The reference rate used in the method comparison is therefore a rate verified after freezing the sample.

[0367] In the end, the simplicate analysis is done on 44 kinetics measured on 44 samples and the triplicate analysis is done on 132 kinetics measured on 44 samples. 4.8.2 Results

[0368] There figure 20gives the results of comparing the dosages of rivaroxaban levels measured using the approach described in this document to the rivaroxaban levels measured using the standard approach on the test set data. These comparisons give a straight line with the equation y =1.04x+1.74 and a coefficient of determination R 2< =0.985 when the analysis is carried out in simplicat; they give a straight line of equation y =1.032x+1.99 and a coefficient R 2< =0.989 when the analysis is performed in triplicate. 4.9 Apixaban dosage (optimized AODs methodology)

[0369] In this section, we give the results of the apixaban level assays on patient samples obtained using the invention described in this document (optimized AODs methodology) in comparison with the levels measured using the standard approach (commercial kit STA ®< - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). 4.9.1 Test Data

[0370] 45 frozen samples from patients treated with Eliquis ® (apixaban) were tested.

[0371] The data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) under the following conditions: each sample is tested in n =3 with the optimized AODs methodology; each sample is tested in n=2with the reference method (commercial methodology of STA ®< - Liquid Anti-Xa). The reference rate used in the method comparison is therefore a rate verified after freezing the sample.

[0372] In the end, the simplicate analysis is done on 45 kinetics measured on 45 samples and the triplicate analysis is done on 135 kinetics measured on 45 samples. 4.9.2 Results

[0373] There figure 21 gives the results of comparing the dosages of apixaban levels measured using the approach described in this document to the apixaban levels measured using the standard approach on the test set data. These comparisons give a straight line with the equation y =1.14x-6.73 and a coefficient of determination R 2< =0.9945 when the analysis is carried out in simplicat; they give a straight line of equation y =1.13x-5.46 and a coefficient of determination R2< =0.9958 when the analysis is performed in triplicate. 4.10 Dosage of apixaban (universal methodology or "improved methodology based on the universal methodology")

[0374] In this section, we give the results of the apixaban level assays on patient samples obtained using the invention described in this document (universal methodology) in comparison with the levels measured using the standard approach (STA ®< commercial kit - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). 4.10.1 Test Data

[0375] 37 frozen samples from patients treated with Eliquis ® (apixaban) were tested.

[0376] The data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) under the following conditions: each sample is tested in n =3 with universal methodology; each sample is tested in n=2 with the reference method (commercial methodology of STA ®< - Liquid Anti-Xa). The reference rate used in the method comparison is therefore a rate verified after freezing the sample.

[0377] In the end, the simplicate analysis is done on 37 kinetics measured on 37 samples and the triplicate analysis is done on 111 kinetics measured on 37 samples. 4.10.2 Results

[0378] There figure 22gives the results of comparing the dosages of apixaban levels measured using the approach described in this document to the apixaban levels measured using the standard approach on the test set data. These comparisons give a straight line with the equation y =1.041 x -2.66 and a coefficient of determination R 2< =0.9792 when the analysis is carried out in simplicat; they give a straight line of equation y =1.041 x -2.41 and a coefficient R 2< =0.9877 when the analysis is performed in triplicate. 4.11 Dosage of edoxaban (optimized AODs methodology)

[0379] In this section, we give the results of the edoxaban level assays on patient samples obtained using the invention described in this document (optimized AODs methodology) in comparison with the levels measured using the standard approach (commercial kit STA ®< - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). 4.11.1 Test Data

[0380] 56 frozen samples from patients treated with Lixiana ® (edoxaban) were tested.

[0381] The data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) under the following conditions: each sample is tested in n =3 with the optimized AODs methodology; each sample is tested in n=2 with the reference method with the commercial methodology of STA ®< - Liquid Anti-Xa. The reference rate used in the method comparison is therefore a rate verified after freezing the sample.

