Method and device for the diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral profile of at least one biological fluid.

Abstract

Title: Method and device for the diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral profile of at least one biological fluid. The method according to the invention consists of: - Implementing a reference database (RDB) comprising data relating to samples of biological fluid, and to the healthy or sick subjects from whom the samples originate; - Implementing a sample Ex from a subject Sx to be tested; - Analyzing each sample to determine at least one Mineral Profile of Sx; - Defining diagnostic (d), prognostic (p), and therapeutic monitoring (t) categories to make at least one prediction of Sx's membership in a category; - Statistically processing the RDB data and graphically visualizing the processed data, then positioning the Sx data on this visualization; - and / or performing machine learning on the RDB data;to make the prediction of membership concerning Sx and thereby assist in the diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, for Sx Figure 1;

Description

Title of the invention: Method and device for diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral profile of at least one biological fluid. Field of invention

[0001] The invention relates to the field of diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral profile of at least one biological fluid, in particular cerebrospinal fluid, whole blood or blood plasma, urine, follicular fluid, bile, etc. These diseases include: . in the case of cerebrospinal fluid, diseases affecting the central nervous system, in particular neurological diseases, such as psychiatric diseases, neurodegenerative diseases such as Parkinson's or Alzheimer's,... . in the case of blood plasma, cancers, neurological diseases, diseases related to fertility,... . in the case of urine, diseases related to surgical complications following complex operations, particularly cardiovascular,... . in the case of follicular fluids, diseases related to endometriosis or diseases causing fertility problems and requiring in particular procreation aids,...

[0002] The invention relates more specifically to a method for diagnosis, prognosis and / or therapeutic monitoring of these diseases.

[0003] The invention also relates to an electronic device, a system comprising the latter and a computer program involved in the implementation of this method. Technological background of the invention

[0004] Neurological diseases [such as neurodegenerative diseases], cancers, fertility problems, heart operations, have a very significant societal impact and represent a growing burden for the community. More than one in two people will be confronted with one of these diseases during their lifetime. Therefore, accurately and early enough diagnosing these diseases, establishing a prognosis or therapeutic follow-up forecasts, are crucial challenges for patients and their caregivers and to help the latter propose the most appropriate therapeutic treatments.

[0005] Due to the progressive aging of the population and the lack of curative treatments, the number of people suffering from neurodegenerative diseases has has increased considerably over the past few decades and is expected to grow steadily in the coming years.

[0006] Furthermore, the fertility rate has never been so low in developed countries, and recourse to medically assisted procreation will increase steadily in the years to come.

[0007] Cancers and heart disease have also increased considerably in recent years and the number of patients treated has more than doubled in less than 20 years.

[0008] It is therefore urgent to have better tools for the diagnosis, prognosis and monitoring of the effectiveness of treatments for these diseases.

[0009] Early, effective and accurate diagnosis is a priority objective.

[0010] But a diagnosis is often very complex, given the wide variety of possible pathologies. In addition, information is difficult to obtain, especially when the main organ concerned is the central nervous system and, more specifically, the brain.

[0011] The structural and functional alterations of the brain that cause these diseases modify the biochemical composition of the cerebrospinal fluid. The latter may therefore contain biomarkers of these diseases.

[0012] Many diagnoses are already based in part on biochemical analyses and the detection and quantification of certain organic molecules present in the cerebrospinal fluid.

[0013] Known biomarkers of cerebrospinal fluid are organic and especially protein molecules. For example, identified biomarkers of Alzheimer's disease are the proteins: amyloid 1-42, total tau, phosphorylated tau isoforms, NF-L, S90B, VILIP-1, YKL-40, PAPP-A, albumin.

[0014] EP1907838B1 describes a method for qualifying the state of Alzheimer's disease in a subject, comprising:(a) measuring the concentration of a VGF-1 peptide consisting of the amino acid sequence of SEQ ID NO:1 in the cerebrospinal fluid (CSF) of the subject; wherein the VGF-1 peptide is measured by capture on an adsorbent surface of a SELDI probe and detecting the captured peptide by laser desorption / ionization mass spectrometry; and (b) correlating the concentration of the VGF-1 peptide with Alzheimer's disease status, or an increased CSF concentration of VGF-1 peptide is an indicator of Alzheimer's disease. EP1907838B1 also discloses a method for determining the progression of Alzheimer's disease comprising: (a) measuring the LCS concentration of a VGF-1 peptide consisting of the amino acid sequence of SEQ ID N degrees 1 in a sample in a first step; (b) measuring the LCS concentration of VGF-1 peptide in a sample in a second step; and (c) comparing the first measurement and of the second measurement; wherein an increased LCS concentration of VGF peptide-1 in the second measurement indicates progression of Alzheimer's disease. EP19078388B1 finally relates to software installed on a computer comprising: (a) code which allows access to data assigned to a sample, the data comprising the measurement of a VGF peptide-1 consisting of the amino acid sequence of SEQ ID No. 1; and (b) code which executes a classification algorithm which assigns the Alzheimer's disease status of the sample, based on the measurement. In this software, the observation of an increase in VGF peptide-1 is an indicator of Alzheimer's disease.

[0015] Identification by trace organic molecules remains tricky in many cases, however. The molecules are often difficult to detect and quantify. They are not very stable and require precautions for sampling and storage. They can also interfere with other molecules, which makes their detection even more complex. Research has been carried out to try to link the presence of certain elements to the existence of a disease. Some results have been proposed. Thus, several studies have measured heavy elements and essential metals in the plasma and / or cerebrospinal fluid of patients with dementia. These studies show that there is a high variability. As a result, metals are still not considered reliable biomarkers of Alzheimer's disease. In addition, the processes of sampling and precise analysis of mineral traces are often complex.Mineral traces are also very indirectly associated with these types of diseases, as causes or consequences of these diseases. For all these reasons, mineral analysis of cerebrospinal fluid is not used, in practice, to diagnose neurological disease. Objectives of the invention

[0016] In this context, the invention aims to satisfy at least one of the following objectives: - to provide an efficient method for diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, in particular cerebrospinal fluid, whole blood or blood plasma, urine, follicular fluid, bile, etc., - to provide a reliable method for diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, in particular cerebrospinal fluid, whole blood or blood plasma, urine, follicular fluid, bile, etc., - to provide an accurate method for diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, in particular cerebrospinal fluid, whole blood or blood plasma, urine, follicular fluid, bile, etc., - provide a method for diagnosis, prognosis and / or therapy or early therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, in particular cerebrospinal fluid, whole blood or blood plasma, urine, follicular fluid, bile. - provide a method for diagnosis, prognosis and / or therapeutic monitoring of diseases affecting the mineral composition of at least one biological fluid, which constitutes a valuable and effective aid for the doctor, allowing the latter to treat and usefully contribute to the fastest and most complete cure possible for patients. - provide an electronic device and a system including this device, which are efficient, reliable, precise and economical for the implementation of the process referred to in the above objectives. Description of the invention

[0017] DEFINITIONS

[0018] Throughout this presentation, any singular designates indifferently a singular or a plural.

[0019] The definitions given below as examples may be used to interpret this disclosure: - “CNS”: Central Nervous System. - “CSF”: cerebrospinal fluid. - “CSF” cerebrospinal fluid. - “Differentiating data”: discriminating data which makes it possible to provide specific information on an Sx subject to be tested who may be a patient. - “AI” Artificial Intelligence. - "ROC": "Receiver Operating Characteristic", also known as performance characteristic (of a test) or sensitivity / specificity curve, which is a measure of the performance of a binary classifier, i.e. a system which aims to categorize elements into two distinct groups on the basis of one or more of the characteristics of each of these elements. - “AUC”: “Area Under the Curve”, area under the ROC curve. - “PCA”: Principal Component Analysis, a statistical technique used to reduce the dimensionality of data while preserving important information. - “t-SNE”: “t-distributed Stochastic Neighbor Embedding”, a dimensionality reduction method. - “RFE”: “Recursive Feature Elimination”, a feature selection method used in the field of machine learning. - “Cross-validation”: Method of evaluating the performance of a model by testing it on several subsets of the data to obtain a more reliable estimate of its performance. - “FPR”: “False Positive Rate”. A measure used in the evaluation of classification models, representing the rate of values ​​incorrectly predicted as positive compared to all actual negative values. - “TPR”: “True Positive Rate”. - “Sensitivity”: Measures the ability to detect true positives, i.e. the proportion of positive cases that are correctly identified by the test. - “Specificity”: Measures the ability to exclude false positives, i.e. the proportion of negative cases that are correctly identified as such by the test. - “approximately”: to within + or - 10%, preferably within 5%.

[0020] METHOD

[0021] The invention satisfies at least one of the above objectives and relates, according to a first aspect, to a method for diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, characterized in that it essentially consists of: -S0- Implement at least one reference database (RDB) comprising data relating, on the one hand, to biological fluid samples, and, on the other hand, to the subjects from whom the biological fluid samples come; - each sample corresponding to a healthy subject Sm "° of reference or to a sick subject Sm - 1 of reference treated or not, m being a positive natural integer; - each reference subject Sm(d>p>t, corresponding to a diagnostic category Sm(d) and / or a prognosis category Sm(p) and / or a therapeutic response category Sm(t); each category may possibly be subdivided into one or more levels; - the database comprising data relating to a total number N' of reference subjects Sm =0( d>p>t ) and Sm>l( d>p>t ); with N' greater than or equal to -in ascending order of preference- 50, 100, 500, 1000; - this data including for each sample: (i) at least one PM mineral profile of concentrations of mineral elements in the biological fluid of the reference subjects Sm "°( d>Pjt ) and / or Sm>l( d>Pjt j; preferably the concentrations of a number Nem mineral elements; * Nem being - in ascending order of preference - greater than or equal to 5, 10, 20, 25, 30; these mineral elements being chosen from the group comprising -advantageously consisting of: Li, B, Na, Mg, Al, Si, P, S, Cl, K, Ca, Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Ga, Ge, As, Se, Br, Rb, Sr, Y, Zr, Nb, Mo, Ru, Rh, Pd, Ag, Cd, In, Sn, Sb, I, Cs, Ba, La, Ce, Pr, Nd, Sm, Eu, Tb, Ta, W, Os, Pt, Au, Tl, Pb, Bi and U; and, preferably from the subgroup comprising -advantageously consisting of: Na, Mg, Al, Si, P, S, Cl, K, Ca, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Br, Rb, Sr, Zr, Sn, I, Cs, Ba, Pb, Bi and U; (ii) preferably, conventional clinical data relating to the reference subjects Sm =0 ( d,P,t ) and / or Sm>l( djP>t ), these data (ii) being chosen from the group comprising - advantageously consisting of -: age, gender, height, weight, data on the type of disease considered provided that it is known, data on the medical history of the reference subjects Sm "°( d,p,t ) and / or Sm>l( d>p>t ,dc, data on the therapeutic treatment(s) of the reference subjects Sm =0( djPjt ) and / or Sm>l( djP>t ), provided that they exist, lifestyle, concentrations of components of the biological fluid other than mineral elements, in particular concentrations of biomarkers, proteins, lipids, lipoproteins, red blood cells, white blood cells, platelets; -SI- Implement at least one Ex sample of biological fluid from at least one Sx subject to be tested; -S2- Optionally assign to each sample Ex of step -SI- DSx data on the subject Sx to be tested, these DSx data meeting the same definition as the data (ii) referred to in SO; -S3- Possibly keep at least part of the Ex samples from step SI under specific conditions; -S4- Analyze each sample from step SI to determine at least one PM as defined in S0(i); -S5- Optionally, complete and / or update the data assigned in step S2, at least once, for all or part of the samples; -S6- Optionally complete and / or repeat the analyses carried out in step S4, at least once, on all or part of the samples from step S2 or S3; -S7- Enrich the database with the data produced in at least one of steps S4, S5, S6; -S8- Define at least two targets to make at least one prediction of membership of the subject Sx to be tested, to at least one target, each target comprising at least one category and / or at least one subdivision of at least one category; -S9- Implement -S9.1 - : 1st protocol consisting of - Carry out statistical processing of the BDR data, preferably of a selection of these data; *among which are the PM mineral profiles (ii) comprising at least X differentiating mineral elements, X corresponding to - in increasing order of preference - 3, 5, 10, 15, 15, 20, 25, 30; *and relating to Sm =°( d>p>t, and / or subjects Sm( d>p>t, ; to assign the subjects Sm( d>Pjt ) to at least one target defined in S8; using a tool for reducing the dimensionality of 2D or 3D data and graphical visualization of targets; - and Position on the graphic visualization of the targets, the DSx data on the Sx subject(s) to be tested, to visualize the positioning of the Sx(s), in relation to the targets; to assist in the diagnosis, prognosis and / or therapeutic monitoring of diseases, in particular neurological diseases, for the Sx subject(s); and / or, -S9.2- Implement a 2 — protocol consisting of: - Implementation of at least one machine learning and AI prediction model; - Learning by this(these) model(s) in the BDR to choose at least one model and / or a selection of data, *among which are the mineral profiles PM (ii) comprising at least X differentiating mineral elements, X corresponding to - in increasing order of preference - 3, 5, 10, 15, 15, 20, 25, 30, these X differentiating mineral elements being included among the most effective for predicting the belonging of the subject Sx to be tested, to at least one target defined in S 8; - Possibly optimization of the parameters of the chosen model(s); - Use of the chosen model(s) to predict the belonging of the subject Sx to at least one target defined in S8 and to help in the diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, for the subject Sx to be tested.

[0022] It is to the credit of the inventors to have proposed, among other things, to use the information (i) mineral profile of the biological fluid of the subjects to be tested, based on a judicious selection of mineral elements and / or on a minimum of mineral elements, and, preferably, the information (ii) conventional clinical data, to feed automatic learning, of the statistical processing type combined with visualization (1st protocol) and / or of the AI ​​type (2nd protocol), to provide assistance with medical decision-making in terms of diagnosis (d) and / or prognosis (p) and / or therapeutic monitoring (t) [effectiveness of a therapy]. The 1st and 2nd protocols of the method according to the invention also have the particularity of being conducted on the basis of medical questions: * defining at least 2 categories and / or subdivisions of at least one category; * and requiring a prediction as to the membership of a subject Sx to be tested in at least 2 categories and / or subdivisions of at least one category, of a population of sick subjects Sm>l( d>p>t ^t / or of healthy subjects Sm =0( d>p>t,, with a given reliability (sensitivity / specificity).

[0023] The attached [Fig.l] summarizes the method according to the invention. A subject Sx to be tested to determine a diagnosis (d), a prognosis (p) and / or therapeutic monitoring (t), has one or more samples Ex of biological fluid taken. An analysis S4 of the mineral profile of this sample of these samples is carried out. The DSx data obtained are data (i) of concentrations of mineral elements in the biological fluid. Other DSx data, which are conventional clinical data (ii), are assigned to the subject Sx. Questions are asked by the doctor concerning the subject Sx to be tested. The answers to these questions consist of whether or not the subject Sx to be tested belongs to at least one target, each target comprising at least one category of diagnosis (d), prognosis (p) and / or therapeutic monitoring (t), and, possibly at least one subdivision of at least one category.The system for implementing the method according to the invention comprises an electronic device described below and shown in the attached [Fig.24], a reference database BDR comprising (i) PM and (ii) conventional clinical data on an entire population of sick subjects Sm>l( d>p >t ^t / or healthy subjects Sm =°( d>Pjt, of reference, as well as at least one terminal for communication with the electronic device. A user defines, on the basis of the questions asked, at least 2 targets to make a prediction of belonging of the subject Sx to be tested to at least one target. A 1st statistical protocol and / or a 2nd IA protocol is launched by the electronic device, at the user's command, to make the expected prediction, with a sensitivity and / or a specificity defined by the user. This prediction makes it possible to answer the questions asked by the doctor and to help the latter to make a diagnosis, a prognosis and / or a therapeutic follow-up.

[0024] According to a remarkable embodiment of the invention, * the biological fluid is chosen from the group comprising - advantageously consisting of - cerebrospinal fluid, whole blood, blood plasma, urine, follicular fluid, bile. * and diseases impacting the mineral profile of the biological fluid, include: . . in the case of cerebrospinal fluid, diseases affecting the central nervous system, in particular neurological diseases, such as psychiatric diseases, neurodegenerative diseases such as Parkinson's or Alzheimer's, . in the case of blood plasma, cancers, neurological diseases, diseases related to fertility, cardiac arrhythmias, . in the case of urine, diseases related to surgical complications following complex operations, particularly cardiovascular, . in the case of follicular fluids, diseases linked to endometriosis or diseases causing fertility problems and requiring in particular procreation aids.

[0025] Advantageously, the BDR comprises data relating to healthy reference subjects Sm =°( d>p >t) and / or sick reference subjects Sm>l( d>p>t) who suffer from one or more diseases impacting at least two biological fluids chosen from the group comprising - advantageously constituted by - cerebrospinal fluid, whole blood, blood plasma, urine, follicular fluid, bile.

[0026] In a particular embodiment, the method according to the invention comprises a step -SOa- of constructing the reference database (RDB) consisting of collecting data, in particular data (i) and / or (ii) and / or DSx data and storing them.

[0027] Preferably, the analysis in S4 is semi-quantitative, with inductively coupled plasma mass spectrometry, or ICP-MS (Inductively Coupled Plasma Mass Spectrometry) being preferred.

