Diagnostic methods and systems for inflammatory bowel disease (IBD)

Abstract

The present invention relates to a method and kit for the diagnosis of inflammatory bowel disease (IBD) in a subject. The diagnostic method is based on the detection of fecal calprotectin and at least one further fecal biomarker selected from PGRP-S and MMP-8 in a fecal sample taken from the subject. In a more preferred embodiment, the obtained fecal biomarker concentration data is analyzed and classified by a supervised machine learning diagnostic model as having or not having IBD.

Description

Technical Field

[0001] The present invention relates to methods and systems for diagnosing inflammatory bowel disease (IBD).

Background Art

[0002] Inflammatory bowel disease (IBD) is a group of chronic inflammatory diseases of the large and small intestines, including Crohn's disease and ulcerative colitis.

[0003] Calprotectin in feces is currently an established marker in the diagnosis of IBD and is useful for differentiating, for example, IBD from irritable bowel syndrome (IBS).

[0004] Calprotectin belongs to the S100 protein family (S100A8 / S100A9 protein) and is an important pro-inflammatory factor in the innate immune system. Calprotectin acts as a marker protein for endogenous cell- and tissue-damage associated molecular patterns via activation of Toll-like receptor 4, and its presence also indicates the activated state of neutrophils in the intestine.

[0005] However, sensitivity and specificity sufficient to reliably differentiate IBD from IBS without endoscopic confirmation are not always obtained solely with calprotectin-based tests.

[0006] In fact, while studies have provided good performance data for fecal calprotectin testing as a diagnostic marker for IBD, a wide gray zone (approximately 50-300 μg / g) still exists. In this range, the clinical performance of calprotectin alone as a biomarker is insufficient, and endoscopy is recommended (Dulai, Parambir S., et al. "Incorporating fecal Calprotectin in clinical practice for patients with moderate to severely active ulcerative colitis treated with biologics or small molecule inhibitors." The American journal of gastroenterology 115.6 (2020): 885).

[0007] However, endoscopy is an invasive procedure. Therefore, there is a need to identify further non-invasive biomarkers that could help improve the diagnosis of IBD.

[0008] Current technological capabilities include various biomarkers that have been individually proposed for the diagnosis of IBD.

[0009] International Publication No. 2013 / 016425 discloses a method for determining disease activity in a patient who has IBD or is at risk of developing IBD, the method comprising measuring the concentration of neopterin in a fecal sample of the patient.

[0010] International Publication No. 2010 / 045180 lists numerous IBD biomarkers, including those for Crohn's disease and ulcerative colitis, but most of the biomarkers listed lack sufficient experimental evidence to at least reliably establish their actual diagnostic efficacy.

[0011] Soomro, Sanam, et al., "Predicting disease course in ulcerative colitis using stool proteins identified through an aptamer-based screen." Nature Communications 12.1 (2021): 3989, discloses a high-throughput aptamer-based targeted proteomics assay that identifies fecal biomarkers for pediatric IBD. The results of this study showed that each of the four fecal biomarkers—fibrinogen, MMP-8, PGRP-S, and TIMP-2—was positively correlated with the PUCAI and PGA disease severity indices. However, Soomro, Sanam, et al. propose the use of each individually identified biomarker as an alternative to fecal calprotectin in pediatric patients. Furthermore, Soomro, Sanam, et al. do not address the issue of improving the diagnostic accuracy of calprotectin in the gray zone. [Prior art documents] [Patent Documents]

[0012] [Patent Document 1] International Publication No. 2013 / 016425 [Patent Document 2] International Publication No. 2010 / 045180 [Non-patent literature]

[0013] [Non-Patent Document 1] Dulai, Parambir S., et al. "Incorporating fecal Calprotectin in clinical practice for patients with moderate to severely active ulcerative colitis treated with biologics or small molecule inhibitors." The American journal of gastroenterology 115.6 (2020): 885 [Non-Patent Document 2] Soomro, Sanam, et al. "Predicting disease course in ulcerative colitis using stool proteins identified through an aptamer-based screen." Nature Communications 12.1 (2021): 3989 [Overview of the project]

[0014] Given the above background, there is still a need to identify additional IBD biomarkers that, when used in combination with fecal calprotectin, improve the diagnostic accuracy of fecal calprotectin, particularly in the aforementioned gray zone (i.e., approximately 50-300 μg / g), compared to when used alone, thereby avoiding or reducing the need for endoscopy in borderline cases.

[0015] These and other requirements are addressed by the present invention, and in a first aspect of the present invention, an in vitro method for diagnosing a target inflammatory bowel disease (IBD) is provided as defined in claim 1 of the appended claims.

[0016] The present invention is based on the use of fecal calprotectin as an IBD biomarker in combination with peptidoglycan-recognizing protein 1 (PGRP-S) and / or matrix metalloproteinase (MMP-8).

[0017] Peptidoglycan-recognizing protein 1 (PGRP-S) is a pattern recognition protein of the innate immune system that regulates host defense against Gram-positive bacteria.

[0018] Matrix metalloproteinase-8 (MMP-8) is a collagen-cleaving enzyme that is involved in the degradation of the extracellular matrix in normal physiological processes (e.g., embryonic development and tissue remodeling), as well as in disease processes (e.g., arthritis and cancer metastasis).