[0382] In the end, the simplicate analysis is done on 56 kinetics measured on 56 samples and the triplicate analysis is done on 168 kinetics measured on 56 samples. 4.11.2 Results

[0383] There figure 23 gives the results of comparing the dosages of edoxaban levels measured using the approach described in this document to the edoxaban levels measured using the standard approach on the test set data. These comparisons give a straight line with the equation y =0.905 x +12.35 and a coefficient of determination R 2< =0.9853 when the analysis is carried out in simplicat; they give a straight line of equation y =0.93 x +8.18 and a coefficient of determinationR 2< =0.9881 when the analysis is performed in triplicate. 4.12 Dosage of edoxaban (universal methodology or "improved methodology based on the universal methodology")

[0384] In this section, we give the results of the edoxaban level assays on patient samples obtained using the invention described in this document (universal methodology) in comparison with the levels measured using the standard approach (commercial kit STA ®< - Liquid Anti-Xa). The results are considered satisfactory when the slope of the linear regression is between 0.9 and 1.1 and the coefficient of determination R 2< is greater than or equal to 0.95 (CLSI EP9-A2 criteria). 4.12.1 Test Data

[0385] 42 frozen samples from patients treated with Lixiana ® (edoxaban) were tested.

[0386] The data were generated on the STA-R ®< AUT06399 (soft version 3.04.07) with the STA ®< - Liquid Anti-Xa (lot 251187) under the following conditions: each sample is tested in n =3 with universal methodology; each sample is tested in n=2 with the reference method (commercial methodology of STA ®< - Liquid Anti-Xa). The reference rate used in the method comparison is therefore a rate verified after freezing the sample.

[0387] In the end, the simplicate analysis is done on 42 kinetics measured on 42 samples and the triplicate analysis is done on 126 kinetics measured on 42 samples. 4.12.2 Results

[0388] There figure 24gives the results of comparing the dosages of edoxaban levels measured using the approach described in this document to the edoxaban levels measured using the standard approach on the test set data. These comparisons give a straight line with the equation y =0.9431 x +4.51 and a coefficient of determination R 2< =0.9968 when the analysis is carried out in simplicat; they give a straight line of equation y =0.9442x+4.26 and a coefficient R 2< =0.9962 when the analysis is performed in triplicate. INHIBITORS I Factor Xa inhibitors

[0389] Tables 17 and 18 list the natural and synthetic factor Xa inhibitors known to date, respectively. [Table 17]

[0390] Table 17: Natural inhibitors of factor Xa Directs Indirect Reversible TFPI Protein S Irreversible Antithrombin Nexin-1 protease [Table 18]

[0391] Table 18: Synthetic factor Xa inhibitors Directs Indirect Reversible Rivaroxaban Apixaban Edoxaban Betrixaban Irreversible HNF HBPM Pentasaccharides Danaparoid sodium (Orgaran) II Factor IIa inhibitors

[0392] Tables 19 and 20 list the natural and synthetic factor IIa inhibitors known to date, respectively. [Table 19]

[0393] Table 19: Natural inhibitors of factor IIa Directs Indirect Reversible Irreversible Antithrombin Heparin cofactor II Nexin-1 protease α2-macroglobulin [Table 20]

[0394] Table 20: Synthetic inhibitors of factor IIa Directs Indirect Reversible Dabigatran Melagatran Argatroban Bivalirudin Irreversible Hirudin HNF Lepirudin Desirudin HBPM Antithrombin (Aclotin ®) References

[0395] Géron, A. (2017). Hands-On Machine Learning with Scikit-Learn and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems. O’Reilly.

[0396] Segel, I. H. (1993). Enzyme kinetics. Behavior and analysis of rapid equilibrium and steady-state enzyme Systems. Wiley Classics Library.

[0397] Bonaccorso, G. (2017). Machine Learning Algorithms: A reference guide to popular algorithms for data science and machine learning. Packt Publishing Limited.

Claims

1. Method of detecting in a biological sample, in particular a blood sample, the presence of an inhibitor of a blood coagulation enzyme selected, independently, from factor Xa (FXa) and factor IIa (FIIa), the method comprising the following steps: a. Measurement of one or more competition kinetics by performing a competitive enzymatic assay on a blood sample previously obtained from a subject, said assay being adapted to perform competitive kinetics with respect to either a factor Xa inhibitor or a factor IIa inhibitor, and then b. Input of one or more kinetics obtained in step a. to a classification decision model A obtained by training a supervised automatic learning model, for example a model chosen from one of the following families: support vector machine, neural networks, decision trees, ensemble methods, and k nearest neighbours model, then b.i. If the decision model A excludes the presence of an inhibitor of the targeted blood coagulation enzyme in the sample analysed, conclusion of the absence of said inhibitor, optionally allocation as output by the model A, for example in a variable, of the corresponding information, or b.ii. If the decision model A attests to the presence of an inhibitor of the targeted blood coagulation enzyme in the sample analysed, conclusion of the presence of said inhibitor, optionally allocation as output by the model A, for example in a variable, of the corresponding information.