[0028] The elemental composition of a biological fluid is simple to measure and analyze quickly, in a semi-quantitative manner, for several dozen mineral elements.

[0029] Semi-quantitative analysis can, for example, be defined as follows: The concentrations of mineral elements are determined on a measurement scale ranging from 0.01 pg / L to 10 g / L, using a semi-quantitative method with a standard ranging from 20 to 40 elements from 10 to 30 pg / L, eg from 28 elements to 20 pg / L and a negative control (a blank) eg consisting of 1% nitric acid. The method thus measures the concentrations of eg 41 different chemical elements. The accuracy and robustness of the method have been validated by multiple tests.

[0030] Semi-quantitative analysis is particularly suitable for the creation of data (i) of mineral profiles, intended to be stored in the BDR and exploitable by statistical processing and / or by AI (chemometrics).

[0031] Indeed, the concentration of mineral elements in patients' biological fluids is often very variable. And if certain main minerals and trace elements elements, evolve in a well-defined domain and have a homeostasis which limits their variation gap; other elements, in particular metallic traces or ultratraces can vary over several orders of magnitude, often with concentration gaps of more than 1000,, on certain elements, between healthy subjects Sm =0 of reference and / or sick subjects Sm>1 of reference treated or not (patients). Semi-quantitative analysis makes it possible to measure sufficiently precisely and above all in a sufficiently reproducible manner, to simply and quickly feed a database and above all to expand it to more than 1000 patients.Using simplified sample preparation (e.g. simple dilution in nitric acid, without mineralization, nor use of EDTA or additives, often critical for trace elements) and a "reasonably priced" (entry-level) ICP-MS analyzer, a simple general multi-elemental calibration (without the necessary use of metered additions), it is possible to obtain a multi-mineral profile of at least X differentiating mineral elements, X corresponding to -in ascending order of preference- 3, 5,10, 15, 15, 20, 25, 30, for example, of about fifty elements, of which at least thirty can be used for a database. With a variability in the value obtained which can be less than 5% for the main elements and less than 20% for the trace elements.

[0032] It is nevertheless possible to carry out a more precise quantification of certain elements of interest. To do this, a calibration range specific to the concentrations of the elements is carried out. The semi-quantitative and quantitative results are then obtained in parallel and in a single acquisition for a sample and can be compared.

[0033] According to remarkable characteristics of the method according to the invention, the prediction of membership of the subject Sx to be tested to at least one category and possibly to at least one subdivision, according to -S8-, is part of a diagnosis of diseases impacting the mineral composition of at least one biological fluid, for the subject Sx to be tested, and the BDR comprises data relating to a number Ns of healthy subjects Sm =°( d>p>t, of reference and a number Nm of sick subjects Sm>l( d,P,t )of reference; with [Ns / (Ns + Nm)] *100 greater than or equal to -according to an increasing order of preference- 20%, 30%, 40%.

[0034] Advantageously, the 1st protocol -S9.1- comprises the following steps: -S9.1.1- Collection of data in the BDR; -S9.1.2- Data standardization; -S9.1.3- Selection from data (i), and possibly data (ii), of differentiating data; -S9.1.4- Reduction of the dimensionality of 2D or 3D data, preferably by a PCA principal component analysis method and / or by a t-SNE method; -S9.1.5-Visualization of targets on graphs; -S9.1.6-Positioning on the graphic visualization of the targets, the DSx data on the Sx subject(s) to be tested, to visualize the positioning of the Sx(s), in relation to the targets.

[0035] The targets are established from the BDR data collected, normalized, selected and statistically processed in -S.9.1.4-.

[0036] According to variants, step -S9.1.3- consists of using: * [Variant VI of the 1st protocol] “volcano plot” curves and / or * [Variant V2 of the 1st protocol] a selection algorithm, preferably a recursive selection algorithm “recursive features elimination-RFE”, possibly associated with a machine learning and AI prediction model chosen from a group of machine learning and AI prediction models, by implementing the following sub-steps: -S9.1.3.1- Definition of models; -S9.1.3.2- Separation of data; -S9.1.3.3- Learning and evaluation loop; -S9.1.3.4- Plotting the ROC curve for each model and calculating the average AUC (AUC score); -S9.1.3.5- choice of machine learning and AI prediction model based on the best AUC score.

[0037] Advantageously, the 2 — protocol in S9.2 comprises the following steps: -S9.2.1- Collection in the BDR of the values ​​of the training data; -S9.2.2- Normalization of training data values; -S9.2.3- Implementation of several machine learning and AI prediction models; -S9.2.4- Separation of data; -S9.2.5- Learning and evaluation loop; -S9.2.6- Plotting the ROC curve for each model and calculating the average AUC (AUC score); -S9.2.7- Selection of the machine learning and AI prediction model based on the best AUC score; preferably using volcano plots and / or the RFE recursive selection algorithm associated with the machine learning and AI prediction model selected in step -S9.2.7-. -S9.2.8- Selection of training data; -S9.2.9- Prediction of subject Sx's belonging to at least one target, using the model selected in -S9.2.7-.

[0038] Advantageously, step -S9.2.8- consists of using “volcano plot” curves and / or the RFE recursive selection algorithm associated with the machine learning and AI prediction model selected in step -S9.2.7-.

[0039] According to remarkable modalities of the method according to the invention, *a prognosis is established consisting of evaluating the risks of clinical complications linked to a renal problem, a pulmonary problem, an infection, and / or the risks of lethal complications, for an Sx subject to be tested, having undergone a surgical operation, in particular a surgical operation in which a long clamping was implemented, for example of more than 1 hour and / or blood transfusions and / or extracorporeal blood circulation; this prognosis constituting an aid to the medical decision of whether or not to keep the Sx subject in intensive care; and / or *a therapeutic monitoring is established consisting of evaluating during a drug treatment of an Sx subject to be tested, the residual level of drug circulating after its administration, on one or more occasions, to the Sx subject to be tested, for one or more durations after the administration or administrations, in particular in oncology;this therapeutic monitoring constituting an aid to the medical decision to modify or not the dose of medication to be administered to the subject Sx to be tested; and / or; * therapeutic monitoring is established consisting of evaluating, during radiotherapy treatment of a cancerous subject Sx to be tested, the clinical response of the tumors and / or the occurrence of clinical complications, in particular necrosis of healthy tissue and / or enteritis, this therapeutic monitoring constituting an aid to the medical decision to modify or not the irradiation doses of the subject Sx to be tested; and / or * a prognosis is established consisting of evaluating, within the framework of medical procreation assistance, for an Sx subject to be tested, the success rate of oocyte punctures, and / or fertilizations and / or embryo transfers and / or the occurrence of a pregnancy; this prognosis constituting an aid in the decision of whether or not to freeze oocytes and / or whether or not to transfer oocytes and / or the decision as to the number of embryos to transfer.

[0040] DEVICE:

[0041] According to a second of its aspects, the invention relates to an electronic device for implementing the method according to the invention, characterized in that it comprises: * modules for implementing the 1st protocol -S9.1-: -M9.1.1- Data collection in the BDR -M9.1.2- Data standardization; -M9.1.3- Selection from data (i), and possibly data (ii), of differentiating data; -M9.1.4- Reduction of the dimensionality of 2D or 3D data, preferably by a PCA principal component analysis method and / or by a t-SNE (t-distributed Stochastic Neighbor Embedding) method; -M9.1.5-Visualization of targets on graphs; -M9.1.6-Positioning on the graphic visualization of the targets the DSx data on the Sx subject(s) to be tested, to visualize the positioning of the Sx(s), in relation to the targets. * implementation modules of the 2nd protocol -S9.2-: -M9.2.1- Collection of training data values ​​in the BDR -M9.2.2- Normalization of training data values -M9.2.3- Implementation of multiple machine learning and AI prediction models -M9.2.4- Data Separation -M9.2.5- Learning and evaluation loop -M9.2.6- Plotting the ROC curve -M9.2.7- Machine learning and AI prediction model selection -M9.2.8- Training data / feature selection -M9.2.9- Prediction of the target of the subject Sx using the model selected in -S9.2.7-

[0042] [Fig.24] shows an example of a simplified architecture of an electronic device capable of carrying out all or part of the method according to the invention as presented previously. This electronic device comprises an electronic assembly comprising a memory 41, a processing unit 42 equipped for example with a microprocessor, and controlled by at least one computer program 43.

[0043] In one embodiment, the method according to the invention is controlled by programs installed in part or in full on the electronic device.

[0044] In another embodiment, the method according to the invention is controlled by a dedicated component (CpX) capable of processing data from the processing unit 42 or other processing units and installed in part or in full on the electronic device.

[0045] Furthermore, the device also comprises communication means (CIE) presented for example in the form of network components (Wi-Fi, 3G / 4G / 5G, wired, RFID / NFC, Bluetooth, BLE, LPWan, VLC, etc.) which allow the device to receive data (I) from entities connected to one or more communication networks and to transmit processed data (T) to such entities. These entities may be terminals allowing one or more users to communicate with the device.

[0046] According to one possibility, the device has the architecture of a computer. It is equipped with one or more processors capable of executing all types of computer programs, from operating systems to application software, written in compiled or interpreted languages. The different components of the device are connected to each other by a communication bus. The device may optionally be equipped with a communication system to communicate via protocols such as Bluetooth, Ethernet or WiFi with other systems and to connect to mobile or non-mobile telecommunications networks. The device also includes memory components which will record the data and programs necessary for the operation of the device.

[0047] SYSTEM

[0048] According to a third of its aspects, the invention relates to a system (rectangle to the right of the diagram in [Fig.l]) comprising the electronic device according to the invention, at least one computer server hosting this electronic device, the BDR and at least one terminal allowing at least one user to communicate with the electronic device, in particular to transmit data DSx relating to a subject Sx to be tested and of the same type as the data (i) & (ii) defined in S0 in the method according to the invention, and to receive in return information, relating to the subject Sx to be tested, to aid in the diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid.

[0049] COMPUTER PROGRAM

[0050] According to a fourth of its aspects, the invention relates to at least one computer program comprising instructions for implementing the method according to the invention, when said instructions are executed by a processor of a computer processing circuit. [0051 ] DATA SUPPORT

[0052] According to a fifth of its aspects, the invention relates to a data medium on which at least one computer program according to the invention is recorded.

[0053] The data carriers may be any entity or device capable of storing the programs. For example, the carriers may comprise a storage means, such as a ROM, for example a CD-ROM or a microelectronic circuit ROM, or a magnetic recording means such as a hard disk, or more often a Flash memory. In addition, the carriers may be transmissible carriers such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means. The programs according to the invention may in particular be downloaded from a network such as the Internet. Alternatively, the information carrier may be an integrated circuit in which the program is incorporated, the circuit being adapted to carry out or to be used in carrying out the method in question. Description of figures

[0054] The attached figures illustrate non-limiting exemplary embodiments, and in which: Fig.l

[0055] [Fig.l] [Fig.l] is a general diagram presenting the method according to the invention. Fig. 2

[0056] [Fig.2] [Fig.2] is a curve giving the distribution of concentrations (pg / L) of 31 mineral elements in all cerebrospinal fluids (CSF) of Example 1. Fig. 3

[0057] [Fig.3] Figures 3A, 3B, 3C, 3D, 3E, 3F attached represent the volcano plots produced in example 1, showing the impact of the mineral concentrations in the CSFs of healthy subjects Sm "° or sick subjects Sm>l as reference. Fig. 4

[0058] [Fig.4] Figures 4A, 4B attached represent the ROC curves of the IA models used in example 1. Fig. 5

[0059] [Fig.5] Figures 5A, 5B, 5C attached illustrate, in example 1, in the form of two-dimensional graphs, the distributions between the different targets: Cl comprising the category Sm "°(d5) of healthy subjects free from neurological diseases, compared to the target C2 comprising the category / subcategories Sm>l(di 44 4.24.3) of subjects suffering from neurodegenerative diseases (figure 5A), C2 compared to the target C3 comprising the category / subcategories Sm>l(d2;2.1;2.2;2.3) of subjects suffering from inflammatory diseases of the CNS (figure 5B), and finally a combined comparison Cl / C2 / C3 (figure 5C). Fig. 6

[0060] [Fig.6] Figures 6A, 6B, 6C attached represent Figures 5A, 5B, 5C with a red cross indicating the positioning of the sample ID 5116 of CSF of the subject Sx to be tested, according to his mineral profile [DSx data (i)] and the conventional clinical data DSx (i): age. Fig. 7

[0061] [Fig.7] Figures 7A (Cl / C2), 7B (C2 / C2) attached represent the confusion matrices for the machine learning and AI prediction model selected in Example 2. Fig. 8

[0062] [Fig-8] [Fig.8] is a curve giving the distribution of concentrations (pg / L) of 34 mineral elements in all the plasma samples of Example 3. Fig. 9

[0063] [Fig.9] The attached [Fig.9] represents the ROC curves of the AI ​​models used in example 3. Fig. 10

[0064] [Fig. 10] The attached [Fig. 10] represents the confusion matrix for the machine learning and AI prediction model selected in Example 3. Fig. 11

[0065] [Fig. 11] [Fig. 11] is a set of graphs giving the distribution of concentrations (pg / L) of 37 mineral elements in all the urine samples of Example 4. Fig. 12

[0066] [Fig. 12] The attached [Fig. 12] represents the volcano plot produced in the first case of example 4: occurrence of renal failure during the stay, showing the impact of the mineral concentrations in the urine of the sick subjects Sm>l(p) of reference in prognosis. Fig. 13

[0067] [Fig. 13] The attached [Fig. 13] represents the volcano plot produced in the 2nd case of example 4: occurrence of respiratory failure during the stay (lowest P / F in the 5 days <200), showing the impact of the mineral concentrations in the urine of the sick subjects Sm>l(p) of reference in prognosis. Fig. 14

[0068] [Fig. 14] The attached [Fig. 14] represents the volcano plot produced in the 3rd case of example 4: occurrence of an infection during resuscitation, showing the impact of the mineral concentrations in the urine of sick subjects Sm>l(p) of reference in prognosis. Fig. 15

[0069] [Fig. 15] The attached [Fig. 15] represents the volcano plot produced in the 4th scenario of example 4: occurrence of at least one of the three complications according to the first 3 scenarios, showing the impact of the mineral concentrations in the urine of the sick subjects Sm>l(p) of reference in prognosis. Fig. 16

[0070] [Fig. 16] The attached [Fig. 16] represents in the form of two-dimensional graphs, the distributions between the targets C5** / C6** of the example, with a red cross indicating the positioning of the sample [Urine2023_CRB002 of the subject Sx to be tested, according to his mineral profile [DSx data (i)] and the Dsx data (ii) including in particular the weight and age. Fig. 17

[0071] [Fig. 17] The attached [Fig. 17] represents the ROC curves of the AI ​​models used in Example 4. Fig. 18

[0072] [Fig. 18] The attached [Fig. 18] represents the confusion matrix for the machine learning and AI prediction model selected in Example 4. Fig. 19

[0073] [Fig. 19] The attached [Fig. 19] is a set of graphs giving the distribution of concentrations (pg / L) of 31 mineral elements in all the follicular fluid samples of Example 5. Fig. 20

[0074] [Fig.20] The attached [Fig.20] represents the volcano plot made in the 1st case of figure of example 5: endometriosis Cl***: Category Sm>l(di) subcategory Sm>l(di.i)endometriosis vs control (linked to a male infertility problem) C2***: Category Sm>l(di) subcategory Sm>l(di ^reference, showing the impact of mineral concentrations in the follicular fluids of the reference Sm>l(p) diseased subjects on prognosis. Fig. 21

[0075] [Fig.21] The attached [Fig.21] illustrates, in example 5, in the form of graphs two-dimensional, the distributions between the targets ci*** / C2*** of example 5, with a red cross indicating the positioning of the sample [follicular fluid CHUL4] of the subject Sx to be tested, according to his mineral profile [DSx data (i)] and the Dsx data (ii) including in particular age. Fig. 22

[0076] [Fig.22] The attached [Fig.22] represents the ROC curves of the AI ​​models used in example 5. Fig. 23

[0077] [Fig.23] The attached [Fig.23] represents the confusion matrix for the model machine learning and AI prediction selected in Example 5. Fig. 24

[0078] [Fig.24] The attached [Fig.24] is a schematic representation of the electronic device for implementing the method according to the invention. EXAMPLES

[0079] Example 1: Diagnosis of a subject S 21 to be tested through comparative statistical analysis _( 1a protocol - S9 - for implementing the method according to the invention ). on the one hand, historical learning data (i) & (ii) originating in particular from CSF samples from a population of healthy reference subjects S “ — or sick Sm> 1 - and, on the other hand, data (i) & (ii) originating in particular from CSF samples from this subject S* to be tested

[0080] - S0- Implementation of a database [0081 ] -S0.1- Implementation of a learning database consisting of conventional clinical data (ii)

[0082] A database is constructed from 1956 CSF samples from more than 2000 reference subjects who consulted the Hamburg hospital in Germany, for suspected neurological disease(s).

[0083] Of these more than 2000 reference subjects, a number Ns = 515 are healthy reference subjects Sm = 0 ( d,P ,t ) not having a specific neurological disease and a number Nm = 781 are sick reference subjects Sm>l ( d>p>t ), suffering from at least one identified neurological disease. 660 subjects who consulted could not have a precise diagnosis and cannot be considered as sick reference subjects Sm>l ( d>p >t ) or healthy reference subjects Sm = 0 ( d>Pjt ).