[0019] To the best of the inventor's knowledge, the combination of calprotectin with PGRP-S and / or MMP-8 has not been disclosed in the prior art, and in particular, has not been disclosed as a solution to the problem of improving the diagnostic performance of in vitro methods for IBD diagnosis.

[0020] According to a first aspect of the present invention, an in vitro method for diagnosing a target IBD is: i. Measure the concentrations of fecal calprotectin and at least one additional fecal biomarker selected from PGRP-S and MMP-8 in fecal samples collected from the subject, and obtain a set of fecal biomarker concentration values ​​characteristic of the subject; ii. Diagnosing IBD in a subject by classifying a set of characteristic fecal biomarker concentrations into at least one class, indicating that the subject has IBD, or a second class, indicating that the subject does not have IBD. This classification is based on multiple sets of fecal biomarker concentration values ​​used in step i, measured in fecal samples of multiple subjects previously diagnosed as having or not having IBD as a result of clinical testing.

[0021] Another aspect of the present invention is a kit for in vitro diagnosis of a target IBD, i. A means for detecting fecal calprotectin in a fecal sample collected from a subject, the means comprising a capture portion that specifically recognizes and binds fecal calprotectin; and ii. A means for detecting at least one additional fecal biomarker selected from PGRP-S and MMP-8 in a fecal sample collected from a subject, the means of ii. comprising a capture portion that specifically recognizes and binds PGRP-S, and / or a capture portion that specifically recognizes and binds MMP-8, the kit.

[0022] A further aspect of the present invention is the use of the kit as defined above for the in vitro diagnosis of IBD in a subject.

[0023] Yet another aspect of the present invention is a computer-implemented method for diagnosing IBD in a subject, comprising receiving a set of fecal biomarker concentration values characteristic of the subject, including the concentration of fecal calprotectin and at least one additional fecal biomarker selected from PGRP-S and MMP-8; and using a supervised learning classifier to classify the received set of fecal biomarker concentration values characteristic of the subject into at least a first class indicating that the subject has IBD or a second class indicating that the subject does not have IBD comprising, where the supervised learning classifier is learning using a training dataset comprising a plurality of sets of concentration values of at least the same fecal biomarkers as the received set obtained from a plurality of subjects, and corresponding classification labels indicating the first class or the second class (where the first class and the second class each indicate, as a result of a clinical trial, that the subject has been pre-diagnosed as having or not having IBD) is a method.

[0024] A further aspect of the present invention is a computer implementation method for training a supervised learning classifier for diagnosing an IBD of a target, Measure the concentration values ​​of fecal biomarkers, including calprotectin and at least one additional biomarker selected from PGRP-S and MMP-8, in each fecal sample of multiple subjects, and obtain each data point containing the fecal biomarker concentration value characteristic of the subject; The training data is generated by assigning a classification label to each data point indicating either a first class or a second class (where the first class and the second class each indicate corresponding subjects among multiple subjects who were previously diagnosed as having or not having IBD as a result of clinical examinations); and The classifier is trained to output classification labels from data points that have been assigned classification labels, based on the training data. This method includes [something].

[0025] A further aspect of the present invention is a data processing device comprising at least one processor designed to perform steps of a computer implementation method for a target IBD diagnosis, and a data processing device comprising at least one processor designed to perform steps of a computer implementation method for training a supervised learning classifier for a target IBD diagnosis.

[0026] A further aspect of the present invention is a computer program product and a computer-readable medium, which includes instructions causing at least one processor to execute a computer implementation method for a subject IBD diagnosis and / or a computer implementation method for training a supervised learning classifier for a subject IBD diagnosis when at least one processor is executed.

[0027] A further aspect of the present invention is a recording medium or computer-readable medium for storing a training dataset and a plurality of data points of the training dataset, wherein each data point includes a concentration value of a fecal biomarker, comprising calprotectin and at least one further fecal biomarker selected from PGRP-S and MMP-8, and a classification label indicating a first class or a second class (wherein the first class and the second class, respectively, indicate that the subject has been previously diagnosed as having or not having IBD as a result of a clinical examination). [Brief explanation of the drawing]

[0028] [Figure 1-5] Not mentioned in the original text. [Modes for carrying out the invention]

[0029] To better understand the present invention, several preferred but non-limiting embodiments thereof will be described by referring to Figure 1 as an example. Figure 1 is a schematic diagram of a system implementing an in vitro method for diagnosing a target IBD according to the present invention.

[0030] As used herein, the term "subject" refers to a person suspected of having IBD or who has been screened for IBD for any reason.

[0031] As used herein, the term "PGRP-S" refers to peptidoglycan-recognizing protein 1, also known as peptidoglycan-recognizing protein (abbreviated form).

[0032] As used herein, the term "MMP-8" refers to matrix metalloproteinase-8 (MMP-8), also known as PMNL collagenase (MNL-CL).

[0033] As used herein, the term "accuracy" refers to the degree to which the test results generated by the method match the true value.

[0034] As used herein, the term "precision" refers to the proportion of true positives among the total number of true positives and false positives classified by a machine learning model.

[0035] As used herein, the term "recall" refers to the proportion of cases classified as true positives out of the total number of cases classified as true positives and false negatives by a machine learning diagnostic model.