2. Method of identifying in a biological sample, in particular a blood sample, an inhibitor of a blood coagulation enzyme selected, independently, from factor Xa (FXa) and factor IIa (FIIa), the method comprising the following steps:

1. Carrying out the steps of the method according to claim 1, then 2. Inputting of the kinetic(s) obtained in the step defined in point a. of claim 1, or step 1. above, and of the result obtained at the end of step b.ii. of claim 1, to a classification decision model B obtained by training a supervised automatic learning model, for example a model chosen from one of the following families: support vector machine, neural networks, decision trees, ensemble methods, and k nearest neighbours model, and allocation at the output by the model B of the category of the inhibitor to one of the following categories: irreversible indirect inhibitor (heparins), or reversible direct inhibitor (AOD), in particular when the data sets used for training model B comprise data relating to these two categories of inhibitor, and outputting, for example by allocation in a variable, the category of inhibitor determined by model B.

3. Method according to claim 2, comprising an additional step of characterizing the inhibitor whose presence has been detected in step b.ii of claim 1, as follows: inputting the kinetic(s) obtained in step a. of claim 1 or step 1. of claim 2, and the data determined, for example the allocated variable, in step 2. of claim 2 concerning the category of inhibitor whose presence has been detected, to a classification decision model C obtained by training a supervised automatic learning model, for example a model chosen from one of the following families: support vector machine, neural networks, decision trees, ensemble methods, and k nearest neighbours model, and output characterisation by the model C of the inhibitor sought, the latter being identified from: a. If the category of inhibitor sought is that of the heparins: UFH or LMWH, or b. If the category of inhibitor sought is that of the AODs: Rivaroxaban, Apixaban, Edoxaban or Dabigatran, and output, for example by allocation in a variable, the characterisation of the inhibitor determined by the C model.

4. Method according to claim 3, comprising an additional step of quantitative assay of the inhibitor characterised, in which are supplied as input to a regression model D, in particular a supervised automatic learning model, for example a model chosen from one of the following families: support vector machines, neural networks, decision trees, ensemble methods, and k nearest neighbours model, the kinetic(s) obtained in step a. of claim 1 or step 1. of claim 2, and the characterisation data obtained according to claim 3 identifying the inhibitor present in the blood sample analysed, said regression model having been trained on a set of data obtained under measurement conditions identical to those in step a. of claim 1 or step 1. of claim 2, and making it possible to determine as output the concentration of the inhibitor identified in the sample analysed and optionally to supply as output, for example by allocation in a variable, the concentration determined by model D.

5. Method according to any one of claims 1 to 4, in which the inhibitor sought is: I. a factor Xa (FXa) inhibitor chosen from: UFH, LMWH, Rivaroxaban, Apixaban, Edoxaban, or II. a factor IIa inhibitor (FIIa) chosen from: UFH, LMWH, Dabigatran.

6. Method according to claim 2, applied to the search for a factor Xa inhibitor whose presence has been detected in step b.ii of claim 1, comprising the following additional characterisation step: I. if the inhibitor class has been allocated in step 2. of claim 2 to the heparin class, then: I.i. supplying as input to a classification decision model C obtained by training a supervised automatic learning model, for example a model chosen from one of the following families: support vector machine, neural networks, decision trees, ensemble methods, and k nearest neighbours model, said determined data, for example the allocated variable, in step 2. of claim 2 concerning the category of inhibitor whose presence has been detected, and of the competitive kinetic(s) with respect to a factor Xa inhibitor, obtained in step a. of claim 1 or step 1. of claim 2, and I.ii. output characterisation by model C of the inhibitor sought, the latter being identified from: UFH or LMWH, and output, for example by allocation in a variable, of the characterisation of the inhibitor determined by model C, or alternatively, II. if the inhibitor class has been allocated in step 2. of claim 2 to the AODs class, then: II.i. performance of a new measurement of one or more competition kinetics by performing a competitive enzymatic assay on a blood sample obtained from the same subject, said assay being adapted to the performance of competitive kinetics with respect to a factor Xa inhibitor, with a sample dilution factor and / or a measurement time adapted to a competitive situation involving the presence of AOD inhibiting factor Xa, in particular a dilution factor and / or a measurement time different from those used for measuring the kinetic(s) obtained in step a. of claim 1 or step 1. of claim 2, then II.ii. supplying as input to a classification decision model C obtained by training a supervised automatic learning model, for example a model chosen from one of the following families: support vector machine, neural networks, decision trees, ensemble methods, and k nearest neighbours model, said determined data, for example the allocated variable, in step 2. of claim 2 concerning the category of inhibitor whose presence has been detected, and of the kinetic(s) obtained in the preceding step i., and II.iii. output characterisation by model C of the inhibitor sought, the latter being identified from: Rivaroxaban, Apixaban, or Edoxaban, and provision at the output, for example by allocation in a variable, of the characterisation of the inhibitor determined by model C.