[0084] The total number N* of reference subjects Sm =°( d>p >t, and Sm>l( d>Pjt ) is therefore 1296. [Ns / (Ns+ Nm)] *100 = [515 / 515+781)] *100 = 39.7%.

[0085] Each of the healthy reference subjects Sm "° or sick Sm>l suffering from at least one neurological disease was assigned to one or more specific diagnostic targets or categories, and possibly sub-categories or sub-targets Sm(d), by the medical team of the Hamburg hospital, via a review of the conventional clinical data concerning the subjects. These categories / sub-categories are "targets" / "sub-targets" or "targets" / "sub-targets", within the framework of the decision-making process involving statistical representations or artificial intelligence, specific to the method according to the invention for diagnosis, prognosis and / or therapeutic monitoring of diseases, in particular neurological diseases, impacting the mineral composition of the cerebrospinal fluid.

[0086] For the 1296 healthy subjects Sm =0( d>Pjt ,or sick Sm>l( d>Pjt )reference subjects of this example, the specific diagnostic categories Sm(d) are five in number, namely: (dl) sick subjects Sm>l( d>Pjt suffering from at least one disease neurodegenerative disease, (d2) sick subjects Sm>l( d>p>t suffering from at least one inflammatory disease of the CNS (d3) sick subjects Sm>l( d>p>t suffering from at least one psychiatric disorder, (d4) sick subjects Sm>l( d>pt suffering from at least one other neurological disease and (d5) healthy subjects Sm "°( d>Pjt not suffering from a specific neurological disease: premium control group.

[0087] Some subcategories are also defined. Table 1 below shows all the diagnostic categories and subcategories assigned to the 1296 healthy Sm "°( d>p ,t )Or sick Sm>l( d>Pjt jreference subjects of this example, by the medical team of the Hamburg hospital, as well as the number of corresponding subjects. Among the reference subjects, 367 Sm(di)present a neurodegenerative disease, 291 Sm(d2)have an inflammatory disease of the CNS, 123 Sm(d3)suffer from psychiatric disorders, 317 Sm(d4)are diagnosed with another neurological disease generating abnormalities in the CSF. Among the study population, 515 Sm "0(d5)healthy subjects are classified as a premium control group, showing no symptoms of neurological disorder and without abnormalities detected in the CSF.In addition, the study population includes 660 subjects with non-specific neurological signs, for whom no precise diagnosis could be established and where no known CSF alterations were observed. This latter group is designated as GNI (Unidentified Group), and the corresponding subjects are designated as Sm(d4,2).

[0088] [Table 1] Distribution of CSF samples from the BDR N* sn N*' ; * '<> -<3 ■> >&hü"î'' OH* L...........: dégmêrâBw ; «H 367 ;•$ : 1 >2 O ks 62 Matos dégénérons (4&ww> AJ / Ne: n ta r ' D 4 rn fer : r-ss Whdigs RsfMrwrL sÿndrL 6®rpms FsrkïJWPi 'psr ■fexe^pse Ataxie rp«^x:&rébo:feu%& Svfêïow =0^^ 177 192. 3® {dïj kfe SN€ 281 M î M 146 ^:to^fpaï<î70<.<iu $NG a u> pa-tK^ne > Wfapfsç 733⁄4 îs'sOisiri^feîi, 7?s ferv'os.2rd:îo^, ^s> rad«xî$s< îhm par axsmpfe a» yzv?' ^hrfi-'nah'^feS du SNC <î <s:JOs<isa an '’æP', y fearffpm s^di'esvî^ MMaèm (nflamnak^s du SNC Jxu'sa ps? (sans SEP) y rom^m n*t»-*upathm a^to S3® 63 123 : 123 TrwStet (WùS^s ^$ydwass dèHodancog) putrss ] no oe> is | "sMC i 977 kt 4.2 1 ? 669 Auïm fwnsksgk^® (smwWîl&m U^| Srmp®: Non Msr^- (GNÇ S SIS ! SI 6 pas 4«? lC^ npîr 'gL

[0089] -S0.2- Implementation of a learning database (i) constituted by the P Mineral profile (PM)

[0090] The database is completed for the 1956 healthy subjects Sm "°(d5) or sick Sm>l(di.i ;i.2;i.3;i.4;2.i ;2.2 ;2.3 ;3 ;4.i) of reference or of the Group of unidentified (GNI) Sm( d4.2) of this example, by the mineral profile PM of concentrations of Nem = 31 mineral elements established for the corresponding CSF samples.

[0091] These CSF samples are stored in the freezer, at a temperature generally below -15°C, and thawed just before preparation for analysis.

[0092] Before being analyzed by inductively coupled plasma mass spectrometry, or ICP-MS, CSF samples are diluted 25 times in high purity 1% nitric acid with indium (In) as an internal standard, according to the following protocol: 100 pL of CSF in 2400 pL of 1% HNO3 with 1.6 ppb of In.

[0093] Nitric acid allows partial, but sufficient, mineralization of the CSF. It helps dissolve and partially degrade proteins, lipids, and other components of CSF to prepare a homogeneous sample. It stabilizes the sample by preventing precipitation or adsorption on the container walls, thus ensuring a homogeneous distribution of elements during analysis. The acid also helps minimize potential interferences during ICP-MS analysis by keeping the elements in a stable ionic form.

[0094] A dilution factor of twenty-five is used to analyze major and trace elements in a single analysis.

[0095] Due to the high sensitivity of mass spectrometry, the volume of CSF required for analysis is extremely low (less than 100 pL), compared to other analytical methods requiring: Between 0.5 and 1 mL for biochemical analyses (biomarkers), Between 1 and 2 mL for cytological analyses (cells), Between 0.5 and 1 mL for genomic analyses (DNA or RNA).

[0096] Analysis by ICP-MS

[0097] ICP-MS analyses are performed using an ICP-MS mass spectrometer. Agilent 7850 Series MS with quadrupole analyzer and integrated autosampler.

[0098] The analyses are carried out, in semi-quantitative mode, with the following conditions: Standard 28 elements at 20 ppb_Blank: HN03 l%_Sample collection time: 50 s_Stabilization time: 30 s Acquisition time: 216 s_Post-acquisition rinsing: 3 x 45 s_Total analysis time: 431 s.

[0099] Semi-quantitative analysis involves estimating the relative concentrations of elements without resorting to the rigorous calibration essential to quantitative analysis. It proves particularly useful for quickly obtaining the elemental composition of samples.

[0100] It is nevertheless possible to carry out a more precise quantification of certain elements of interest. To do this, a calibration range specific to the concentrations of the elements is carried out. The semi-quantitative and quantitative results are then obtained in parallel and in a single acquisition for a sample and can be compared.

[0101] For each analysis series, the following protocol is used: Systematic cleaning of ICP-MS supply lines to avoid any contamination and ensure good analysis of trace elements in particular. Systematic verification of device performance. Verification of the semi-quantitative model through the analysis of a reference wine at the beginning, middle and end of each analysis sequence. Monitoring of analyses through regular analysis of the 28-element standard throughout the sequence. Monitoring of the concentration of the internal indium standard to prevent equipment-related drifts.

[0102] [Fig.2] shows the distribution of concentrations [pg / L] of 31 mineral elements in all CSF samples from the BDR.

[0103] -SL

[0104] A sample (stored at -18°C -S3-) from an Sx subject whose diagnosis is still undetermined within the GNI group according to Table 1 above, is used.

[0105] -S2-

[0106] The following Dsx data are assigned: the ID identification number and the conventional clinical data (ii): gender and age of Sx corresponding to this sample.

[0107] -S4- _The PM of this sample is analyzed as described above for the (NSm =0( d,p ,t ) + NSm>l( d>p>t,) 1296 healthy subjects Sm ~°kH or sick Sm>l(di.i ;i.2;i.3 ;i.4 ;2.i ;2.2;2.3;3;4.i) of reference of this example.

[0108] The detailed results are presented in Table 2 below, specifying the concentrations of minerals measured in ppb, in this sample corresponding to the subject Sx to be tested as well.

[0109] [Table 2] Concentration in ppb of the different minerals ID sm B 103.926 Na 3217183 Mg 23304 Al 51.83 P 16044 S 59726 Cl 5358740 K 103890 32024 Ti 2.4 > a Cr 0.267 Mn 0.464 Fe 26.75 Ce 0 Cu 5.9© Zn 25.08 Br 1584 Rb 55.2 Sr w.œ Zr 0 Sn 0.472 [ 9.37 Cs 0.15 Ba 0. La 0 Ce 0 Gd 0 Tb 3.09 Pb 0.43 Genus f

[0110] -S8- Define at least two targets each comprising at least one category and / or at least one subdivision (eg subcategory) of at least one category, to make at least one prediction of membership of the subject Sx to be tested to at least one target.

[0111] In this example 1, we define 3 categories, 3 subcategories of one of the categories and 3 subcategories of the other category. These 3 categories and these 6 subcategories are targets with respect to which the method makes it possible to predict whether Sx belongs to one or more of them.

[0112] The membership of Sx to none, one or more targets is not predefined. Depending on the data and the categories / subcategories chosen, Sx can belong to 0, 1 or more categories / subcategories. The physician can participate in defining the targets (categories / subcategories) of interest to aim at on the subject Sx.

[0113] In this example, we predict that Sx belongs to none, one or more of the following targets: Cl: Category Sm "°(d5) healthy subjects free from neurological diseases: premium controls C2: Category_ Sm>l(di) subjects suffering from neurodegenerative diseases and subcategories: Sm>l(du, Alzheimer's dementia; Sm>l(di 2) non-Alzheimer's dementia; Sm>l(di.3) Parkinson's (disease / syndrome); Sm>l(di.4) other neurodegenerative diseases C3: Category- Sm>l(d2) subjects suffering from inflammatory diseases of the CNS and subcategories Sm>l(d2.i) inflammatory diseases of the CNS caused by a pathogen, including meninges, encephalitis, radiculitis, due for example to VZV; Sm>l(d22) inflammatory diseases of the CNS multiple sclerosis -MS-, including CIS; Sm>l(d23) inflammatory diseases of the CNS not caused by the pathogen, (without MS), including immune neuropathies, autoimmune encephalitis, myasthenia gravis. C4: Subcategory_ Sm>l(d4.i) subjects suffering from other neurological diseases.

[0114] -S9.1- Statistical processing;

[0115] -S9.1.1^ Collection of training data used: *Data (i) concentrations of mineral elements *Conventional clinical data (ü): ​​age; gender: 0 for male and 1 for female

[0116] -S9.1.2- Normalization of training data values ​​using StandardScaler

[0117] -S9.1.3- [Variant VI of the 1st protocol] Selection of training data by using “Volcano Plot” curves:

[0118] For the selection of the most relevant, i.e. most differentiating, data (i) concentrations of mineral elements and conventional clinical data (ii), we use in this example, the Volcano Plot feature selection tool for the PM of the CSF of healthy subjects Sm "0(d5) or sick subjects Sm>l(di.i ;i,2 ;i.3 ;i.4 ;2.i ;2.2 ;2.3 ;3 ;4.i) of reference of this example.

[0119] The impact of the concentration of each measured mineral element is studied for each target category Cl, C2, C3 or C4, for example the categories / subcategories corresponding to the subjects Smd i,d 2 or d3,oud4.i- The so-called "volcano plot" curves represent the statistical impact of each metallic element by comparing the concentrations measured for this element for each individual of a category compared to other categories. This type of graph allows to quickly and efficiently visualize the important differences between two categories / subcategories or targets / subtargets. On a "volcano plot", the abscissa axis (horizontal axis) generally represents the degree of change, while the ordinate axis (vertical axis) indicates the statistical significance of this change, is expressed as P value.The points at the upper ends of the graphs indicate the elements with the most marked and statistically significant differences. The curve represents the P value as a function of the impact of the concentration of that element. P values ​​can be calculated using the spearmanr, mannwhitneyu or pearsonr functions from the scypi.stats library on python3.

[0120] Thus the attached figures 3A, 3B, 3C, 3D, 3E, 3F represent volcano plots which compare: - either the concentrations of mineral elements in the CSF of healthy subjects Sm "°d5 of reference Cl, to the concentrations of mineral elements in the CSF of sick subjects Sm>l of reference suffering from neurological diseases C2; - or, between them, the concentrations of mineral elements in the CSF of sick subjects Sm>l of reference suffering from different neurological diseases C2,C3; - [Fig.3]A the concentrations of mineral elements in the CSF of healthy subjects Sm "°d5 of reference Cl [Cl_category d5], to the concentrations of mineral elements in the CSF of sick subjects Sm>ldi.2 of reference C2 (non-Alzheimer's dementia) [C2_subcategory dl.2]; - [Fig.3]B the concentrations of mineral elements in the CSF of healthy subjects Sm "°d5 of reference Cl [Cl_category d5], to the concentrations of mineral elements in the CSF of sick subjects Sm>ldi.3 of reference C2 (Parkinson's disease, Parkinson's syndrome) [C2_subcategory dl.3]; - [Fig.3]C the concentrations of mineral elements in the CSF of healthy subjects Sm "°d5 of reference Cl [Cl_category d5], to the concentrations of mineral elements in the CSF of sick subjects Sm>ldide reference C2 (neurodegenerative diseases) [C2_category dl]; - [Fig.3]D the concentrations of mineral elements in the CSF of healthy subjects Sm "°d5 of reference Cl [Cl_category d5], to the concentrations of mineral elements in the CSF of sick subjects Sm>ld2 of reference C3 (inflammatory diseases of the CNS), [C3_category d2] - [Fig.3]E the concentrations of mineral elements in the CSF of healthy subjects Sm "°d5 of reference Cl [Cl_category d5], to the concentrations of mineral elements in the CSF of sick subjects Sm>ldi.ide reference C2 (Alzheimer's dementia) [C2_subcategory dl.l]; - [Fig.3]F the concentrations of mineral elements in the CSF of sick subjects Sm>ldi of reference C2 (neurodegenerative diseases) [C2_category dl], to the concentrations of mineral elements in the CSF of sick subjects Smd2 of reference C3 (inflammatory diseases of the CNS) [C3_category d2].

[0121] For each of the categories identified in relation to the controls, we see that certain mineral elements have differentiating roles of different intensity and direction, that is to say that the concentration of a given element is generally influenced by the given category. This influence can be in a positive direction (the more there is of this element, the greater the probability of belonging to this category) or negative (the more there is of this element, the less the probability of belonging to the category). This link can also be of variable intensity in absolute value, for a link of high intensity. This means that a small variation in the concentration of an element will strongly influence the probability of belonging or not to the given category. These differentiating roles are not the same depending on the categories.The dotted red line in Figures 2A, 2B, 2C, 2D, 2E, 2F represents the limit corresponding to a P Value of 0.05; commonly considered a significant limit for an individual value.

[0122] Figure 3A: Dementia (excluding Alzheimer's dementia) [C2_subcategory dl.2] versus premium controls [Cl_category d5]. Differentiating minerals identified as significantly associated with dementia include Cu, Fe, P, Sr, Zn, Cr, S, Al, Pb, and are selected based on their correlation strength according to Mann-Whitney U statistical analysis, while Gd, Mg and Tb have a negative effect on the course or prevalence of dementia, as evidenced by the negative correlation values ​​obtained and the corresponding positions on the volcano plot.

[0123] Figure 3B: Parkinson's disease and Parkinson's syndrome [C2_subcategory dl.3] compared to premium controls [Cl_category d5]. The differentiating minerals identified as significantly associated with Parkinson's disease include Cu, Fe, P, S, Al, Cr, V, Zn and are selected based on their correlation strength according to Mann-Whitney U statistical analysis, while Mg and Na, have a negative effect on the progression or prevalence of Parkinson's disease, as evidenced by the negative correlation values ​​obtained and the corresponding positions on the volcano plot.

[0124] Figure 3C: Neurodegenerative diseases [C2_category dl] versus premium controls [Cl_category d5]. Differentiating minerals identified as significantly associated with neurodegenerative diseases include Cu, Fe, P, Sr, Zn, Cr, S, Al, Pb, and are selected based on their correlation strength according to Spearman statistical analysis, while Gd, Mg, Tb, Cl and Na have a negative effect on the progression or prevalence of neurodegenerative diseases, as evidenced by the negative correlation values ​​obtained and the corresponding positions on the volcano plot.

[0125] Figure 3D: Inflammatory CNS diseases [C3_category d2] compared to premium controls [Cl_category d5]. The differentiating minerals identified as significantly associated with this type of diseases include Cu, Zn, S, P, K, Fe, Rb, Cs, Sr, and are selected based on their correlation strength, according to the Mann-Whitney U statistical analysis, while Mg and Cl have a negative effect on the evolution or prevalence of this type of disease, as evidenced by the negative correlation values ​​obtained and the corresponding positions on the volcano plot.

[0126] Figure 3E: Alzheimer's dementia [C2_subcategory dl.l] versus premium controls [Cl_category d5]. Differentiating minerals identified as significantly associated with this type of diseases include Cu, Fe, Cr, Ni, Zn, Cs, and are selected based on their correlation strength according to Mann-Whitney U statistical analysis, while Mg, Tb, Na, Gd, Cr have a negative effect on the evolution or prevalence of this type of disease, as evidenced by the negative correlation values ​​obtained and the corresponding positions on the volcano plot.