[0036] In this specification, the term "AUC" refers to the "Area Under the ROC curve." Here, the ROC (Receiver Operated Characteristic) curve is a performance metric obtained by plotting the "TPR" (True Positive Rate) against the "FPR" (False Positive Rate) in a specific trained machine learning model.

[0037] Fecal samples collected from subjects are generally denoted as S, and a set of fecal biomarker concentrations characteristic of the subject can be obtained from these fecal samples. These fecal biomarkers include calprotectin, as well as at least one additional fecal biomarker selected from PGRP-S and MMP-8. These biomarkers are detected, for example, by immunoassay methods using capture moieties that specifically recognize and bind to fecal calprotectin, PGRP-S, and / or MMP-8.

[0038] According to one embodiment, the method of the present invention includes measuring the concentrations of fecal calprotectin and PGRP-S in a fecal sample of interest.

[0039] According to another embodiment, the method of the present invention includes measuring the concentrations of fecal calprotectin and MMP-8 in a fecal sample of interest.

[0040] In a preferred embodiment, the method of the present invention includes measuring the concentrations of fecal calprotectin, PGRP-S, and MMP-8 in a fecal sample of the subject, and a schematic diagram in Figure 1 shows a system for performing an in vitro method for diagnosing IBD of the subject according to a preferred embodiment.

[0041] The concentrations of the aforementioned fecal biomarkers in a fecal sample may be measured by any suitable method (e.g., a method known in the field of protein detection). Preferably, the concentrations of calprotectin, PGRP-S, and / or MMP-8 are measured by their respective immunoassays. The immunoassays include both homogeneous and heterogeneous assays, as well as both competitive and non-competitive sandwich assays. Immunoassays included in the scope of the present invention may be any suitable format, such as radioimmunoassays (RIAs), chemiluminescent or fluorescent immunoassays, enzyme-linked immunoassays (ELISAs), Luminex-based bead arrays, protein microarray assays, or rapid test formats (e.g., immunochromatographic strip tests). The use of any type of immunoassay format selected within the scope of common technical knowledge of those skilled in the art is included in the scope of the present invention.

[0042] According to a preferred embodiment, the immunoassay for measuring the concentrations of calprotectin, PGRP-S, and / or MMP-8 is a sandwich immunoassay, more preferably a chemiluminescent immunoassay (CLIA).

[0043] Immunoassay methods for measuring the concentrations of calprotectin, PGRP-S, and / or MMP-8 involve the use of capture moieties (e.g., antibodies) that can specifically recognize and bind to each protein.

[0044] As used herein, the term “antibody” includes whole antibody molecules (including polyclonal antibodies, monoclonal antibodies, chimeric antibodies, humanized antibodies, or human antibodies having full-length heavy and light chains), as well as antigen-binding antibody fragments or single-chain variable region fragments. “Antibody fragment” includes any immunoglobulin fragment having the same binding specificity as the corresponding full-length antibody. Such fragments are prepared according to standard methods; see, for example, Harlow and Lane, “Antibodies, A Laboratory Manual”, CSH Press, Cold Spring Harbor, USA, 1988. Examples of antibody fragments include, but are not limited to, F(ab), F(ab')2, F(v), and F(d). Single-chain variable region fragments (scFv) are fusion proteins of the variable regions of the heavy (VH) and light (VL) chains of immunoglobulins, typically linked by a short linker peptide consisting of, for example, 10 to about 25 amino acids. Furthermore, scFVs may be designed as divalent single-chain variable region fragments (e.g., tandem bi-scFvs and diabodies) or trivalent single-chain variable region fragments (e.g., tandem tri-scFvs and tribodies).

[0045] In a preferred embodiment of the method of the present invention, the capture portion is a monoclonal antibody.

[0046] According to the method of the present invention, the specific immunological binding of a capture moiety (e.g., an antibody) to a fecal biomarker is detected via a detectable signal. This signal is generated either directly, for example, by using a labeled capture moiety, or indirectly, for example, via a labeled detection molecule that can specifically bind to the fecal biomarker captured by the capture moiety. Generally, the detection molecule is an antibody against an epitope on the fecal biomarker that is different from the epitope recognized by the capture moiety.

[0047] The detectable label may be any substance capable of generating a visually or effectively detectable signal. Suitable labels for use in the present invention include, for example, fluorescent compounds, chemiluminescent compounds, radioactive compounds, enzymes, enzyme substrates, molecules suitable for colorimetric detection, binding proteins, and epitopes. In practice, embodiments of the methods and kits of the present invention may incorporate any signal molecule or label known in the art.

[0048] In the method of the present invention, depending on the format of the immunoassay, the capture portion that specifically recognizes and binds fecal calprotectin, PGRP-S, and / or MMP-8 may be immobilized on a solid support. Suitable solid supports include, but are not limited to, wells of a microtiter plate, the surface of microparticles (e.g., latex, polystyrene, silica, chelated Sepharose, or magnetic beads), membranes, strips, or chips.

[0049] The solid support is preferably a bead, and more preferably a paramagnetic microparticle (PMP). The capture portion (e.g., antibody) may be covalently bonded to the solid support, or it may be non-covalently bonded by nonspecific binding.