7. Method according to claim 6, comprising an additional step of quantitative assay of the inhibitor identified at the end of steps I. or II. of claim 6, in which, respectively: I. If the inhibitor identified in step I. of claim 6 is an UFH or LMWH, then: supplying as input to a regression model D, in particular a supervised automatic learning model, for example a model chosen from one of the following families: support vector machines, neural networks, decision trees, ensemble methods, and k nearest neighbours model, said model having been trained on a data set obtained under measurement conditions identical to those in step a. of claim 1, or step 1. of claim 2 of the kinetic(s) obtained in step a. of claim 1 or step 1. of claim 2 and of the data determined in step I.ii of claim 6 identifying the inhibitor present in the analysed blood sample, said regression model making it possible to determine as an output the concentration of inhibitor identified in the sample analysed, and optionally supplying as an output, for example by allocation in a variable, the concentration determined by model D, or II. If the inhibitor identified in step II. of claim 6 is Rivaroxaban, Apixaban, or Edoxaban, then: supplying as input to a regression model D, in particular a supervised automatic learning model, for example a model chosen from one of the following families: support vector machines, neural networks, decision trees, ensemble methods, and k nearest neighbours model, said model having been trained on a data set obtained under measurement conditions identical to those of step II. i. of claim 6, the kinetic(s) obtained in step II. i. of claim 6 and the characterisation data, for example the allocated variable, supplied at the end of step II. iii. of claim 6 identifying the inhibitor present in the analysed blood sample, said regression model making it possible to determine as output the concentration of inhibitor identified in the sample analysed and optionally supplying as output, for example by allocation in a variable, the concentration determined by model D.

8. Method according to step II. of claim 7, in which if the concentration of inhibitor identified in the sample analysed, as determined by model D, is less than or equal to 200 ng / mL, then input to a regression model D2, in particular a supervised automatic learning model, for example a model chosen from one of the following families: support vector machines, neural networks, decision trees, ensemble methods, and k nearest neighbours model, said model having been trained on a set of data obtained under measurement conditions identical to those in step a. of claim 1, of the kinetic(s) obtained in step a. of claim 1, said regression model D2 making it possible to recalculate as output the concentration of inhibitor identified in the sample analysed and optionally supplying as output, for example by allocation in a variable, the concentration determined by model D2.

9. Method according to any one of claims 1 to 8, wherein the in vitro measurement of competition kinetics by competitive enzyme assay on a blood sample obtained from a subject comprises the following steps: a. supply of a blood sample, diluted or undiluted, then b. addition to the blood sample of a substrate specific for either factor Xa or factor IIa, depending on the inhibitor sought, in particular a chromogenic or fluorogenic substrate, c. incubation with raising the temperature of the mixture obtained in b. to between 35 and 39°C, in particular 37°C, d. addition to the reaction mixture obtained from c. of factor Xa or factor IIa, depending on the substrate added in step b., so as to trigger competition between an inhibition reaction and the induced enzymatic reaction, e. measurement by an instrument, over time, of the quantity of product resulting from the transformation of the substrate as a result of the action of the enzyme analysed on the latter, chosen from factor Xa or factor IIa, where appropriate by measuring a marker associated with the substrate, released during said enzymatic reaction, and recording of the kinetics obtained.