[0127] Figure 3F: Neurodegenerative diseases [C2_category dl] versus inflammatory CNS diseases [C3_category d2]. Differentiating minerals identified as significantly associated with neurodegenerative diseases and negatively correlated with inflammatory diseases include Fe, Sr, Cr, Al, and are selected based on their correlation strength according to Spearman's statistical analysis, and Gd, Mg, Tb, Cl and Na are identified as significantly associated with inflammatory diseases and negatively correlated with neurodegenerative diseases.

[0128] The correlations between the different data (ii) conventional clinical: age; gender: 0 for male and 1 for female; and the different targets are analyzed. Age is not displayed on the different volcano plots, because its correlation with certain pathologies is too important and overwhelms the other information of the volcano plot. We also note that for figures 3A, 3B, 3C and 3D, gender has a significant effect on the evolution or prevalence of certain diseases.

[0129] The differentiating minerals selected from Figures 3A to 3F allow to better clarify the relationship between various minerals and the associated diseases. They are not necessarily selected for the statistical processing -S9.1- and the IA calculation -S9.2- described below in the example, the choice of minerals is made using an RFE model associated with the calculation model used.

[0130] -S9.1.3- [V2 variant of the 1st protocol] This variant V2 is tested on three scenarios of definition, according to step -S8-, of the targets used for the prediction of membership of Sx to one of these targets, which each comprise at least one category and / or at least one subdivision of at least one category. These 3 scenarios correspond respectively to: - 1st scenario: Cl Sm targets "°(d5) healthy subjects free from neurological diseases [Cl_category d5]: premium controls & C2 Sm>l(di ;ii ;i,2;i.3> subjects suffering from neurodegenerative diseases[C2_category dl]; - 2nd case: targets C2 Sm>l(di ;ii ;i,2;i.3> subjects suffering from neurodegenerative diseases [C2_category dl] & C3 Sm>l(d2;2 | 2.22.3) subjects suffering from inflammatory diseases of the CNS [C3_category d2]. - 3rd case: targets C2 Sm>l(di;Li;L2;i.3) subjects suffering from neurodegenerative diseases [C2_category dl], C3 Sm>l(d22.i 2.22.3) subjects suffering from inflammatory diseases of the CNS [C3_category d2], & Cl Sm =°(d5) healthy subjects free from neurological diseases [Cl_category d5]: premium controls.

[0131] -S9.1.3.1- Definition of models A group of 8 machine learning and AI prediction models are defined as follows: Random Forest, Logistic Regression, Gradient Boosting Machine, Decision tree, Conditional Random Forest, Linear SVM, Ridge Classifier and XGBoost. These models are configured with their default parameters

[0132] -S9.1.3. 2 - Data Separation We divide the data set (i) & (ii) for the 1st and 2nd cases into training (80%) and test (20%) parts.

[0133] -S9.1.3. 3 - Learning and evaluation loop Application of 5-fold cross-validation with StratifiedKFold for model training and evaluation, calculating for each iteration the false positive rates (FPR) and true positive rates (TPR).

[0134] -S9.1.3. 4 - Plotting the ROC curve for each model and calculating the average of AUC (AUC score) The ROC curve is plotted for each model using the averages of the TPRs and FPRs, and the average AUC is calculated over the cross-validation iterations.

[0135] The ROC curves of the 2nd case for the 8 models tested are shown in Figure 4A (curves A to I).

[0136] The ROC curves of the 2nd case for the 8 models tested are shown in Figure 4B (curves A to I).

[0137] In the legend of these two figures 4A and 4B, the AUCs obtained for each model tested are shown as well as the number n corresponding to the number of categories of training data used for each model.

[0138] -S9.1.3. 5 z_ Ç choice of machine learning and AI prediction model based on best AUC score.

[0139] The best AUC score measuring the capacity of each model to differentiate the targets C2 / C1 (figure 4A) and CH C3 (figure 4B), is obtained, respectively, for the Conditional Random Forest model [targets C2 / C1 (figure 4A)] and for the Gradient Boosting Machine model [C2 / C3 (figure 4B)], which are therefore chosen.

[0140] The RFE recursive selection algorithm, combined with the chosen machine learning and AI prediction models, is used for the selection of the differentiating data (i) & (ii).

[0141] The data (i) & (ii) selected according to -S9.1.3-are: - for the 1st case C1 / C2-: . Data (i) mineral elements: B, Al, Fe, Rb, S . Conventional clinical data (ii): age - for the 2nd case C2 / C3-: . Data (i) mineral elements: P, K, Cr, Zn, Rb. . Conventional clinical data (ii): age - for the 3rd case C1 / C2 / C3-: . Data (i) mineral elements: P, S, Cl, Cu, Zn, Rb. . Conventional clinical data (ii): age These data selected by the RFE algorithm [Variant V2 of the 1st protocol], rather than by the volcano plot curves [Variant VI of the 1st protocol], are displayed and used for dimensionality reduction-S9.1.4-.

[0142] -S9.1.4- 2D or 3D stochastic dimension reduction The t-SNE (t-distributed stochastic neighbor embedding) algorithm is implemented to - reduce the dimensionality of the targeted data by compressing the multi-dimensional data into two or three principal components, - and facilitate their visualization in 2D or 3D depending on the selected data and improve the distinction between the targeted categories.

[0143] -S9.1.5-Visualization on graphs of targets (targeted data) (-(i)- concentrations of the different minerals and -(ii)- conventional clinical data) most differentiating / relevant for the performance of the process.

[0144] A KDE kernel density plot (Kernel Density Estimate) is produced to visualize the distribution of the targets, 1st case C2 / C1, 2nd case CH C3, and 3rd case C1 / C2 / C3, as defined previously. To do this, the components extracted following the application of the t-SNE algorithm are plotted respectively on the abscissa axes and the ordinate axis.

[0145] A scatter plot is also overlaid to show the individual points.

[0146] The attached Figures 5A, 5B and 5C illustrate, in the form of graphs two-dimensional, the distributions between the different combinations of targets, respectively, C2 / C1; C2 / C3; C1 / C2 / C3 defined above: - [Fig.5]A: C1 Sm "0(d5) [Cl_category d5] compared to C2 Sm>l(di ;ii ;i.2;i.3) [Fig.5]B: C2 Sm>l(dl ;ii ;i.2;i.3)[C2_category dl] compared to C3 Sm>l(d2;2.i ;2.2;2.3) [C3_category d2] ; [Fig.5]C: combined comparison of the three targets Cl Sm "°(d5) [Cl_category d5];C2 Sm>l(dl ;1.1 ;1.2;1.3) [C2_category dl] and C3 Sm>l(d2;2.i ;2.2;2.3) [C3_category d2],

[0147] Regions where the estimated density is equal to or greater than 50% of the maximum value will be displayed, with different shades or colors representing different density ranges.

[0148] The statistical processing -S9.1- allows an optimal separation of the categories, for example by t-SNE. Figures 5A / 5B / 5C represent on 2-dimensional graphs, the different targets, in order to best separate in space, the points representing the different subjects Sm "0(d5)[Cl_category d5] ; Sm>l(di ;Li ;L2;L3) [C2_category dl] Sm>l(d2;2.i ;2.2;2.3)[C3_category d2].

[0149] This visualization helps to distinguish between different targets based on selected parameters, such as concentrations of certain mineral elements.

[0150] -S9.1.6-Positioning

[0151] The location of the Dsx data (i) formed by the ppb concentrations of the different minerals (Table 2 above) and the Dsx data (ii) formed by the gender and age corresponding to the subject Sx to be tested, are integrated into the three t-SNE graphs previously established (Figures 5A / 5B / 5C), with a distinctive red cross to facilitate its identification. These are the attached figures 6A / 6B / 6C.

[0152] Figure 6A: Healthy Sm=0(d5) subjects [Cl_category d5] versus Sm>l(dl;L1;1.2;1.3) subjects with neurodegenerative diseases [C2_category dl]. Selected DSx (i) & (ii) training data: [B, Al, Fe, Rb, Sr, age]. The red cross indicates the positioning of sample ID 5116 on this map according to its mineral profile.

[0153] Figure 6B: Sm>l(dl;L1;1.2;1.3) subjects with neurodegenerative diseases [C2_category dl] compared to Sm>l(d2;2.1;2.2;2.3) subjects with CNS inflammatory disease [C3_category d2]. DSx training data (i) & (ii) selected: [P, K, Cr, Zn, Rb, age]. The red cross indicates the positioning of sample ID 5116 on this map according to its mineral profile.

[0154] Figure 6C: Healthy Sm=0(d5) subjects [Cl_category d5] compared to Sm>l(dl;L1;1.2;1.3) subjects with neurodegenerative diseases [C2_category dl] and Sm>l(d2;2.1;2.2;2.3) subjects with CNS inflammatory disease [C3_category d2. Selected characteristics: [P, S, Cl, Cu, Zn, Rb, age] the red cross indicates the positioning of sample ID 5116 on this map according to its mineral profile. On each of the t-SNE maps in Figures 5A / 5B / 5C, the red cross indicating the position of sample ID5116 of the subject Sx to be tested, is clearly located within the target (region) associated with Sm>l(dl;L1;1.2;1.3) subjects with neurodegenerative diseases. This localization strongly suggests that the PM mineral profile of the CSF sample ID5116 from subject Sx to be tested corresponds to the typical characteristics of neurodegenerative disorders, which could point towards a diagnosis in this category of diseases.

[0155] Example 2: Diagnosis of a subject S 21 to be tested by means of an automatic learning and AI prediction model (2 — _ protocol for implementing step -S9- _ of the method according to the invention), on the one hand, from historical learning data (i) & (ii) originating in particular from CSF samples from a population of healthy reference subjects S m _ and / or sick Sm>l , and, on the other hand, from data (i) & (ii) originating in particular from CSF samples from this subject S~ to be tested.

[0156] -Sl-

[0157] A sample (stored at -18°C -S3-) from subject Sx to be tested in example 1 (category GNI according to table 1 above) is used.

[0158] -S2-

[0159] The following DSx data are assigned: the ID identification number and the conventional clinical data (ii): gender and age of Sx corresponding to this sample.

[0160] -S4- The PM of this sample is analyzed as described above for the N* 1296 subjects healthy Sm "°(d5) or sick Sm> 1 (dL1; L2; i,3; i.4; 2.i; 2.2; 2.3; 3; 4.i) of reference of this example.

[0161] The detailed results are presented in Table 2 presented previously, specifying the concentrations of minerals measured in ppb, in this sample corresponding to the subject Sx to be tested.

[0162] -S8- Define at least two targets to make at least one prediction of the subject Sx to be tested belonging to at least one target

[0163] In this example 2, we take, in the same way as in example 1, the 3 targets Cl comprising the category [Sm "°(d5)], C2 comprising the category [Sm>l(di] with its 4 sub-categories Sm>l(di.i), Sm>l(di.2), Sm>l(di.3) & Sm>l(di.4) & C3 comprising the category [Sm>l(d2], with its 3 sub-categories Sm>l(d2 i), Sm>l(d22), Sm>l(d23). These 3 categories and these 6 sub-categories are targets, with respect to which the method makes it possible to predict whether Sx belongs to one or more of them.

[0164] The membership of Sx to none, one or more targets is not predefined. Depending on the data and the categories / subcategories chosen, Sx can belong to 0, 1 or more categories / subcategories. The physician can participate in defining the targets (categories / subcategories) of interest to aim at on the subject Sx.

[0165] In this example 2, the membership of Sx to none, one or more targets, is predicted according to the same first 2 cases C1 / C2 & C2 / C3 as those of example 1.

[0166] 2 — protocol _ S9,2- of _ step - S 9 - of implementing several models machine learning and AI prediction.

[0167] -S9,2.1- Collection of training data values: *Data (i) concentrations of mineral elements *Conventional clinical data (ii): age; gender: 0 for male and 1 for female.

[0168] -S9.2 ,2- Normalization of the values ​​of the training data using StandardScaler

[0169] -S9.2,3- Implementation of 7 machine learning models and AI prediction.

[0170] This repeats step - S9.1.3.1 - of example 1.

[0171] - S9.2,4- Data separation:

[0172] This repeats step - S9.1.3.2- of example 1.

[0173] - S9 . 2 ,5- Learning and evaluation loop:

[0174] This repeats step - S9.1.3.3 - of example 1.

[0175] - S9.2,6- Plotting the ROC curve:

[0176] This repeats step -S9.1.3.3- of example 1.

[0177] The AUC for the 8 models in Figure 4A (curves A to I) is given in this figure.

[0178] The AUC for the 8 models in Figure 4B (curves A to I) is given in this figure.

[0179] In the legend of these two figures 4A and 4B, n corresponds to the number of categories of training data used for each model.

[0180] By studying the PM mineral profile of the CSFs, and using machine learning and AI prediction models, we can determine the membership of the subject Sx to be tested, with certain reliability, in the Cl, C2 and / or C3 category.

[0181] -S9.2.7- Selection of machine learning and prediction model by AI (idem -S9.1.3.5-) based on the best AUC score measuring the ability of each machine learning and AI prediction model to differentiate categories / targets. The Conditional Random Forest model is selected for targets C2 / C1 (Figure 4A) and the Gradient Boosting Machine model C2 / C3 (Figure 4B), as it is the one that achieves the best AUC score in these 2 cases.

[0182] -S9. 2^8 - Selection of training data (i) & (ii):

[0183] The data to be selected are the categories / subcategories of the reference sick subjects Sm>l and / or the reference healthy subjects Sm ", the concentrations of the different minerals [data (i)] and the conventional clinical data (ii) associated with these reference subjects that are the most differentiating / relevant for the performance of the automatic learning and prediction model by AI Conditional Random Forest or Gradient Boosting Machine selected in -S9.2.7-, thanks to the RFE estimator associated with these models.

[0184] The RFE algorithm, associated with the chosen Conditional Random Forest or Gradient Boosting Machine AI prediction and machine learning model, is also used for the selection of the differentiating data (i) & (ii). These data are: - for the 1st case C1 / C2-: . Data (i) mineral elements: B, Al, S, Cl, Ni, Zn, Rb, I, Tb . Conventional clinical data (ii): age - for the 2nd case C2 / C3-: . Data (i) mineral elements: B, Na, Mg, P, Cr, Co, Cu, Rb, Tb . Conventional clinical data (ii): age These data selected by the RFE algorithm are displayed and used for training the chosen AI model.

[0185] The confusion matrices (Figures 7A & 7B attached), showing the true positives, false positives, true negatives and false negatives, are generated from the test data set (representing 20% ​​of the entire data set).

[0186] The sensitivity and specificity of the model are calculated. The hyperparameters are then adjusted in order to optimize the performance of the model.

[0187] The confusion matrices (Figures 7A & 7B attached) for this example 2 were carried out on 20% of the 815 subjects belonging to at least one of the two targets C1 / C2 or C2 / C3, i.e. 163 subjects. The average sensitivity obtained is 0.79. The average specificity obtained is 0.78. In addition, the average AUC score obtained is 0.85.

[0188] -S9. 2^9 - Prediction of membership of subject Sx using the model selected in - S9.2.7-

[0189] We implement the Conditional Random Forest or Gradient Boosting Machine prediction model on the selected data of the subject Sx and the selected data of the target categories of the reference subjects.

[0190] The prediction is displayed, with the sensitivity, specificity and AUC score of the Conditional Random Forest or Gradient Boosting Machine model for the analyzed data.

[0191] As shown in the figures: 7A: Confusion matrix target C2 subjects suffering from neurodegenerative diseases versus target Cl healthy subjects; Sensitivity: 0.7903225806451613; Specificity: 0.7821782178217822; AUC score: 0.8510060683487704 & 7B: Confusion matrix target C2 subjects suffering from neurodegenerative diseases versus target C3 subjects suffering from an inflammatory disease of the CNS; Sensitivity: 0.7903225806451613; Specificity: 0.7821782178217822; AUC score: 0.8510060683487704

[0192] The Conditional Random Forest and Gradient Boosting Machine models predict the belonging of the subject Sx to the target neurodegenerative diseases C2, reliably AUC-ROC = 85%, based on the data selected for the chosen targets.

[0193] -S9.2.9- Membership prediction of 655 Sx subjects from the subcategory d,4 (unknowns from [Table 1] above) using the model selected in -S9.2.7 -

[0194] The conventional clinical data (ii) are supplemented by the sex of the 655 subjects and by quantitative data on plasma components measured in the plasma samples of these 655 Sx subjects. [Table 3] below lists these plasma components.

[0195] [Tables3] Sodium mmol / î Alkaline phosphatase, U / l Platelets, 10 il Basophils, % Potassium, mmol / l Bilirubin (total), mg / dl VPM. Phosphate, mg / dl Chloride, mmol / f CHE, kU / i Quick % Cholesterol, mg / dl Calcium, mmol / l Alpha-amylase, U / l iNR LDL-Cholesterol, mg / dl Creatinine, mg / dl Pancreatic amylase, U / l TTP, s HDL-Cholesterol, mg / dl Krea-GFR (CKD), M'min Lipase, U / l Thrombin time, s Triglycerides, mg / dl Urea, mg / dl LDH, U / l Fibrinogen, mg / dl Normobiasts, / 100 Leu Uric acid, mg;di Troponin T. pg / ml GRP, mg / 1 Albumin, % Glucose, mg / dl Leukocytes, 10*9 / 1 IgG (Serum), g / dl Globulin, alpha. % HbAlc, % Erythrocytes. W4ÙI IgA (Serum), mg / dl Globulin, beta, % Protein, g / l Hb, g / di IgM (Serum), mg / dl Globulin; gamma,. % Albumin, g / l H et, % CK-MB, U / l TSH (Hormones), plU / mi CK, U / l VCM. fi Neutrophils, % free T3 (Hormones), pg / mi AST (GOT), U / l TCMH. pg Lymphocytes, % free T4 (Hormones), ng / dl AL.AT (GPT), U / l CCMH, g / di Monocytes, % Erythrocyte Sedimentation Rate 1st hour, mm Gamma-GT, U / l RDW. % Eosinophils, % .