[0050] According to a preferred embodiment of the present invention, quantification of fecal calprotectin, PGRP-S, and / or MMP-8 in a fecal sample of interest is performed by creating a standard curve plotting known concentrations of fecal biomarkers against immunoassay measurements (e.g., optical density (OD) or mean fluorescence intensity (MFI)). Quantification of unknown fecal biomarker concentrations in the fecal sample of interest may be estimated from the standard curve based on detectable signals measured in this unknown fecal sample.

[0051] In the method of the present invention, the concentrations of fecal calprotectin, PGRP-S, and / or MMP-8 in fecal samples collected from the subject may be measured separately or simultaneously.

[0052] If necessary, dilution of the fecal sample is performed by mixing a specific amount of sample (pre-weighed as appropriate) with a predetermined amount of suitable extraction buffer. In this embodiment, the sample dilution ratio is preferably 10 to 1000 times, more preferably 50 to 750 times, and even more preferably about 475 times.

[0053] Suitable extraction buffers for fecal immunoassays are well known to those skilled in the art and include, for example, phosphate, maleate, and TRIS-based buffers.

[0054] In a preferred embodiment of the present invention, fecal calprotectin is present in the fecal sample to be tested at a concentration in the range of about 50 μg / g to about 300 μg / g of the sample. More preferably, the concentration of fecal calprotectin in the fecal sample to be tested is about 50 μg / g to about 250 μg / g of the sample, and even more preferably about 70 μg / g to about 200 μg / g of the sample. Examples of fecal calprotectin concentration values ​​in the fecal samples being tested include 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, and 104. ,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153,154,155,15 6, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167, 168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181, 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195, 196, 197, 198, 199, 200, 201, 202, 203, 204, 205, 206, 207, 2 08, 209, 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237, 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251, 252, 253, 254, 255, 256, 257, 258, 259,Examples include 260, 261, 262, 263, 264, 265, 266, 267, 268, 269, 270, 271, 272, 273, 274, 275, 276, 277, 278, 279, 280, 281, 282, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293, 294, 295, 296, 297, 298, 299, and 300 μg / g. Therefore, in this embodiment, the subjects of the examination are those whose fecal calprotectin concentration values ​​fall within the so-called "gray zone," in which it is not easy to relatively accurately distinguish between IBD and non-IBD cases, and endoscopy is usually required. Therefore, a preferred embodiment of the present invention is the in vitro diagnostic method defined above, wherein the fecal calprotectin concentration measured in the target fecal sample is in the range of approximately 50 μg / g to approximately 300 μg / g, preferably approximately 50 μg / g to approximately 250 μg / g, and more preferably approximately 70 μg / g to approximately 200 μg / g. Examples of fecal calprotectin concentration values ​​measured in the target fecal sample are listed above.

[0055] A further aspect of the present invention is the kit as defined above. In a preferred embodiment, the kit is i. A means for detecting fecal calprotectin in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to calprotectin; ii. A means for detecting PGRP-S in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to PGRP-S; and iii. A means for detecting MMP-8 in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to MMP-8. Includes.

[0056] When the kit of the present invention is used in an immunoassay, the kit may further include means for detecting the specific immunological binding of a capture moiety (e.g., an antibody) to each fecal biomarker.

[0057] Suitable capture components and detection means used in the kit of the present invention are as described above in relation to the method of the present invention.

[0058] The kit of the present invention may further include a solid support (for example, but not limited to, beads, microparticles, nanoparticles, superparamagnetic particles, microtiter plates, cuvettes, lateral flow devices, flow cells, or any surface to which the capture portion can be passively or covalently bonded).

[0059] Further, the use of the kit defined above for performing a method for diagnosing IBD according to the present invention is provided.

[0060] In an example illustrating a preferred embodiment of the present invention, three stool biomarkers, calprotectin, PGRP-S, and MMP-8, are all used, and a set of stool biomarker concentration values ​​is shown as data BM1, BM2, and BM3, which are provided to a data processing device P. The data processing device P includes at least one processor designed to execute instructions for at least one computer program product stored on at least one computer-readable medium in order to implement a diagnostic model based on a supervised learning classifier C. The computer-readable medium may be a tangible computer program carrier stored in or remotely accessible from the data processing device P, or it may be an electronic carrier such as a signal that transmits computer program code from a remote storage location to the data processing device and causes it to be executed locally on the device. This may include, for example, a remote storage location or the data processing device being configured for cloud storage or cloud computing.

[0061] According to a preferred embodiment of the present invention, the supervised learning classifier is a nonlinear classifier (e.g., a classifier based on decision trees), and is preferably a decision tree ensemble classifier.

[0062] Decision tree ensemble classifiers are preferably boosting decision tree classifiers, and more preferably gradient boosting decision tree classifiers (e.g., those generated by the known XGBoost algorithm). XGBoost is a boosting algorithm that combines multiple weak classifiers to construct a single strong classifier, improving performance by sequentially constructing decision trees and adjusting the weights of misclassified samples, thereby making the classifier more robust to outliers and noise.

[0063] In embodiments where all three stool biomarkers, calprotectin, PGRP-S, and MMP-8, are used, the decision tree ensemble may include multiple decision trees, each trained on a different combination of biomarkers. For example, the first decision tree is generated from training data containing only the concentration values ​​of calprotectin and PGRP-S; the second decision tree is generated from training data containing only the concentration values ​​of calprotectin and MMP-8; and the third decision tree is generated from training data containing the concentration values ​​of calprotectin, PGRP-S, and MMP-8. The classification results may be obtained, for example, based on a predetermined voting scheme applied to the classification results of the three decision trees.