10. Method according to claim 9, wherein the competitive enzyme assay is specific for factor Xa, and wherein: a. in step a. the blood sample is a plasma sample diluted 1 / 2 in an Owren Koller buffer, b. in step b. the substrate is MAPA-Gly-Arg-pNA reagent, c. in step c. the incubation time is 240 seconds at 37°C, d. factor Xa added to the mixture in step d. is bovine factor Xa, e. the release of paranitroaniline (pNA) in step e. is measured by colorimetry at 405 nm every two seconds for 156 seconds, on an appropriate instrument, such as STA-R®, or alternatively: a. in step a. the blood sample is a plasma sample diluted 1 / 8th in an Owren Koller buffer, b. in step b. the substrate is MAPA-Gly-Arg-pNA reagent, c. in step c. the incubation time is 240 seconds at 37°C, d. factor Xa added to the mixture in step d. is bovine factor Xa, e. the release of paranitroaniline (pNA) in step e. is measured by colorimetry at 405 nm every two seconds for 86 seconds, on an appropriate instrument, such as STA-R®,11. Method according to claim 9, in which the competitive enzyme assay is carried out on a miniaturised device, for example a device using microfluidics, in a reaction volume of between 1 and 20 µL.

12. Method according to any one of claims 1 to 11, in which: a. the blood sample is a plasma sample, and / or b. the input to a supervised automatic learning model of a kinetic obtained experimentally consists of supplying the pairs of values constituted by each value measured for each discrete measurement point taken during the measurement time.

13. Data processing system or device for detecting, in a biological sample, the presence or identification of an inhibitor of factor Xa (FXa) or of factor IIa (FIIa), said data processing system or device comprising means for implementing at least step b. of claim 1 and means for supplying as input one or more competition kinetics obtained by performing a competitive assay against either a factor Xa inhibitor or a factor IIa inhibitor on a blood sample previously obtained from a subject, in particular via a measuring device, and means for supplying as output the variables generated during this step, and where appropriate also means for implementing step 2. of the method according to claim 2, and optionally also comprising means for providing as input and / or output the variables generated during this step, and optionally also comprising means for implementing the steps involving the classification or regression decision models C, D, and where appropriate D2, of one of claims 3 to 12, in particular means making it possible to make use of a matrix resulting from the learning which has made it possible to set the parameters of these models, in order to return the classification or regression result taking into account the variables supplied to the model(s), and optionally also comprising a processor adapted to implement the steps indicated in the present claim.

14. Computer program comprising instructions which lead a data processing system or a device according to claim 13, in particular a device including an apparatus for measuring kinetics as defined in any one of claims 1 or 6 or a data processing system or a device calling upon such a measuring apparatus, in particular a remote one, to execute at least step b. of claim 1, and where appropriate also step 2. of the method according to claim 2, and optionally leading to execution of the steps involving the classification or regression decision models C, D, and where appropriate D2, of one of claims 3 to 12.

15. Computer-readable recording medium comprising instructions which, when executed by a computer, cause the latter to implement at least step b. of claim 1, and where appropriate also step 2. of the method according to claim 2, and optionally also make it possible to recover as input and / or provide as output the variables generated during these steps, and optionally cause the latter to implement the steps involving the classification or regression decision models C, D, and where appropriate D2, of one of claims 3 to 12.

16. Computer-readable data medium on which the computer program according to claim 14 is recorded, or a signal from a data medium carrying the computer program according to claim 14.

17. Kit for implementing a method according to one of claims 1 to 12, comprising: a. A substrate specific to FXa and / or Flla, for example the MAPA-Gly-Arg-pNA substrate for factor Xa, for example the EtM-SPro-Arg-pNA substrate for factor IIa, and b. Optionally, FXa and / or Flla, for example bovine or human factor Xa, for example bovine or human factor IIa, and c. Optionally, one or more suitable buffers, e.g. Owren Koller buffer, e.g. Tris EDTA buffer and d. Optionally instructions for implementing measurement(s) of one or more competition kinetics by performing a competitive enzymatic assay using the substrate, and e. the system and / or device and / or computer program and / or computer-readable data medium according to one of claims 13 to 15, f. and optionally instructions allowing to implement the method according to one of claims 1 to 12, g. and optionally instructions relating to the use of a signal from a data medium according to claim 16, for the implementation of a method according to one of claims 1 to 12.