[0196] The conventional clinical data (ii) as well as the data (i) formed by the concentrations of the most differentiating mineral elements, are chosen by the recursive RFE selection algorithm, associated with the machine learning model used: Conditional Random Forest or Gradient Boosting Machine for each of the categories / targets chosen, the Conditional Random Forest or Gradient Boosting Machine model predicts the membership of 252 of the 655 Sx subjects with a specificity of more than 95%, as follows: * 93 to target C2 subjects category Sm>l(di) suffering from neurodegenerative diseases, * 29 to target C3 subjects category Sm>l(d2) suffering from inflammatory diseases of the CNS, * 40 target C2 subjects category Sm>l(di) subcategories: Sm>l(di i, Alzheimer's dementia; Sm>l(di.2) non-Alzheimer's dementia, * 31 to target C2 subjects category Sm^l^ subcategory Sm>l(di 4) other neurodegenerative diseases * and 59 to the target Cl subjects category Sm "°(d5) healthy, free from neurological diseases: premium controls.

[0197] Example 3; _ Prognosis / Therapeutic monitoring of a subject S 21 to be tested through comparative statistical analysis (1 protocol -S9.1- for implementing the method according to the invention:) and using an automatic learning and AI prediction model (2 protocol -S9.2- for implementing the method according to the invention:) on the one hand, from historical learning data (il & (ii) originating in particular from plasma samples from a population of sick reference subjects Sm>l -, and, on the other hand, from data (il & (ii) originating in particular from plasma samples from this subject S~ to be tested

[0198] -S0- Implementation of a database

[0199] -S0.1- Conventional clinical data (ii)

[0200] A database is constructed from 379 plasma samples from 379 reference female subjects with ERB2-positive breast cancer treated with antibodies (Trastuzumab_Herceptin) and from the TROÏKA study (NCT03013504) presented in the article by X. Pivot et al. (JAMA Oncol. 2022 May 1;8(5):698-705. ) "Ejjicacy of HD201 vs Referont Trastuzumab in Patients WithERBB2-Positive Breast Cancer Treated in the Neoadjuvant Setting. A Multicenter Phase 3 Randomized Clinical Trial ”.

[0201] Each of the reference subjects Sm>l suffering from breast cancer (ERB2 positive) was assigned, through conventional clinical data concerning the subjects: . to a target including the Sm(dij) category corresponding to a specific Sm(d) diagnostic criterion (criterion 1): Performance status according to the ECOG (Eastern cooperative oncology group) ECOG grid; and to a Sm(di.i) ECOG 0 subcategory or a Sm(di.2) ECOG 1 subcategory. . to a target including the Sm(d2) category corresponding to a specific diagnostic criterion (criterion 2)5: Largest clinical diameter of the tumor; and to a Sm(d2.i) subcategory >30mm, or to a Sm(d2.2) subcategory between 10 and 30 mm or to a Sm(d2 3)Non-interpretable subcategory. . to a target including the category Sm(d3)corresponding to a specific diagnostic criterion (criterion 3): number of lymph nodes; and to a subcategory Sm(d3.i) >3, or to a subcategory Sm(d3.2) between 1 and 3, or to a subcategory Sm(d33)none, or to a subcategory Sm(d3.4)Not interpretable; . to a target including the Sm(pi) category corresponding to a specific prognostic criterion (criterion 1): Complete local pathological response (tpCR) and to a subcategory S” p U) Yes p=0, or to a subcategory Sm( p i.2) No p=l, or to a subcategory S” pL3 j Not interpretable; . to a target including the Sm(p2) category corresponding to a specific prognostic criterion (criterion 2): Complete pathological response of the breast at the time of surgery (bpCR) and to a subcategory Sm(p2.i) Yes p=0, or to a subcategory Sm(P2.2) No p=l, or to a subcategory Sm(p2.3) Not interpretable; . to a target including the category Sm(p3)corresponding to a specific prognostic criterion (criterion 3): Resumption during the study or similar (Presence of an event) and to a subcategory Sm(p3.i) Yes p=0, or to a subcategory Sm(p3.2) No p=l; . to a target including the category Sm(ti)Corresponding to a specific therapeutic monitoring, given by the measured residual trastuzumab concentration (CtrOUgh), and to a subcategory Sm(ti d} Ctrough on cycle 5, or to a subcategory Sm(t. i 2) G^gh on cycle 8, or to a subcategory Sm(ti 3) Ctrough at the end of treatment. These targets are taken into account in the decision-making process involving statistical representations or artificial intelligence, specific to the method according to the invention.

[0202] For the sick subjects Sm>l( d>p >t ylc reference of this example 3, the targets or categories and the sub-targets or sub-categories are given in table 4 below.

[0203] [Tables4] Total of reference diseased subjects analyzed N* 379 Main category Subcategory Number (dOnyiçg.of performance of (ECOG - Eastern Cooperative Oncology Group) (dl.l)ECOGO 279 (d1.2)ECOG 1 100 (d2) Larger tumor size at inclusion (d2.1) >30 mm 193 (d2.2) between 10 and 30 mm 163 (d2.3) Not interpretable 23 (d3) Number of lymph nodes at screening (d3.1) >3 31 (d3.2) between 1 and 3 207 (d3.3) none 108 (d3.4) Not interpretable 33 (p1) Complete local pathological response (M® (pi .1) p=0 Yes 165 (p1.2) p=1 No 203 (p1.3) No interpretable 11 (p2) Pathological Complete Breast Response at the time of surgery (teSB) (p2.1) p=0 Yes 188 (p2.2) p-1 No 180 (p2.3) Not interpretable 11 (p3) Disease recurrence on study or similar (Presence of an event) within 24 months (p.3.1) p=0 Yes 81 (p3.2) p-1 No 318 (tl)Patient with measured residual trastuzumab concentration (t1 .1 ) Measured Cycle 5 371 (4 BLQ) (tl .2) Measured Cycle 8 366(1 BLQ) (tl.3^3n of treatment 344 (14 BLQ). BLQ: Below the limit of quantification.

[0204] -S0 ,2 - Implementation of a database: training data (i) constituted by the Mineral Profile (MP)

[0205] The database is completed for the N = 379 subjects Sm>l( d,p >t} sick Sm>l(di i ;dL2; dl.3 ; d2.1 ;d2.2;d2.3 ;d3.1 ;d3.2 ;d3.3 ;d3.4 ;pl.l ; pl.2;pl.3 p2.1 ; p2.2 ;p2.3 p3.1 ; p3.2;tl.l ;tl.2;tl.3 i de reference of this example 3, by the PM mineral profile of concentrations of Nem = 34 mineral elements, established for the corresponding plasma samples.

[0206] This mineral profile on the plasma samples was carried out as described in Example 1 for the CSF samples.

[0207] [Fig.8] shows the distribution of concentrations [pg / L] of 34 mineral elements in all BDR plasma samples.

[0208] -SL

[0209] A sample (stored at -18°C -S3-) from a subject Sx 112-003-001 whose therapeutic response is still undetermined is used.

[0210] -S2-

[0211] The following Dsx data are assigned: identification number ID 112-003-001 and conventional clinical data (ii): weight and age of Sx corresponding to this sample.

[0212] -S4- The PM of this sample is analyzed as described above for the N= 379 subjects Sm>l(d,p,t)) sick Sm>l(dl.l ;dl.2; dl.3 ;d2.1 ;d2.2;d2.3 ;d3.1 ;d3.2;d3.3;d3.4;pl.l ; pl.2 ;pl.3 p2.1 ; p2.2 ;p2.3 p3.1 ; P3.2 ; ti.1 ;ti.2 ;ti.3 from this example 3.

[0213] The detailed results are presented in Table 5 below, specifying the concentrations of minerals measured in ppb, in this sample corresponding to the subject Sx to be tested as well.

[0214] [Table 5] Concentrations in ppb of different minerals / Weight / Age Sx Weight (kg) Age (years) Na Mg Al PS Cl 2057233.3 40797.3 197.4 677143.3 302191.7 2718744 K Ca Ti Cr Fe Cu 1317188.3 32371.8 13.6 11.2 160229 4 775.5 Zn Ga Ge As Se Br 5977.7 0.9 6.4 2.1 24.4 3082.7 112- 003-001 70 60 151 Rb 1496.4 Sr 20 Zr 0.8 Cd 0.5 Sn 24.7 'I 6.7 Cs Ba La Ce Eu Gd 2.1 15.1 0.1 0.2 0.1 0.2 Tb W Hg Tl Pb U 0.1 2.6 1.3 0 15.9 0

[0215] -S8- Define at least two targets _ (each comprising at least one category and possibly to at least one subdivision of at least one category X to make at least one prediction of membership of the subject Sx to be tested to at least one target

[0216] In this example 3, we define 3 categories f Sm>l(pl); Sm>l(p2); Sm>l(p3) 3 subcategories of the category Sm>l(pij :Sm>l(Pi.i);Sm>l(Pi.2); 3 subcategories of the category Sm>l(p2):Sm>l(p2i);Sm>l(p22); 2 subcategories of the category Sm>l(p3): Sm>l(p3.i); Sm>l(p3.2}. These 3 categories and 6 subcategories are intended to form targets for which the process makes it possible to predict whether Sx belongs to one or more of them.

[0217] The membership of Sx to none, one or more targets is not predefined. Depending on the data and the categories / subcategories chosen, Sx can belong to 0, 1 or more targets. The physician can participate in defining the targets (categories / subcategories) of interest to target on the subject Sx.

[0218] In this example, the membership of Sx to none, to one or more targets, is predicted in 3 cases.

[0219] 1st scenario: Complete local pathological response (tpCR) Target Cl*: Category Sm>l(pi, subcategory Sm>l(pi i,p=0 Yes Target C2*: Category Sm>l(pi) subcategory Sm>l(pi 2)p=l No.

[0220] 2nd scenario: Complete pathological response of the breast at the time of surgery (bpCR) Target C3*: Category Sm>l(p2} subcategory Sm>l(p2.i)P=0 Yes Target C4*: Category Sm>l(p2) subcategory Sm>l(p22)p=l No.

[0221] 3rd scenario: Recurrence of the disease during the study or similar (Presence of an event) within 24 months Target C5*: Category Sm>l(p3} subcategory Sm>l(p3.i)P=0 Yes Target C6*: Category Sm>l(p3) subcategory Sm>l(p32)p=l No.

[0222] 2 — protocol S9.2- of _ step - S9 - of implementation of several models machine learning and AI prediction.

[0223] -S9.2.1- Collection of training data values: *Data (i) concentrations of mineral elements: see [Fig.8] & table 5. *Conventional clinical data (ii): age; weight: table 5.

[0224] -S9.2.2- Normalization of the values ​​of the learning data using StandardScaler

[0225] -S9.2.3- Implementation of 4 machine learning and prediction models by AI.

[0226] These 4 models are random forest, logistic regression, XGBoost and decision tree, configured with their default parameters.

[0227] -S9.2.4- Data separation: Division of the entire data set into training (80%) and testing (20%) parts.

[0228] -S9.2.5- Learning and evaluation loop:

[0229] Application of 5-fold cross-validation with StratifiedKFold for model training and evaluation, calculating for each iteration the false positive rates (FPR) and true positive rates (TPR).

[0230] -S9.2.6- Plotting the ROC curve:

[0231] Plotting the ROC curve for each model using the averages of the TPRs and FPRs, and calculating the average of the AUC over the iterations of the cross-validation.

[0232] The attached [Fig.9] shows the 8 ROC curves (curves A to I) corresponding to the 8 machine learning and AI prediction models, for the recurrence of the disease during the study or similar (Presence of an event) within 24 months, Target C5*: Category Sm>l(p3, subcategory Sm>l(p3 i)P=0 Yes Target C6*: Category Sm>l(p3) subcategory Sm>l(p32)p=l No.

[0233] The AUCs for the 8 models of [Fig.9] (curves A to I) are given in this figure.

[0234] In the legend of [Fig.9], n corresponds to the number of training data categories used for each model.

[0235] By studying the PM mineral profile of the plasmas of the N* = 379 subjects Sm>l( d>p >t,) patients Sm>l(dl.l ;dl.2 ; dl.3 ;d2.1 ;d2.2 ;d2.3 ;d3.1 ;d3.2 ;d3.3 ;d3.4 ;pl.l ; pl.2 ; p2.1 ; p2.2 ; p3.1 ; p3.2 ; tl.l ;tl.2 ;tl.3 ) of reference of this example 3, and with the help of these 4 models of automatic learning and prediction by AI, we can determine the belonging of the subject Sx to be tested, with a certain reliability, to the target C5* to the target C6*.

[0236] -S9.2.7- Machine learning and AI prediction model selection based on the best AUC score measuring the ability of each machine learning and AI prediction model to differentiate categories / targets. The Gradient Boosting Machine AI model is selected for the C5* / C6* targets ([Fig.9]), as it is the one that achieves the best AUC score.

[0237] -S9.2.8- Selection of training data:

[0238] The data to be selected are the categories / subcategories C5*: Category Sm>l(p3) subcategory Sm>l(p3.i)P=0 Yes & C6*: Category Sm>l(p3) subcategory Sm>l(p3 2)P=l No of the reference Sm>l sick subjects, the concentrations of the different minerals [data (i)] and the conventional clinical data (ii) associated with these reference subjects which are the most differentiating / relevant for the performance of the IA logistic regression model selected in -S9.2.7-, using the RFE estimator associated with the logistic regression model.

[0239] The RFE algorithm, associated with the logistic regression model, is used for the selection of differentiating data (i) & (ii). These data are: - for the 3rd scenario C5* / C6*: Recurrence of the disease during the study or similar (Presence of an event) within 24 months . Data (i) mineral elements: Na, K, Fe, Zn, Ge, As, Sn, Gd, Pb. . Conventional clinical data (ii): age in years; These data selected by the RFE algorithm are displayed and used for training the chosen AI model.

[0240] The confusion matrix ([Fig. 10] attached), showing true positives, false positives, true negatives and false negatives, is generated from the test data set (representing approximately 20% of the entire data set).

[0241] The sensitivity and specificity of the model are calculated. The hyperparameters are then adjusted in order to optimize the performance of the model.

[0242] The confusion matrix ([Fig. 10] attached) for this example 3 is carried out on 20% of the 379 subjects belonging to at least one of the two targets C5* / C6*, i.e. 76 subjects. The average sensitivity obtained is 0.45. The average specificity obtained is 0.93. In addition, the average AUC score obtained is 0.69. From the confusion matrix, we see that the chosen logistic regression AI model has predictive capabilities to know the therapeutic response of patients to the treatment. Thus, with an average AUC of 69% and a detection sensitivity of 93%, this model makes it possible to identify more than 90% of the Sx subjects to be tested, whose prognosis is a recurrence of their disease during the 24 months with the treatment carried out.

[0243] -S9.2.9- Prediction of membership of the subject Sx _ to be tested whose sample has the identification number 12-003-001 using the model selected in -S9.2.7-

[0244] The chosen logistic regression prediction AI model is implemented on the selected data of the subject Sx to be tested and the selected data of the targets C5* / C6* of the reference subjects.

[0245] The prediction is displayed, with the sensitivity, specificity and AUC score of the chosen logistic regression model, for the analyzed data.

[0246] The prediction thus made for sample 112-003-001 on the recurrence of the disease, using the most efficient AI model found previously (here logistic regression "Logistic Regression") is as follows: no recurrence of the disease during the study or similar (Presence of an event) within 24 months. Depending on the data selected for sample 112-003-001, the model, trained on the corresponding database, gives a probability score. If the score indicating the probability of the model is higher than the threshold of the model's prediction, defined according to the need, in terms of specificity and sensitivity, within the limit of the model, then the model predicts that the subject will belong to the selected category. This threshold can be chosen to increase the sensitivity or specificity of the model according to the needs. In this example, we choose a threshold that will allow us to obtain the best prediction score (compromise between sensitivity and specificity). In this case, it can be stated that sample 112-003-001 will not have a recurrence of the disease with a reliability of at least 72%. Additionally, the sensitivity, specificity, selected features, AUC score, and the total number of patients used to make this prediction and the number of targets were displayed. Model Probability: 0.18; Prediction Threshold: 0.20; Model Sensitivity: 0.72, Model Specificity: 0.57. Selected Features: Na, Mg, S, Cl, Ga, Ge, Zr, Cs, Ba, La. Average AUC Score: 0.65. Total N: 378. Target N: 61.

[0247] Example 4: Prognosis / Therapeutic monitoring of a subject S 21 to be tested through comparative statistical analysis (1 g protocol S9.1 for implementing the method according to the invention :) and using an automatic learning and AI prediction model (2 ~ protocol S9.2 for implementing the method according to the invention), on the one hand, from historical learning data (i) & (ii) originating in particular from urine samples from a population of sick reference subjects Sm>l, and, on the other hand, from data (i) & (ii) originating in particular from urine samples from this subject S 21 to be tested

[0248] -S0- Implementation of a database

[0249] -S0.1- Conventional clinical data (ii)

[0250] A database is constructed from 197 urine samples from 197 reference patients Sm>l, collected after cardiac surgery. This is the TRANSNEPHRON study (NCT 02763410) (Impact of the Composition of Packed Red Blood Cell Supernatant on Renal Dysfunction and Posttransfusion Immunomodulation -TRANSNEPHRON-). The investigators selected cardiac surgery patients for the transfusion frequency and significant post-surgical renal morbidity.