[0064] The data processing device P is designed to classify the received set of stool biomarker concentration values ​​into at least a first class C1 indicating that the subject has inflammatory bowel disease, or a second class C2 indicating that the subject does not have inflammatory bowel disease.

[0065] A supervised learning classifier C is pre-trained using a training dataset DS, which contains training data including multiple sets of concentration values ​​for stool biomarkers identical to the received set, obtained from stool samples S1, ..., Sn of multiple different subjects. Each training data is assigned a corresponding classification label L1, ..., Ln indicating whether or not the subject has IBD as a result of clinical testing.

[0066] In one embodiment, the dataset DS is preprocessed before being used to train a supervised learning classifier. For example, to improve the statistical modeling of the data distribution, values ​​below the measurement range may be set to 0, and / or value normalization may be applied.

[0067] In one embodiment, the received set of fecal biomarker concentration values ​​is pre-processed before being used to train a supervised learning classifier. For example, to improve the statistical modeling of the data distribution, values ​​below the measurement range may be set to 0 and / or value normalization may be applied.

[0068] In the method of the present invention, preferably, the clinical examination performed on multiple subjects is an endoscopic examination. Those skilled in the art are well aware of the standard clinical methods for diagnosing IBD.

[0069] In the currently preferred embodiment, where the classifier is a decision tree ensemble, the decision trees and their parameters are automatically adjusted using the training data as input, based on the model's output in the previous epoch. For example, by using a loss function, the classifier outputs classification labels from the corresponding data points to which the classification labels have been assigned. The hyperparameters may be the default parameters provided by the XGBoost algorithm.

[0070] As an example, a decision tree classifier was generated from a training dataset of 83 samples with the following IBD distribution, as determined by clinical testing: 30 samples with IBD and 53 samples without IBD.

[0071] Figures 2a, 2b, and 2c show the distribution of concentration values ​​for three biomarkers in box plots for 83 samples, for subjects without IBD (indicated as N) and subjects with IBD (indicated as Y). A box plot is a type of graph used to show the distribution of a set of data. The box in the plot represents the central 50% of the data, and the median line is drawn inside the box. The whiskers extend from the box and indicate the range of the data, with their longest length corresponding to the interquartile range. More extreme points are displayed as outliers.

[0072] Preferably, the preliminary classification data is divided into a first actual training dataset containing approximately 75% of the collected data (62 samples in this case) and a second test dataset containing the remaining data (21 samples in this case). It is desirable that the training dataset be randomly divided into actual training datasets and test datasets. This ensures that both datasets include features of subjects with IBD (i.e., sets of concentration values) and features of subjects without IBD.

[0073] As an example, the training set available to the inventor constructed the decision tree ensemble described at the end of this specification. The thresholds are particularly dependent on the training data. The rules and thresholds for the root node and decision nodes are specific to the training set used.

[0074] The tree shown in Figure 3 should be considered as just one example of a tree that contributes to a classification process, and it includes a tree ensemble model.

[0075] Specifically, based on the concentration distribution of three biomarkers in 83 samples shown in Figures 2a, 2b, and 2c (fecal biomarkers include calprotectin, PGRP-S, and MMP-8), the decision tree has a root node that evaluates the concentration value of PGRP-S. More specifically, the decision tree compares the input PGRP-S concentration value to the first PGRP-S concentration threshold of 47.92 pg / mL.

[0076] The first branch growing from the root of the decision tree is defined for PGRP-S concentration values ​​smaller than the first PGRP-S concentration threshold and includes a first-level intermediate node that evaluates the calprotectin concentration value against the calprotectin concentration threshold of 154 μg / g, and a second-level intermediate node that evaluates the MMP-8 concentration value against the MMP-8 concentration threshold of 39.7 pg / ml.

[0077] The second branch growing from the root of the decision tree is defined for PGRP-S concentration values ​​greater than the first PGRP-S concentration threshold and includes a first-level intermediate node that evaluates the PGRP-S concentration value in comparison to the second PGRP-S concentration threshold of 70.4 pg / ml, and a second-level intermediate node that evaluates the PGRP-S concentration value in comparison to the third PGRP-S concentration threshold of 62.9 pg / ml.

[0078] Preferably, in another embodiment, a third validation dataset can be separated from the training dataset (for example, 80% of the training dataset is selected for actual training, 10% for testing, and 10% for validation). Classifier validation can be performed in multiple cycles, for example, by exchanging validation data with training data according to a cross-validation procedure with a predetermined number of divisions (k-fold cross-validation). Preferably, the data for the test, training, and validation datasets can be selected manually or automatically from the collected data while maintaining as much of the proportional division for each known class as possible.

[0079] The validation process can be useful for tuning hyperparameters such as tree depth and iteration count.

[0080] In another embodiment, a random forest classifier can be used. In this case, a set of decision trees is used, and each tree is trained on a random subset of the training data and a random subset of its features. By combining multiple decision trees, the algorithm can obtain higher accuracy and generalization results. During prediction, i.e., during the data analysis or diagnostic phase, the class is determined by aggregating the results of all the trees in the forest, for example, by majority vote.

[0081] In yet another embodiment, various decision tree models may be used, and their outputs may be combined based on a predetermined voting scheme.

[0082] As will be apparent to those skilled in the art, a computer implementation method for diagnosing inflammatory bowel disease via a supervised learning classifier, and a computer implementation method for training a supervised learning classifier for diagnosing inflammatory bowel disease (IBD), may be implemented independently of each other, on different processing units, at different locations and at different times, and by different entities, and these are included in the claims.

[0083] Of course, without impairing the principles of the present invention and without departing from the scope of protection of the present invention as defined by the appended claims, the embodiments and implementation details may be broadly modified from those described and illustrated as non-limiting examples.

[0084] (Declaration based on Article 170-2, Paragraph 3 of the Italian Industrial Property Law) This invention was completed in accordance with the provisions of Article 170-2, paragraph 3 of the Italian Industrial Property Law concerning the acquisition of informed consent. [Examples]

[0085] (1. Substances and Methods)

[0086] (1.1 Sample) A subset of 83 clinical stool samples was selected from the study by Campbell et al. (2021) (Campbell, James P., et al. "Clinical performance of a novel LIAISON fecal calprotectin assay for differentiation of inflammatory bowel disease from irritable bowel syndrome" Journal of clinical gastroenterology 55.3 (2021): 239). To address clinically low-accuracy ranges, samples were selected within the range of 10–300 μg / g based on the original stool calprotectin measurements.

[0087] Clinical diagnoses based on endoscopy included IBD in 30 samples (fecal calprotectin (FC) range 25–295 μg / g) and non-IBD in 53 samples (IBS or other GI disorders, FC range 11–265 μg / g).

[0088] (1.2 Assay Description) The IBD diagnostic assay consists of three separate chemiluminescent immunoassays for the quantitative measurement of fecal calprotectin, peptidoglycan-recognizing protein 2 (PGRP-S), and matrix metalloproteinase 8 (MMP-8). The three fecal protein assays are integrated using a computer-based machine learning algorithm to assist in the diagnosis of IBD.

[0089] (1.3 DiaSorin LIAISON® Calprotectin Assay) The DiaSorin LIAISON® calprotectin assay is a sandwich assay using two monoclonal antibodies for capturing and detecting calprotectin. The LIAISON® calprotectin assay was performed on a LIAISON® XL analyzer, a fully automated system with continuous feeding capabilities, according to the manufacturer's instructions. Calprotectin was first extracted from human fecal samples using either the weight method or a LIAISON® QSET device with LIAISON® QSET buffer. In the assay, the extracted sample, calibrator, control, or calibration validater was incubated with assay buffer and paramagnetic particles coated with a monoclonal antibody that specifically recognizes the calprotectin heterocomplex. After incubation, a washing cycle was performed to remove all unbound material. Next, an isoluminol-conjugate monoclonal antibody that recognizes calprotectin was added to the reactants and incubated. Unbound conjugates were removed in a second washing step. Next, the starter reagent was added to initiate the flash chemifluorescence reaction. The light signal was measured as relative luminescence (RLU) using a photomultiplier tube and was proportional to the concentration of calprotectin in the calibrator, control, or sample. All assay steps and incubations were performed using a LIAISON® XL analyzer. The analyzer software automatically calculated the concentration of calprotectin in the sample. This concentration was expressed in units of μg / g.

[0090] These measurements were processed in conjunction with stool PGRP-S and MMP-8 for the purpose of assisting in the diagnosis of IBD.

[0091] (1.4 PGRP-S assay) The PGRP-S assay is a self-sandwich format using polyclonal goat antibodies for PGRP-S capture and detection. PGRP-S was first extracted from human fecal samples using either gravimetric methods or a LIAISON® QSET device with DiaSorin LIAISON® QSET buffer. In the assay, the extracted sample, calibrator, or control was incubated with assay buffer and paramagnetic particles coated with polyclonal goat antibodies that specifically recognize PGRP-S. After incubation, a washing step was performed to remove unbound material. Next, isoluminol-conjugate polyclonal goat antibodies that recognize PGRP-S were added to the reactants and incubated. Unbound conjugates were removed in a washing step. Then, a starter reagent was added to initiate the flash chemifluorescence reaction. The optical signal was measured as relative luminescence (RLU) using a photomultiplier tube and was proportional to the concentration of PGRP-S in the calibrator, control, or sample. This concentration was expressed in units of pg / mL.

[0092] The results of the PGRP-S assay were computationally processed in conjunction with fecal calprotectin and MMP-8 for the purpose of assisting in the diagnosis of IBD.

[0093] (1.5 MMP-8 assay) The MMP-8 assay is a sandwich assay using two monoclonal antibodies for capturing and detecting MMP-8. MMP-8 was first extracted from human fecal samples using either gravimetric methods or a LIAISON® QSET device with DiaSorin LIAISON® QSET buffer. In the assay, the extracted sample, calibrator, control, or calibration validater was incubated with assay buffer and paramagnetic particles coated with a monoclonal antibody (MAB908) that specifically recognizes human MMP-8. After incubation, a washing cycle was performed to remove all unbound material. Next, the isoluminol-conjugate monoclonal antibody that recognizes MMP-8 was added to the reactants and incubated. Unbound conjugates were removed in a second washing step. Then, the starter reagent was added to initiate the fillash chemifluorescence reaction. The optical signal was measured as relative luminescence (RLU) using a photomultiplier tube and was proportional to the concentration of MMP-8 in the calibrator, control, or sample. This concentration was expressed in units of pg / mL.