[0251] Each of the sick reference subjects Sm>l was assigned, through conventional clinical data concerning the subjects: . to a target including the category Sm(pi) corresponding to a specific prognostic criterion (criterion 1): Renal failure during the stay and to a subcategory Sm(pi i) Yes p=0, or to a subcategory Sm(pi.2) No p=l; . to a target including the category Sm(p2) corresponding to a specific prognostic criterion (criterion 2): Respiratory failure during the stay (lowest P / F in the 5 days <200) and to a subcategory Sm(p2.0 Yes p=0, or to a subcategory Sm(p2 2) No p=1, or to a subcategory Sm(p2.3) Not interpretable. . to a target including the category corresponding to a specific prognostic criterion Sm(p3) criterion 3: Infection during resuscitation and to a subcategory Sm(p3.i) Yes p=0, or to a subcategory Sm(p3.2) No p=l, or to a subcategory Sm(p3.3) Not interpretable; These categories are "targets" or "targets", within the framework of the decision-making process involving statistical representations or artificial intelligence, specific to the method according to the invention for diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, in particular plasma.

[0252] For the sick subjects Sm>l( d>p >t ,dc reference of this example, the targets or categories and the sub-targets or sub-categories are given in table 6 below.

[0253] [Tableauxô] Total patients analyzed 197 Main category Subcategory Name Specific prognostic criterion criterion 1: Renal failure during the stay YES p=0 60 ^1.2? No p-1 137 Specific prognostic criterion criterion 2: Respiratory failure during the stay (P / F lowest in the 5 days <200) ^,2.-5.5 YES p=û 68 Yes p=0 122 *^>2 3? Not interpretable 6 Specific prognostic criterion S 6 criterion 3: Infection in intensive care Yes p=Q 18 No p=1 175 Not interpretable 4 Specific prognostic criterion criterion 4: At least (one of the categories Yes pQ ^>45$ No p=1 Death during the study 8 P / F is an index of severity of hypoxia in cases of ventilation / perfusion disturbances at the pulmonary level resulting in a shunt effect. The normal value is 95mmHg / 21% = 452. A value below 300 indicates severe impairment of gas exchange within the lungs.

[0254] -S0.2- Implementation of a database: training data (i) constituted by the Mineral profile (PM)

[0255] The database is completed for the N = 197 subjects Sm>l( d>p >t,) sick Sm>l(pi.i ;pi.2; P2.1 ;P2.2;p2.3 ; P3.i ;P3.2; P3.3 ,p4.i ; p4.2) of reference of this example 4, by the mineral profile PM of concentrations of Nem = 37 mineral elements, established for the corresponding urine samples.

[0256] This mineral profile on the urine samples was carried out as described in Example 1, for the CSF samples.

[0257] [Fig. 11] shows the distribution of concentrations [pg / L] of 37 mineral elements in all urine samples in the database.

[0258] -SC

[0259] A sample (stored at -18°C -S3-) from a subject Sx S2023_CRB002 whose prognosis is still undetermined is used.

[0260] -S2-

[0261] The subject Sx to be tested corresponding to this sample is assigned the following data Dsx: the identification number S2023_CRB002 and the conventional clinical data (ii): weight (kg);age (years), smoking (yes= l / no=0), diabetes (yes= l / no=0), hemoglobin (in g / dL, preoperatively on the assessment before surgery), platelets (in number of platelets / mm3 preoperatively on the assessment before surgery), lymphocytes (in number of lymphocytes / mm3 preoperatively on the assessment before surgery), preoperative creatinine (in micromol / L preoperatively on the assessment before surgery), extracorporeal circulation duration (in minutes: duration of circulatory assistance during the block), clamp duration (in minutes: duration of interruption of aortic flow in minutes), dobutamine maximum dosage (dose in microgram / kg / min in continuous infusion), high lactate (in mmol / L hypoxia marker), low hemoglobin (same in g / dL: lowest value postoperatively), neutrophil count,. ;

[0262] -S4- The PM of this sample is analyzed as described above for the N = 197 subjects Sm>l(d>p>t)) sick Sm>l(Pn;pi.2;Pi.3, P2.i;P2.2;P2.3; P3.i;P3.2; P3.3;P4.i;P4.2) of this example 4.

[0263] The detailed results are presented in Table 7 below, specifying the measured mineral concentrations in ppb, in this sample corresponding to the subject Sx to be tested as well as the conventional clinical data (ii) for Sx to be tested whose identification number is S2023_CRB002.

[0264] [Table 7] Concentration in ppb of different minerals Sample Name Lt S Na Mg Ai Urine2G23_CRBœ2 14.676 172.192 3261246 90612.8 20.084 Mn Fe Co Ni Cu Zn 0 6 763 0168 0.7S7 5.103 79.652 Cr Cd Sn 1 Cs Ba 1.426 ©<246 3.038 8866.807 1.688 5.626 WEIGHT AGE SMOKING DIABETES HEMOGLOBAL PLATELETS .83 74 0 0 14.5 151000 Sample Name Si PS Cl Urine2023_CRB002 6696.932 343379.1 261257.6 7814680 Mh OS As Se Br 0 0.215 9.032 o 5422.522. Cr Gd W Ti Pb 1.426 0 0.225 0.037 0.878 LYMPHOCYTE WEIGHT CREAT PRE OP CEC DURATION CLAMPING DURATION 63 1476 128.2 76 Cl Name of feçhaïUilîon K sa Ti Udne2G23_CRB0D2 1029616 39320.61 2.7GS Mb Rb Sr Zr ü 600.541 88.032 0.259 Cr UV SEX 1.426 0.07 0.178 WEIGHT DOBUTAM1NEPOSOMAX • PO HIGH LACTATE LOW HEMOGLOBIN 63 •J" 1.5 9.6

[0265] -S8- Define at least two targets to make at least one prediction of membership of the subject Sx to be tested to at least one target (each target comprising at least one category and possibly at least one subdivision of at least one category)

[0266] In this example 4, we define 4 categories Sm>l(Pi);Sm>l(p2);Sm>l(p3);Sm>l(p4). 3 subcategories of the category Sm>l(pi): Sm>l(Pi.i); Sm>l(Pi.2); Sm>l(pi.3); 3 subcategories of the Sm>l(p2) category: Sm>l(p2i); Sm>l(p22); Sm>l(p23); 2 subcategories of the category Sm>l(p3):Sm>l(p3.i);Sm>l(p3.2); 2 subcategories of the category Sm>l(p4):Sm> 1(^.1);Sm> 1(^.

[0267] These 3 categories and these 10 sub-categories make it possible to form targets with respect to which the method makes it possible to predict whether Sx belongs to one or more of them.

[0268] The membership of Sx to none, one or more targets is not predefined. Depending on the data and the categories / subcategories chosen, Sx can belong to 0, 1 or more targets. The physician can participate in defining the targets (categories / subcategories) of interest to target on the subject Sx.

[0269] In this example 4, the membership of Sx to none, one or more categories and subcategories, is predicted in 4 cases.

[0270] 1st scenario: occurrence of renal failure during the stay Target Cl**: Category Sm>l(pi, subcategory Sm>l(pi i)P=0 Yes Target C2**: Category Sm>l(pi) subcategory Sm>l(pi.2)p=l No.

[0271] 2nd scenario: occurrence of respiratory failure during the stay (lowest P / F in the 5 days <200) Target C3**: Category Sm>l(P2 ) subcategory Sm>l(p2 i)P=0 Yes Target C4**: Category Sm>l(p2) subcategory Sm>l(p2.2)p=l No.

[0272] 3rd scenario: occurrence of an infection during resuscitation Target C5**: Category Sm>l(p3, subcategory Sm>l(p3 i)P=0 Yes Target C6**: Category Sm>l(p3) subcategory Sm>l(p32)p=l No.

[0273] 4th scenario: occurrence of at least one of the three complications according to the first 3 scenarios, respectively renal, pulmonary or infection complication during the stay Target C7**: Category Sm>l(p 4} subcategory Sm>l(p 4 i)P=0 Yes Target C8**: Category Sm>l(p 4 , subcategory Sm>l(p 4 2)p=l No.

[0274] -S9.1- Statistical processing _ j.

[0275] -S9.1.1- Collection of training data used: *Data (i) concentrations of mineral elements: see [Fig. 1] 1. *Conventional clinical data (ii): weight (kg); age (years), smoking (yes= l / no=0), diabetes (yes= l / no=0), hemoglobin (in g / dL, preoperatively on the assessment before surgery), platelets (in number of platelets / mm3 preoperatively on the assessment before surgery), lymphocytes (in number of lymphocytes / mm3 preoperatively on the assessment before surgery), preoperative creatinine (in micromol / L preoperatively on the assessment before surgery), extracorporeal circulation duration (minutes: duration of circulatory assistance during the block), clamping duration (in minutes: duration of interruption of aortic flow in minutes), dobutamine maximum dosage (dose in microgram / kg / min in continuous infusion), high_lactate (in mmol / L hypoxia marker), low_hemoglobin (same in g / dL: lowest value postoperatively), neutrophil count.

[0276] -S9.1.2- Normalization of training data values ​​using StandardScaler

[0277] -S9.1.3- [Variant VI of the 1 a protocol] Selection of training data by using “Volcano Plot” curves:

[0278] For the selection of the most relevant, i.e. most differentiating, data (i) concentrations of mineral elements and data (ii) conventional clinical data, we use, in this example, the Volcano Plot selection tool for the N* = 379 subjects subjects Sm>l( d>p >t,) sick Sm> l.pl, :pl 3: pl î: p3, :p;;:p2; . p3.i ;p3.2; p4.i ; p4.2) of reference of this example 4.

[0279] The volcano plots are plotted for the 4 cases in the same way as in example 1.

[0280] Thus, figures 12, 13, 14 & 15 attached represent volcano plots which correspond to the 4 cases mentioned above.

[0281] We observe a good variability of the impact of mineral elements on the occurrence of a complication after the operation. We observe a different impact of conventional clinical parameters already known, such as weight, sex, duration of the operation, platelet count. But we also observe impacts of the concentrations of certain mineral elements, thus statistically the presence of cadmium will favor the occurrence of a renal complication, the presence of zirconium (Zr) or tungsten (W) seems to protect against pulmonary insufficiency, while the presence of iodine (I) and lead (Pb) is linked to the occurrence of more infections. Overall high concentrations of zirconium (Zr) and selenium (Se) are linked to the presence of fewer complications, while high concentrations of potassium (K) and lead (Pb) are linked to the occurrence of fewer complications.

[0282] -S9.1.4- 2D or 3D stochastic dimension reduction We implement the t-SNE (t-distributed stochastic neighbor embedding) algorithm for - reduce the dimensionality of the targeted data by compressing multidimensional data into two or three main components, - and facilitate their visualization in 2D or 3D depending on the selected data and improve the distinction between the targeted categories.

[0283] -S9.1.5-Visualization on graphs of targets (targeted data) (-(i)- concentrations of the different minerals and -(ii)- conventional clinical data) most differentiating / relevant for the performance of the process.

[0284] A KDE kernel density plot (Kernel Density Estimate) is produced to visualize the distribution of the 3rd case targets C5** / C6**, as defined previously. To do this, the components extracted following the application of the t-SNE algorithm are plotted respectively on the abscissa axes and the ordinate axis.

[0285] A scatter plot is also overlaid to show the individual points.

[0286] The attached [Fig. 16] illustrates, in the form of two-dimensional graphs, the distributions between the C5** / C6** targets defined above:

[0287] Regions where the estimated density is equal to or greater than 50% of the maximum value will be displayed, with different shades or colors representing different density ranges.

[0288] The statistical processing -S9.1- allows an optimal separation of the categories, for example by t-SNE. [Fig. 16] represents on 2-dimensional graphs, the targets C5** / C6**, in order to best separate in space, the points representing the different subjects Sm(p3.i infection during resuscitation: yes p=0 / target C5**; and Sm(p3.2) Infection during resuscitation: no p=1 / target C6**.

[0289] This visualization helps to distinguish C5** / C6** targets based on selected data, such as concentrations of certain mineral elements.

[0290] -S9.1.6-Positioning

[0291] The location of the Dsx data (i) formed by the ppb concentrations of the different minerals (Table 7 above) and the Dsx data (ii) including (Table 7 above) in particular the weight and age corresponding to the subject Sx to be tested [Urine2023_CRB002], are integrated into the t-SNE graph of [Fig. 16], with a distinctive red cross to facilitate its location.

[0292] This location strongly suggests that the PM mineral profile of the Urine2023_CRB002 sample from subject Sx to be tested corresponds to the typical characteristics of the prognostic criterion Sm(p3.i) specific criterion 3: Infection during resuscitation yes p=0 / target C5**, which could point towards this prognosis.

[0293] 2 — protocol 2 S9.2- of step -S9- of implementation of several models machine learning and AI prediction.

[0294] -S9.2.1- Collection of training data values: *Data (i) concentrations of mineral elements concentrations of mineral elements: see [Fig.l]l *Conventional clinical data (ii): see -S9.1.1- above.

[0295] -S9.2.2- Normalization of the values ​​of the training data using StandardScaler.

[0296] -S9.2.3- Implementation of 4 machine learning and prediction models by AI,

[0297] These 4 models are random forest, logistic regression, XGBoost and decision tree, configured with their default parameters.

[0298] -S9.2.4- Data separation:

[0299] Division of the entire data set into training (80%) and testing (20%) parts.

[0300] -S9.2.5- Learning and evaluation loop:

[0301] Application of 5-fold cross-validation with StratifiedKFold for model training and evaluation, calculating for each iteration the false positive rates (FPR) and true positive rates (TPR).

[0302] -S9.2.6- Plotting the ROC curve:

[0303] Plotting the ROC curve for each model using the averages of the TPRs and FPRs, and calculating the average of the AUC over the iterations of the cross-validation.

[0304] The attached [Fig.l]7 shows the 4 ROC curves corresponding to the 4 machine learning and AI prediction models, for the 4th scenario: occurrence (prognosis) of at least one of the three complications according to the first 3 scenarios, respectively renal, pulmonary or infection complications during the stay. Target C7**: Category Sm>l(p4 , subcategory Sm>l(p4 i)P=0 Yes Target C8**: Category Sm>l(p4) subcategory Sm>l(p42)p=l No.

[0305] In [Fig. 1]7, the AUC for the random forest, XGBoost logistic regression and decision tree models is 0.636, 0.754, 0.556 and 0.560 respectively. By studying the PM mineral profile of the plasmas of the N' = 182 subjects Sm>l( d>p >t,) sick Sm> 1( pL1. pL2;pi,3 ; P2.1 ; P2.2 ;P2.3 ; P3.i ;P3.2 ;P4.i ; p4.2) of reference of this example 4, and with the help of these 4 models of automatic learning and prediction by AI, we can determine the belonging of the subject Sx to be tested, with a certain reliability, to: Target C7**: Category Sm>l(p4) subcategory Sm>l(p4.i)P=0 Yes Target C8**: Category Sm>l(p4) subcategory Sm>l(p42)p=l No.

[0306] - S9.2.7- Selection of machine learning and AI prediction model based on the best AUC score measuring the ability of each machine learning and AI prediction model to differentiate categories / targets. The logistic regression AI model is selected for the C5* / C6* targets ([Fig. 17]), as it is the one that achieves the best AUC score.

[0307] -S9.2.8- Selection of training data: The data to be selected using the RFE estimator associated with the regression model are the categories / subcategories of the target C7**: Category Sm>l(p4) subcategory Sm>l(p4 i)P=0 Yes & C8**: Category Sm>l(p4) subcategory Sm>l(p42)p=l No, of the reference Sm>l sick subjects, the concentrations of the different minerals [data (i)] and the conventional clinical data (ii) associated with these reference subjects that are most differentiating / relevant for the performance of the machine learning and AI prediction model logistic regression selected in -S9.2.7-.

[0308] The RFE algorithm, associated with the logistic regression model, is used for the selection of differentiating data (i) & (ii). These data are: - for the 4th scenario: occurrence (prognosis) of at least one of the three complications according to the first 3 scenarios, respectively renal, pulmonary or infection complication during the stay. Target C7**: Category Sm>l(p4) subcategory Sm>l(p4i)P=0 Yes Target C8**: Category Sm>l(p4) subcategory Sm>l(p4.2)P=l No. * Data (i) mineral elements: Li, S, Cl, K, Cu, Se, Rb, Zr, Gd. * Conventional clinical data (ii): weight, smoking, platelets, neutrophil count, duration of extracorporeal circulation, maximum noradrenaline dosage.

[0309] These data selected by the RFE algorithm are displayed and used for training the chosen AI model.

[0310] The confusion matrix ([Fig.21] attached), showing the true positives, false positives, true negatives and false negatives, is generated from the test data set (representing approximately 20% of the entire data set).

[0311] The sensitivity and specificity of the model are calculated. The hyperparameters are then adjusted in order to optimize the performance of the model.

[0312] The confusion matrix ([Fig. 18] attached) for this example 4 was carried out on 25% of the 182 subjects belonging to at least one of the two targets C7** / C8**, i.e. 45 subjects. The average sensitivity obtained is 0.78. The average specificity obtained is 0.71. In addition, the average AUC score obtained is 0.78. From the confusion matrix, we see that the model has predictive capabilities to know the prognosis after the operation, with the occurrence or not of at least one complication such as renal failure, hepatic failure or infection. Thus, with an average AUC of 78%, we can make a good prognosis on nearly 4 out of 5 patients.

[0313] -S9.2.9- Prediction of membership of the subject Sx to be tested whose sample has the identification number S2023 CRB002 using the model selected in-S9.2.7-

[0314] The chosen logistic regression prediction model is implemented on the selected data of the subject Sx to be tested and the selected data of the target categories C7** / C8** of the reference subjects.

[0315] The prediction is displayed, with the sensitivity, specificity and AUC score of the chosen logistic regression model, for the analyzed data.

[0316] The prediction thus made for the sample S2023_CRB002 on the appearance of at least one complication of the type renal failure, hepatic failure or infection, using the most efficient AI model found previously (logistic regression "Logistic Regression")

[0317] Depending on the data selected for the sample S2023_CRB002, the model, trained on the corresponding database, gives a probability score. If the score indicating the probability of the model is greater than the prediction threshold of the model, defined according to the need in terms of specificity and sensitivity, within the limit of the model, then the model predicts that the subject will belong to the selected target. This threshold can be chosen to increase the sensitivity or specificity of the model as needed.

[0318] Here we opted for high sensitivity. In this case, we can say that the sample S2023_CRB002 will not have any of these 3 complications with a reliability of at least 88%. In addition, we have displayed the sensitivity, specificity, selected characteristics, AUC score as well as the total number of patients used to make this prediction and the number of targets. Urine Sample Prediction S2023_CRB002: PF ratio > 200, no renal failure or infection: - in Target C8**, Category Sm>l(p4) subcategory Sm>l(p42)P=l No; - outside Target C7**, Category Sm>l(p4) subcategory Sm>l(p4.i)P=0 Yes

[0319] Model probability: 0.26; prediction threshold: 0.36; Model sensitivity: 0.88, Model specificity: 0.42

[0320] Selected characteristics: ['S', 'Cl', 'K', 'Cu', 'Se', 'Rb', 'Zr', 'Gd', 'WEIGHT', 'PLATELETS', 'NEUTROPHIL COUNT', 'NORADRENALINE_POSO_MAX -PO]

[0321] Average AUC Score: 0.77

[0322] N Total: 181

[0323] N Target: 95

[0324] Example 5: Prognosis / Therapeutic monitoring of a subject Sx to be tested through comparative statistical analysis (1st protocol -S9.1- for implementing the method according to the invention :) and using a machine learning and AI prediction model (2nd protocol -S9.2- for implementing the method according to the invention), on the one hand, historical learning data derived in particular from samples of follicular fluid from a population of sick reference subjects Sm>l, and, on the other hand, data derived in particular from samples of follicular fluid from this subject Sx to be tested

[0325] -S0- Implementation of a database

[0326] -S0.1- Conventional clinical data (ii)

[0327] A database is constructed from 64 (individually collected at the time of puncture at the level of a follicle) + 22 (pooled at the level of all the samples taken from the same sick reference subject Sm>l) samples of follicular fluid from 64 + 22 female reference subjects, undergoing oocyte puncture for in vitro fertilization (IVF).

[0328] Each of the sick reference subjects Sm>l was assigned, through conventional clinical data concerning the subjects: . to a target including the category corresponding to a specific diagnostic criterion Sm(d) criterion 1 Sm(di): pathologies; to a subcategory Sm(di.i) endometriosis or to a subcategory Sm(di.2) tubal; to a subcategory Sm(di.3) DOR or to a subcategory Sm(di.4) reference; or to a subcategory Sm(di.5) miscellaneous or unknown; . to a target or category corresponding to a specific diagnostic criterion Sm(d) criterion 2 Sm(d2): Oocyte status (individual fluid); and to a subcategory Sm(d2.i) M2, or to a subcategory Sm(d2.2) empty; . to a target including the category corresponding to a specific diagnostic criterion Sm(d) criterion 3 Sm(d3): fertilization; and to a subcategory Sm(d3.i) yes, or to a subcategory Sm(d3.2) no, or to a subcategory Sm(d3.3)lysed, or to a subcategory Sm(d3.4)3PN; . to a target including the category corresponding to a specific diagnostic criterion Sm(d) criterion 4 Sm(d4): transfer; and to a subcategory Sm(d4.i) yes, or to a subcategory Sm(d4 2) no, . to a target including the category corresponding to a specific prognostic criterion Sm(pi) criterion 1: outcome of IVF and to a subcategory Sm(pi.0 pregnancy yes p=0, or to a subcategory Sm(pi.2) pregnancy no p=1, or to a subcategory Sm(pi.3) white puncture; or to a subcategory Sm(pi.3) IVF failure. These targets are taken into account in the decision-making process involving statistical representations or artificial intelligence, specific to the method according to the invention, in this example 5.

[0329] For the reference sick subjects Sm>l( d>p >t , of this example, the categories and subcategories are given in table 8 below.

[0330] [Tables8] Total N* of patients analyzed Individual fluid 64 Fluid 22 Main category Subcategory number (dl [Pathologies (d 1.1 ) Endometriosis 13+4 (d1.2) Tubal 12 (d1.3) DOR 8+4 (d1.4) Reference 24+9 (d.1.5) Various or unknown 12 (d.2) Oocyte status (individual fluid) (d2.1)M2 39 (d2.2) Empty 23 (d3) Fertilization (d3.1 ) yes 27 (d3.2) no 5 (d3.3) Lysed 5 (d3.4)3PN 2 (d4) T ransfer (d4.1 ) Yes 5 (d4.2) No 21 (ç1 ) from IVF (p.1 ) Pregnancy* 18 (p.2) No pregnancy 40 (p.3) White puncture 2 (p.4) Fertilization failure 4

[0331] - S0.2- Implementation of a database: training data (i) constituted by the Mineral profile (PM)

[0332] The database is completed for the Nt= 64+22 subjects Sm>l( d>p >t jnaladcs of reference Sm>l( d>p ) (dL1 ; dL2 ; d1.3 ; dl.4 ; dl.5 ; d2.1 ; d2.2 ; d3.1 ; d3.2 ; d3.3 ; d3.4 ; d 4.1 ; d4.2 ; pl.l ; pl.2 ; pl.3 p 1.4); of this example 5, by the mineral profile PM of concentrations of Nem = 31 mineral elements, established for the corresponding follicular fluid samples.

[0333] This mineral profile on the follicular fluid samples was performed as described in Example 1 for the CSF samples.

[0334] [Fig. 19] shows the distribution of concentrations [pg / L] of 31 mineral elements in all follicular fluid samples in the database.

[0335] -Sl-

[0336] A sample (stored at -18°C -S3-) from a Sx CHUL4 subject whose diagnosis and IVF outcome prognosis are still undetermined is used.

[0337] -S2-

[0338] The following Ds x data are assigned: the CHUL4 identification number and the conventional clinical data (ii): weight and age of Sx corresponding to this sample.

[0339] -S4- The PM of this sample is analyzed as described above for the Nt= 64+22 subjects Sm>l(d>p,t,reference patients Sm>l(d>p, (di.i;di.2; di.3; di.4; di.5; d2.i;d2.2; d3.1;d3.2;d3.3;d3.4; d4.1; d4.2;pl.l; pl.2;pl.3 pl.4) from This example 5.

[0340] The detailed results are presented in Table 9 below, specifying the concentrations of minerals measured in ppb, in this sample corresponding to the subject Sx to be tested as well as the conventional clinical data (ii): age of the subject Sx to be tested CHUL4.

[0341] [Table 9]: Concentration in ppb of the different minerals Identification ILM 11 B (ppb) 23 Na (ppb) 24 Mg (ppb) 27 Ai (ppb) 23 Si (ppb) 31 P (ppb) CHUM < LOD 3843446.57 17632.98 <LOD 3826,52 122788,81 Age 137 Ba(ppb) 139 La (ppb) 140 Ce (ppb) 141 Pr(ppb) 146 Nd (ppb) 153 Eu (ppb) 38 0 < LOD < LOD < LOD < LOD < LOD 51 V(ppb) 52 Cr(ppb) 55 Mn(ppb) 56 Fe (ppb) 59 Co (ppb) 60 Ni (ppb) 63 Cu (ppb) <LOD 0,35 1,68 231,05 0,46 0.15 1226,4 identification fLM 34 S (ppb) 35 Ci (ppb) 39 K (ppb) 43 Ca (ppb) 45 Sc(ppb) 47 Ti(ppb) CHUL4 737298.11 4553966.26 138568.49 62735.43 < LOD 2.25 Age 157 Gd(ppb) 159 Tb (ppb) 182 W (ppb) 205 Tl (ppb) 208 Pb (ppb) 238 U (ppb) 38 0.22 < LOD < LOD < LOD 0.08 0.01 51 V (ppb) 66 Zn (ppb) 75 As (ppb) 78 Se (ppb) 79 8r(ppb) 85 Rb (ppb) 88 Sr(ppb) < LOD 531.95 0.11 42.03 5512.08 160.59 39.35 LOD: Limit Of Detection

[0342] -S8- Define at least two targets to make at least one membership prediction from the subject Sx to be tested to at least one target.

[0343] In this example 5, we define 5 categories Sm>l(di)Sm>l(d2)Sm>l(d3)Sm>l(d4)Sm>l (pD 5 subcategories of the Sm>l(di) category: Sm>l(di.i); Sm>l(di.2); Sm>l(di.3); Sm>l(di.4); Sm>l(di 5). 2 subcategories of the category Sm>l(d2):Sm>l(d2.i);Sm>l(d2.2)); 4 subcategories of the category Sm>l(d3):Sm>l(d3 i);Sm>l(d3 2); Sm>l(d3 3) Sm>l(d3 4) 2 subcategories of the category Sm>l( d 4):Sm>l( d 4;Sm>l( d 4 2); 4 subcategories of the Sm>l(pi) category: Sm>l(pi.i); Sm>l(pi.2); Sm>l(pi.3); Sm>l(pi.4).

[0344] These 5 categories and these 17 sub-categories make it possible to form targets with respect to which the method makes it possible to predict whether Sx belongs to one or more of them.

[0345] The membership of Sx to none, one or more targets is not predefined. Depending on the data and the categories / subcategories chosen, Sx can belong to 0, 1 or more targets. The physician can participate in defining the targets (categories / subcategories) of interest to aim at on the subject Sx.

[0346] In this example, the membership of Sx in none, one or more categories and subcategories is predicted in 4 cases.

[0347] 1st scenario: Target Cl***: Category Sm>l( di, subcategory Sm>l( d U)endometriosis Target C2***: Category Sm>l( d 0 subcategory Sm>l( d L 4} reference: control (linked to a male infertility problem), the woman's follicular fluid has no known problem.

[0348] 2nd scenario: Target C3***: Category Sm>l(pi) resulting from IVF subcategory Sm^l^u, pregnancy yes p=0 Target C4***: Category Sm>l(pi) subcategory Sm>l(pi2)reference: pregnancy not p=l.

[0349] -S9.1- Statistical processing _

[0350] -S9.1.1- Collection of training data used: . Data (i) concentrations of mineral elements: see [Fig.2]l. . Conventional clinical data (ii): age.

[0351] -S9.1.2- Normalization of the values ​​of the training data using StandardScaler

[0352] -S9.1.3- [Variant VI of the 1st protocol] Selection of training data by using “Volcano Plot” curves:

[0353] For the selection of the most relevant, i.e. most differentiating, (i) mineral element concentration data and (ii) conventional clinical data, we use, in this example, the volcano Plot selection tool for the Nt= 64+22 subjects Sm> 1( d,p ,t jnaladcs of reference Sm> 1( d,p ,(di,i ;di,2 ; di3 ; di.4 ; di.5 ; d2.i ;d2.2 ; d3.1 ;d3.2 ;d3.3 ;d3.4 ; d4.1; d4.2 ;pl.l ; pl.2 ;pl.3 pl.4) of This example 5.

[0354] The volcano plot is plotted for this example 5, in the same way as in the example 1.

[0355] Thus the attached [Fig.20] represents a volcano plot which corresponds to the 1st case of this example 5: Target Cl***: Category Sm>l(di) subcategory Sm>l(du) endometriosis versus Target C2***: Category Sm>l(dij subcategory Sm>l(di.4) reference: control (linked to a male infertility problem), the woman's follicular fluid has no known problem.

[0356] The more cobalt there is in the follicular fluid, the more likely the fluid is to come from an Sx subject of the Sm>l(dj type, subcategory Sm>l(di i)endometriosis: Target Cl***.

[0357] S9.1.3- [Variant V2 of the 1 H protocol]: we implement this variant such that described in example 1 above

[0358] -S9.1.3.1- Definition of models

[0359] -S9.1.3.2- Data Separation

[0360] -S9.1.3.3- Learning and evaluation loop

[0361] -S9.1.3.4- Plotting the ROC curve for each model and calculating the average of AUC (AUC score)

[0362] -S9.1.3.5- Choice of machine learning and AI prediction model based on the best AUC score.

[0363] - S9.1.4- 2D or 3D stochastic dimension reduction

[0364] -S9.1.5-Visualization on graphs of the data used (concentration of different minerals and conventional clinical data) most differentiating / relevant for the performance of the process.

[0365] A KDE kernel density plot, (Kernel Density Estimate) is produced to visualize the distribution of C3*** / C4*** targets as previously defined using the t-SNE algorithm

[0366] A scatter plot is also overlaid to show the individual points.

[0367] The attached [Fig.21] illustrates, in the form of two-dimensional graphs, the distributions between the categories C3*** / C4*** Target C3***: Category Sm>l(pi} resulting from IVF subcategory Sm>l(pi d)pregnancy yes p=0 Target C4***: Category Sm>l(pi) subcategory Sm>l(pi.2)reference: pregnancy not p=l.

[0368] Regions where the estimated density is equal to or greater than 50% of the maximum value will be displayed, with different shades or colors representing different density ranges.

[0369] The statistical processing -S9.1- allows an optimal separation of the categories, for example by t-SNE. [Fig.21] represents on 2-dimensional graphs, the 2 targets C3*** / C4***, in order to best separate the points representing the different target subjects C3***: Category Sm>l(pl) resulting from IVF subcategory Sm>l(pi 0 pregnancy yes p=0 and target C4***: Category Sm>l(pij subcategory Sm>l(pi.2) reference: pregnancy no p=l.

[0370] This visualization helps to distinguish between different targets based on selected parameters, such as concentrations of certain mineral elements. [0371 ] -S9.1.6-Positioning of the subject Sx CHUL4 to be tested

[0372] The location of the Dsx data (i) formed by the ppb concentrations of the different minerals (Table 9 above) and the Dsx data (ii) including (Table 9 above) in particular the weight and age corresponding to the Sx subject to be tested [CHUL4 follicular fluid], are integrated into the t-SNE graph of [Fig.21], with a distinctive red cross to facilitate its location. This location strongly suggests that the PM mineral profile of the test sample [CHUL4 follicular fluid] of the Sx subject to be tested, corresponds to the typical characteristics of the C3*** target: Category Sm>l(pi} from IVF subcategory Sm>l(pi i) pregnancy yes p=0, which could point towards this prognosis.

[0373] 2nd protocol - S9.2- of step -S9- of implementation of several machine learning and AI prediction models.

[0374] -S9.2.1- Collection of training data values: Data (i) concentrations of mineral elements. Table 9 above. (ii) Conventional clinical data: age. Table 9 above.

[0375] -S9.2.2- Normalization of training data values ​​using StandardScaler

[0376] -S9.2.3- Implementation of 4 machine learning and prediction models by AI,

[0377] These 4 models are random forest, logistic regression, XGBoost and decision tree, configured with their default parameters.

[0378] -S9.2.4- Data separation:

[0379] Division of the entire data set into training (80%) and testing (20%) parts.

[0380] -S9.2.5- Learning and evaluation loop:

[0381] Application of 5-fold cross-validation with StratifiedKFold for model training and evaluation, calculating for each iteration the false positive rates (FPR) and true positive rates (TPR).

[0382] -S9.2.6- Plotting the ROC curve:

[0383] Plotting the ROC curve for each model using the averages of the TPRs and FPRs, and calculating the average of the AUC over the iterations of the cross-validation.

[0384] The attached [Fig.22] shows the 8 ROC curves corresponding to the 8 machine learning and AI prediction models, for the outcome of IVF: pregnancy or non-pregnancy: Target C3***: Category Sm>l(pi, resulting from IVF subcategory Sm>l(pi pregnancy yes p=0 Target C4***: Category Sm>l(pi) subcategory Sm>l(pi.2)reference: pregnancy not p=l.

[0385] The AUCs for the 8 ROC curves of [Fig.22] (curves A to I) are given in this figure.

[0386] In the legend of [Fig.22], n corresponds to the number of training data categories used for each model.