[0094] These measurements were processed in conjunction with fecal calprotectin and PGRP-S for the purpose of assisting in the diagnosis of IBD.

[0095] (1.6 Prediction using supervised learning classifiers) A gradient boosting decision tree classifier was used as a supervised machine learning diagnostic model, similar to those generated by the XGBoost algorithm. The set of 83 samples was divided into a training set of 62 samples and a test set of 21 samples. The confusion matrix in the test set was used as an index to evaluate the ability to distinguish cases using single biomarkers and combinations thereof. This included precision (the number of predicted IBDs that were actually IBDs), recall (the number of times the model was able to detect actual IBDs), accuracy (the number of times the model was correct overall), and the AUC parameter.

[0096] (2.Results) The individual concentration values ​​for each biomarker were aggregated and computed using a supervised machine learning diagnostic model to obtain a classification of IBD or not IBD. Figure 4 shows the magnitude of the influence each feature has on the model output: calprotectin, PGRP-S, and MMP-8 contribute to the model in the order shown in Figure 4.

[0097] Table 1 summarizes the diagnostic capabilities of each marker individually and in combination. [Table 1]

[0098] Figure 5 shows the confusion matrices from which the performance parameters in Table 1 were derived for calprotectin (A), PGRP-S (B), MMP-8 (C), and combinations of all three markers (D).

[0099] Overall, the results above indicate that all combinations of fecal biomarkers tested significantly improved the diagnostic accuracy of fecal calprotectin alone. These preliminary data are based on a very small number of tests, and it is not possible to identify the combination of fecal biomarkers with the highest diagnostic accuracy at this time. However, as shown in Figures 4 and 5, each of the three biomarkers (fecal calprotectin, PGRP-S, and MMP-8) significantly contributes to the diagnostic accuracy of the present invention's method, suggesting that the optimal combination would include all three.

[0100] The experimental data obtained suggests that a combination of three biomarkers (fecal calprotectin, PGRP-S, and MMP-8), using a gradient-boosting decision tree generated by a supervised machine learning model, preferably the XGBoost algorithm, can accurately identify IBD patients even in areas with low clinical accuracy, thereby reducing the number of unnecessary endoscopies.

[0101] (Example of a decision tree ensemble model)

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

change

Claims

1. An in vitro method for diagnosing the target inflammatory bowel disease (IBD), i. Measure the concentrations of fecal calprotectin in fecal samples collected from the subject, as well as at least one additional fecal biomarker selected from peptidoglycan-recognizing protein 1 (PGRP-S) and metalloproteinase 8 (MMP-8), to obtain a set of fecal biomarker concentrations characteristic of the subject; ii. Diagnosing IBD in a subject by classifying a set of characteristic fecal biomarker concentrations into at least one class, indicating that the subject has IBD, or a second class, indicating that the subject does not have IBD. Includes, A method in which the classification is performed based on multiple sets of concentration values ​​of the same fecal biomarkers used in step i, measured in fecal samples of multiple subjects previously diagnosed as having or not having IBD as a result of clinical examination.

2. The in vitro method according to claim 1, wherein step i comprises measuring the concentrations of fecal calprotectin and PGRP-S.

3. The in vitro method according to claim 1, wherein step i comprises measuring the concentrations of fecal calprotectin and MMP-8.

4. The in vitro method according to claim 1, wherein step i comprises measuring the concentrations of fecal calprotectin, PGRP-S, and MMP-8.

5. The in vitro method according to any one of claims 1 to 4, wherein the measured fecal calprotectin concentration value in the fecal sample collected from the subject is in the range of 50 μg / g to 300 μg / g of the fecal sample.

6. The in vitro method according to any one of claims 1 to 5, wherein the classification in step ii is performed based on a supervised learning model trained with multiple sets of concentration values ​​of the same fecal biomarkers used in step i, measured in fecal samples of multiple subjects previously diagnosed as having or not having IBD as a result of clinical examination.

7. The in vitro method according to any one of claims 1 to 6, wherein the clinical examination is an endoscopic examination.

8. An in vitro method according to any one of claims 1 to 7, wherein fecal calprotectin is measured by an immunoassay.

9. The in vitro method according to any one of claims 1 to 8, wherein PGRP-S and / or MMP-8 in stool are measured by their respective immunoassays.

10. This is a kit for the in vitro diagnosis of IBD, i. A means for detecting fecal calprotectin in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to fecal calprotectin; and ii. Means for detecting at least one further fecal biomarker selected from PGRP-S and MMP-8 in a fecal sample collected from a subject, A kit comprising, ii. a means comprising a capture portion that specifically recognizes and binds to PGRP-S in feces, and / or a capture portion that specifically recognizes and binds to MMP-8 in feces.

11. i. A means for detecting fecal calprotectin in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to calprotectin; and ii. A means for detecting PGRP-S in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to PGRP-S. The kit according to claim 10, comprising:

12. i. A means for detecting fecal calprotectin in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to calprotectin; and ii. A means for detecting MMP-8 in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to MMP-8. The kit according to claim 10, comprising:

13. i. A means for detecting fecal calprotectin in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to calprotectin; ii. A means for detecting PGRP-S in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to PGRP-S; and iii. A means for detecting MMP-8 in a fecal sample collected from a subject, comprising a capture portion that specifically recognizes and binds to MMP-8. The kit according to claim 10, comprising:

14. The capture portion that specifically recognizes and binds to calprotectin is a monoclonal antibody against calprotectin, preferably conjugated to a solid support. The kit according to any one of claims 10 to 13.

15. The capture portion that specifically recognizes and binds to PGRP-S is a monoclonal antibody against PGRP-S, preferably bound to a solid support. The kit according to any one of claims 10, 11, or 13.

16. The capture portion that specifically recognizes and binds to MMP-8 is a monoclonal antibody against MMP-8, preferably conjugated to a solid support. The kit according to any one of claims 10, 12, or 13.

17. A kit according to any one of claims 10 to 16, comprising instructions for use for performing the method described in any one of claims 1 to 9.

18. Use of the kit according to any one of claims 10 to 17 for the in vitro diagnosis of the target IBD.

19. A computer implementation method for diagnosing IBD, To receive a set of fecal biomarker concentration values ​​characteristic of the subject (wherein the set of fecal biomarker concentration values ​​includes fecal calprotectin and at least one additional fecal biomarker selected from PGRP-S and MMP-8); Using a supervised learning classifier, the received set of stool biomarker concentration values ​​is classified into at least one class indicating that the subject has IBD, or a second class indicating that the subject does not have IBD. Includes, The supervised learning classifier in question Multiple sets of fecal biomarker concentration values ​​obtained from multiple subjects, at least the same as the received set, and A corresponding classification label indicating Class 1 or Class 2 (where Class 1 and Class 2 indicate, respectively, that the subject had or was pre-diagnosed as having or not having IBD as a result of the clinical trial). A method that trains using a training dataset that includes [specific data / features].

20. A computer implementation method for training a supervised learning classifier to diagnose a target IBD, The concentration values ​​of fecal biomarkers, including calprotectin and at least one additional fecal biomarker selected from PGRP-S and MMP-8, are measured in each fecal sample of multiple subjects, and data points containing fecal biomarker concentration values ​​characteristic of each subject are obtained; The method involves generating training data by assigning a classification label to each data point indicating either a first or second class (where the first and second classes represent corresponding subjects from a group of subjects previously diagnosed as having or not having IBD as a result of a clinical trial); and The classifier is trained to output classification labels from data points that have been assigned classification labels, based on the training data. Methods that include...

21. The method according to claim 19 or 20, wherein the supervised learning classifier is a nonlinear classifier.

22. The method according to claim 21, wherein the nonlinear classifier is a classifier based on a decision tree.

23. The method according to claim 22, wherein the classifier using decision trees is a decision tree ensemble classifier.

24. The method according to claim 23, wherein the decision tree ensemble classifier is a boosting decision tree classifier.

25. The method according to claim 24, wherein the boosting decision tree classifier is a gradient boosting decision tree classifier.

26. The method according to any one of claims 19 to 25, wherein the fecal biomarker comprises calprotectin and PGRP-S.

27. The method according to any one of claims 19 to 25, wherein the fecal biomarker comprises calprotectin and MMP-8.

28. The method according to any one of claims 19 to 25, wherein the fecal biomarker comprises calprotectin, PGRP-S, and MMP-8.

29. Setting values ​​below a predetermined measurement range to zero; and / or Apply value normalization. The method according to claim 19, further comprising pre-treating the received set of fecal biomarkers.

30. The generation of training data is Setting values ​​below a predetermined measurement range to zero; and / or Apply value normalization. The method according to claim 20, comprising a pretreatment step including the following.

31. The method according to any one of claims 19 to 30, wherein the fecal calprotectin concentration is in the range of 50 μg / g to 300 μg / g.

32. A data processing apparatus comprising at least one processor designed to perform a step of the method according to any one of claims 19 to 31.

33. A computer program product that includes instructions causing at least one processor to perform at least one of the methods described in any one of claims 19 to 31 when the program is executed on at least one processor.

34. A computer-readable medium that, when executed on at least one processor, includes instructions causing at least one processor to perform at least one of the methods described in any one of claims 19 to 31.

35. A training dataset containing multiple data points, where each data point is Concentration values ​​of fecal biomarkers, including calprotectin and at least one further fecal biomarker selected from PGRP-S and MMP-8; Classification labels indicating Class 1 or Class 2 (where Class 1 and Class 2 indicate, respectively, that the subject has or has been previously diagnosed with IBD as a result of a clinical trial). A training dataset that includes [the specified data].

36. A recording medium or computer-readable medium storing a training dataset containing multiple data points, wherein each data point is Concentration values ​​of fecal biomarkers, including calprotectin and at least one further fecal biomarker selected from PGRP-S and MMP-8; A classification label indicating either Class 1 or Class 2 (where Class 1 and Class 2, respectively, indicate that the subject was previously diagnosed as having or not having IBD as a result of a clinical trial), Recording media or computer-readable media including [the specified text].

Images