[0387] By studying the PM mineral profile of the follicular fluids of the Nt = 64+22 Sm>l( d,p ,t jpatients of reference Sm> 1( d,p ) (dl.l ; dl.2 ; dl.3 ; dl.4 ; dl.5 ; d2.1 ; d2.2 ; d3.1 ; d3.2 ; d3.3 ; d3.4 ; d4.1 ; d4.2; Pi.i ; pi.2 ; pi.3 Pi.4) of this example 5, and with the help of these 4 machine learning and AI prediction models, we can determine the belonging of the subject Sx to be tested CHUL4, with certain reliability, to C3*** or C4***.

[0388] -S9.2.7- Machine learning and AI prediction model selection based on the best AUC score measuring the ability of each machine learning and AI prediction model to differentiate categories / targets. The logistic regression model is selected for targets C3*** / C4*** ([Fig.22]), as it is the one that achieves the best AUC score.

[0389] -S9.2.8- Selection of training data: The data to be selected are those of the categories / subcategories Category Sm>l(pi) from IVF subcategory Sm>l(Pi 0 pregnancy yes p=0 Category Sm>l(pi) subcategory Sm>l(pi.2)reference: pregnancy no p=l, sick subjects Sm>l or healthy subjects Sm "of reference, the concentrations of the different minerals [data (i)] and the conventional clinical data (ii) associated with these reference subjects the most differentiating / relevant for the performance of the machine learning and AI prediction model logistic regression selected in -S9.2.7-, thanks to the RFE estimator associated with the logistic regression model. The RFE algorithm, associated with the logistic regression model, is used for the selection of differentiating data (i) & (ii).

[0390] These data are: - for the 2nd scenario C3*** / C4***: pregnancy or not pregnancy. . Data (i) mineral elements: 1a test Example 5: Mg, P, Cr, Mn, Zn, Rb, Zr, Cd, I, Tb. 2nd test Example 5: Mg, P, Cr, Mn, Zn, Rb, Y, Zr, Cd, I, Ce, Tb. . Conventional clinical data (ii): age Table 9 above.

[0391] These data selected by the RFE algorithm are displayed and used for training the chosen AI model.

[0392] The confusion matrix ([Fig.23] attached) from IVF pregnancy vs. no pregnancy, showing true positives, false positives, true negatives and false negatives, is generated from the test data set (representing approximately 20% of the entire data set).

[0393] The sensitivity and specificity of the model are calculated. The hyperparameters are then adjusted in order to optimize the performance of the model.

[0394] The confusion matrix ([Fig.23] attached) for this example 5 was constructed on 25% of the 58 subjects belonging to at least one of the two targets C3*** / C4***, i.e. 14 subjects. The average sensitivity obtained is 0.84. The average specificity obtained is 0.80. In addition, the average AUC score obtained is 0.82.

[0395] From the confusion matrix, we see that the chosen logistic regression model has predictive capabilities to know at this stage, with 80% reliability, from the mineral profile, the occurrence of a pregnancy after in vitro fertilization.

[0396] -S9.2.9- Prediction of membership of the subject Sx to be tested whose sample has the CHUL4 identification number using the template selected in -S9.2.7-

[0397] The chosen logistic regression prediction model is implemented on the selected data of the subject Sx to be tested and the selected data of the target categories C3*** / C4*** of the reference subjects.

[0398] The prediction is displayed, with the sensitivity, specificity and AUC score of the chosen logistic regression model, for the analyzed data.

[0399] Depending on the data selected for the CHUL4 sample, the model, trained on the BDR, gives a probability score. If the score indicating the probability of the model is greater than the prediction threshold of the model, defined according to the need in terms of specificity and sensitivity, within the limit of the model, then the model predicts for the subject Sx to be tested whose sample has the identification number CHUL4, the occurrence of a pregnancy after In Vitro fertilization. This threshold can be chosen to increase the sensitivity or specificity of the model according to the needs.

[0400] Here we have opted for a threshold that will allow us to obtain the best prediction score (compromise between sensitivity and specificity). In this case, we can affirm that the subject corresponding to the CHUL4 sample will not have a pregnancy at the end of the IVF attempt, with a reliability of at least 78%.

[0401] XH test Example 5: Prediction sample CHUL4: no pregnancy

[0402] Selected data (i): Mg, P, Cr, Mn, Zn, Rb, Zr, Cd, I, Tb

[0403] Model Probability: 0.209; Prediction Threshold: 0.421; Model Sensitivity: 0.78; Model Specificity: 0.85; Mean AUC Score: 0.76; N* Total Subjects: 57;

[0404] 2nd test Example 5: Prediction sample CHUL4 high sensitivity: no pregnancy

[0405] Selected data (i): Mg, P, Cr, Mn, Zn, Rb, Y, Zr, Cd, I, Ce, Tb

[0406] Model Probability: 0.19; Prediction Threshold: 0.20; Model Sensitivity: 0.89; Model Specificity: 0.49; Average AUC Score: 0.80; N* Total Subjects: 57;

[0407] The threshold was reduced to increase the sensitivity of the model, we then still obtain a prognosis of pregnancy failure at the end of the IVF attempt for the subject Sx corresponding to the CHUL4 sample, we have for this second test a sensitivity of 0.89, that is to say a reliability close to 89%.

Claims

1. Claims Method for diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, characterized in that it essentially consists of: -S0- Implement at least one reference database (RDB) comprising data relating, on the one hand, to biological fluid samples, and, on the other hand, to the subjects from whom the biological fluid samples come; - each sample corresponding to a healthy reference subject Sm"° or to a sick reference subject Sm -1 treated or not, m being a positive natural integer; - each reference subject Sm(djPjt) corresponding to a diagnostic category Sm(d) and / or a prognosis category Sm(p) and / or a therapeutic response category Sm(t); each category may possibly be subdivided into one or more levels; - the database comprising data relating to a total number N* of reference subjects Sm"°(djPjt) and Sm>l(djPjt); with N* greater than or equal to - in increasing order of preference - 50, 100, 500, 1000; - these data comprising for each sample: (i) at least one PM mineral profile of mineral element concentrations in the biological fluid of reference subjects S"K,(dpl) and / or Sm>l(djPjt); preferably the concentrations of a number Nem mineral elements; * Nem being - in ascending order of preference - greater than or equal to 5, 10, 20, 25, 30; these mineral elements being chosen from the group comprising -advantageously consisting of: Li, B, Na, Mg, Al, Si, P, S, Cl, K, Ca, Sc, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, Ga, Ge, As, Se, Br, Rb, Sr, Y, Zr, Nb, Mo, Ru, Rh, Pd, Ag, Cd, In, Sn, Sb, I, Cs, Ba, La, Ce, Pr, Nd, Sm, Eu, Tb, Ta, W, Os, Pt, Au, Tl, Pb, Bi and U; and, preferably from the subgroup comprising -advantageously consisting of: Na, Mg, Al, Si, P, S, Cl, K, Ca, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Br, Rb, Sr, Zr, Sn, I, Cs, Ba, Pb, Bi and U; (ii) preferably, conventional clinical data relating to the reference subjects Sm"°(djPjt) and / or Sm>l(djPjt), these data (ii) being chosen from the group comprising -advantageously consisting of-: age, gender, height, weight, data on the type of disease considered provided that it is known, data on the medical history of the reference subjects Sm"0(djPjt) and / or Sm>l(djPjt), data on the therapeutic treatment(s) of the reference subjects S"K,(dpl) and / or Sm>l(djPjt), provided that they exist, lifestyle, concentrations of components of the biological fluid other than mineral elements, in particular concentrations of biomarkers, proteins, lipids, lipoproteins, red blood cells, white blood cells, platelets; -SI- Implement at least one Ex sample of biological fluid from at least one Sx subject to be tested; -S2- Optionally assign to each sample Ex of step -Sl-DSx data on the subject Sx to be tested, these DSx data meeting the same definition as the data (ii) referred to in SO; -S3- Possibly keep at least part of the Ex samples from step SI under specific conditions; -S4- Analyze each sample from step SI to determine at least one PM as defined in S0(i); -S5- Optionally, complete and / or update the data assigned in step S2, at least once, for all or part of the samples; -S6- Optionally complete and / or repeat the analyses carried out in step S4, at least once, on all or part of the samples from step S2 or S3; -S7- Enrich the database with the data produced in at least one of steps S4, S5, S6; -S8- Define at least two targets to make at least one prediction of membership of the subject Sx to be tested, to at least one target, each target comprising at least one category and / or at least one subdivision of at least one category; -S9- Implement -S9.1-: 1H protocol consisting of - Carry out statistical processing of the BDR data, preferably of a selection of these data;

2. *among which are the PM mineral profiles (ii) comprising at least X differentiating mineral elements, X corresponding to - in increasing order of preference - 3, 5, 10, 15, 15, 20, 25, 30; *and relating to Sm"°(djPjt) and / or subjects Sm(djPjt); to assign the subjects Sm(djPjt) to at least one target defined in S8; using a tool for reducing the dimensionality of the data in 2D or 3D and for graphical visualization of the targets; - and Position on the graphic visualization of the targets, the DSx data on the Sx subject(s) to be tested, to visualize the positioning of the Sx(s), in relation to the targets; to assist in the diagnosis, prognosis and / or therapeutic monitoring of diseases, in particular neurological diseases, for the Sx subject(s); and / or, -S9.2- Implement a 2 — protocol consisting of: - Implementation of at least one machine learning and AI prediction model; - Learning by this(these) model(s) in the BDR to choose at least one model and / or a selection of data, *among which are the mineral profiles PM (ii) comprising at least X differentiating mineral elements, X corresponding to - in increasing order of preference - 3, 5, 10, 15, 15, 20, 25, 30, these X differentiating mineral elements being included among the most effective for predicting the belonging of the subject Sx to be tested, to at least one target defined in S8; - Possibly optimization of the parameters of the chosen model(s); - Use of the chosen model(s) to predict the belonging of the subject Sx to at least one target defined in S8 and to help in the diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid, for the subject Sx to be tested. Method according to claim 1, characterized . in that the biological fluid is chosen from the group comprising - advantageously consisting of - cerebrospinal fluid, whole blood, blood plasma, urine, follicular fluid, bile, . and in that the diseases impacting the mineral profile of the biological fluid include: . in the case of cerebrospinal fluid, diseases affecting the central nervous system, in particular neurological diseases, such as psychiatric diseases, neurodegenerative diseases such as Parkinson's or Alzheimer's, . in the case of blood plasma, cancers, neurological diseases, diseases related to fertility, cardiac arrhythmias, . in the case of urine, diseases related to surgical complications following complex operations, in particular cardiovascular, . in the case of follicular fluids, diseases related to endometriosis or diseases causing fertility problems and requiring in particular procreation aids.

3. Method according to claim 1, characterized in that the BDR comprises data relating to reference healthy subjects Sm"°(djPjt) or reference sick subjects Sm>l(djPjt) who suffer from one or more diseases impacting at least two biological fluids chosen from the group comprising - advantageously constituted by - cerebrospinal fluid, whole blood, blood plasma, urine, follicular fluid, bile.

4. Method according to claim 1, characterized in that it comprises a step -SOa- of constructing the reference database consisting of collecting data, in particular data (i) and / or (ii) and / or DSx data and storing them.

5. Method according to at least one of the preceding claims, characterized in that the analysis in S4 is semi-quantitative, inductively coupled plasma mass spectrometry, or ICP-MS (Inductively Coupled Plasma Mass Spectrometry) being preferred.

6. Method according to at least one of the preceding claims, characterized in that the prediction of membership of the subject Sx to be tested to at least one target according to -S8-, is part of a diagnosis of diseases impacting the mineral composition of at least one biological fluid, for the subject Sx to be tested, and in that the BDR comprises data relating to a number Ns of healthy subjects S m"0(d,P,t) of reference and a number Nm of sick subjects Sm>l(djPjt) of reference; with [Ns / (Ns+ Nm)] *100 greater than or equal to -in an increasing order of preference- 20%, 30%, 40%.

7. Method according to at least one of the preceding claims, characterized in that the 1st protocol -S9.1- comprises the following steps: -S9.1.1- Collection of data in the BDR; -S9.1.2- Normalization of the data; -S9.1.3- Selection from the data (i), and possibly the data (ii), of differentiating data; -S9.1.4- Reduction of the dimensionality of the data in 2D or 3D, preferably by a PCA principal component analysis method and / or by a t-SNE (t-distributed Stochastic Neighbor Embedding) method; -S9.1.5- Visualization of the targets on graphs; -S9.1.6- Positioning on the graphic visualization of the targets, the DSx data on the subject(s) Sx to be tested, to visualize the positioning of the Sx(s), relative to the targets.

8. Method according to claim 7 characterized in that step -S9.1.3- consists of using: * [Variant VI of the 1st protocol] “volcano plot” curves and / or * [Variant V2 of the 1st protocol] a selection algorithm, preferably a recursive selection algorithm “recursive features elimination-RFE”, possibly associated with a machine learning and AI prediction model chosen from a group of machine learning and AI prediction models, by implementing the following sub-steps: -S9.1.3.1- Definition of the models; -S9.1.3.2- Separation of the data; -S9.1.3.3- Learning and evaluation loop; -S9.1.3.4- Plotting the ROC curve for each model and calculating the average of the AUC (AUC score); -S9.1.3.5- choice of machine learning and AI prediction model based on the best AUC score.

9. Method according to at least one of claims 1 to 5 characterized in that the 2 — AI protocol in S9.2 comprises the following steps: -S9.2.1- Collection in the BDR of the values ​​of the learning data; -S9.2.2- Normalization of training data values; -S9.2.3- Implementation of multiple machine learning and AI prediction models; -S9.2.4- Data separation; -S9.2.5- Training and evaluation loop; -S9.2.6- Plotting the ROC curve for each model and calculating the average AUC (AUC score); -S9.2.7- Selection of the machine learning and AI prediction model based on the best AUC score; -S9.2.8- Selection of training data; -S9.2.9- Prediction of subject Sx belonging to at least one target, using the model selected in -S9.2.7-.

10. Method according to claim 9 characterized in that step -S9.2.8- consists of using “volcano plot” curves and / or the RFE recursive selection algorithm associated with the machine learning and AI prediction model selected in step -S9.2.7-.

11. Method according to at least one of the preceding claims, characterized in that: *a prognosis is established consisting of evaluating the risks of clinical complications linked to a renal problem, a pulmonary problem, an infection, and / or the risks of lethal complications, for a subject Sx to be tested, having undergone a surgical operation, in particular a surgical operation in which clamping and / or blood transfusions and / or extracorporeal blood circulation have been implemented; this prognosis constituting an aid to the medical decision of whether or not to maintain the subject Sx in intensive care; and / or *a therapeutic monitoring is established consisting of evaluating, during a drug treatment of a subject Sx to be tested, the residual level of drug circulating after its administration, on one or more occasions, to the subject Sx to be tested, for one or more durations after the administration or administrations, in particular in oncology;this therapeutic monitoring constituting an aid to the medical decision to modify or not the dose of medication to be administered to the subject Sx to be tested; and / or * therapeutic monitoring is established consisting of evaluating during radiotherapy treatment of a cancer subject Sx to be tested, the; 67 clinical response of tumors and / or the occurrence of clinical complications, in particular necrosis of healthy tissue and / or enteritis, this therapeutic monitoring constituting an aid to the medical decision to modify or not the irradiation doses of the Sx subject to be tested; and / or * a prognosis is established consisting of evaluating, in the context of medical procreation assistance, for an Sx subject to be tested, the success rate of oocyte punctures, and / or fertilizations and / or embryo transfers and / or the occurrence of a pregnancy; this prognosis constituting an aid to the decision of whether or not to freeze oocytes and / or whether or not to transfer oocytes and / or the decision as to the number of embryos to be transferred.

12. Electronic device for implementing the method according to at least one of the preceding claims, the device comprising * modules for implementing the 1st protocol -S9.1-: -M9.1.1- Collection of data in the BDR -M9.1.2- Normalization of the data; -M9.1.3- Selection from the data (i), and possibly the data (ii), of differentiating data; -M9.1.4- Reduction of the dimensionality of the data in 2D or 3D, preferably by a PCA principal component analysis method and / or by a t-SNE (t-distributed Stochastic Neighbor Embedding) method; -M9.1.5- Visualization of the targets on graphs; -M9.1.6- Positioning on the graphic visualization of the targets the DSx data on the subject(s) Sx to be tested, to visualize the positioning of the Sx(s), relative to the targets. * implementation modules of the 2nd protocol -S9.2-: -M9.2.1- Collection in the BDR of the values ​​of the learning data -M9.2.2- Normalization of training data values ​​-M9.2.3- Implementation of multiple machine learning and AI prediction models -M9.2.4- Data separation -M9.2.5- Training and evaluation loop -M9.2.6- Plotting the ROC curve -M9.2.7- Selection of the machine learning and AI prediction model.

13.

14.

15. -M9.2.8- Selection of training data / features -M9.2.9- Prediction of the target of the subject Sx using the model selected in -S9.2.7-. System comprising the electronic device according to claim 12, at least one computer server hosting this electronic device, the BDR and at least one terminal allowing at least one user to communicate with the electronic device, in particular to transmit data DSx relating to a subject Sx to be tested and of the same type as the data (i) & (ii) defined in SO in the method according to one of claims 1 to 11, and to receive in return information, relating to the subject Sx to be tested, to aid in the diagnosis, prognosis and / or therapeutic monitoring of diseases impacting the mineral composition of at least one biological fluid. Computer program comprising instructions for implementing the method according to one of claims 1 to 11, when said instructions are executed by a processor of a computer processing circuit. Data carrier on which a computer program according to claim 4 is recorded.