RU2782359C2 - Systems and methods for improvement of diagnostics of diseases, using measured analytes - Google Patents

Abstract

FIELD: biotechnology.

SUBSTANCE: invention relates to biotechnology, in particular, systems and methods for diagnostics of diseases, such as prostate cancer, breast cancer, lung cancer, ovarian cancer, and their stages, are disclosed. In accordance with defined embodiments, disclosed systems and methods provide for collection of patient’s samples, calculation of concentrations and proximity indicators of biomarkers, and use of the results of such calculations for creation of a training sample model, which is used for establishment of a correlation of biomarker concentrations and proximity indicators with disease diagnoses and disease statuses (for example, cancer stages). In accordance with defined embodiments, simple regression, area maximization under ROC curve, topology stabilization, and correlation analysis by special proximity relate to used correlation methods.

EFFECT: obtaining systems and methods for diagnostics of diseases.

20 cl, 26 dwg, 5 tbl, 4 ex

Description

Link to related applications

[0001] This application claims benefit under U.S. Provisional Application No. 62/542,865, filed Aug. 9, 2017, which is incorporated herein by reference in its entirety.

[0002] A related patent application, International Application No. PCT/US2014/000041, filed March 13, 2014 (incorporated herein by reference in its entirety), describes methods for improving disease prediction using an independent variable for correlation analysis that is not is directly the concentration of the analytes being measured, but is a calculated value called a "proximity score" that is calculated from the concentration but also normalized to a specific age (or other physiological parameters) to eliminate age bias and non-linearities in how concentration values shift or shift from physiological parameter (eg, age, menopausal status, etc.) when the disease status shifts from the absence of the disease to the presence of the disease.

The field of technology to which the present invention relates

[0003] The present invention relates to methods for improving the accuracy of diagnosing diseases and to related diagnostic tests that involve the use of correlation of measured analytes with binary output results (for example, the absence of disease or the presence of disease), as well as higher order results (for example, one of several phases of the disease).

State of the art of the present invention

[0004] Methods for determining correlations, where three or more independent variables are used to establish a correlation of binary output results (such as the presence or absence of a given disease), usually involve the use of a method for determining correlation by spatial proximity (also called a cluster or neighborhood search method), regression method and wavelet methods. In the case of predicting a disease, total blood or serum composition is measured and a correlation is attempted using these concentration values as independent variables to predict the status of various diseases. In the case of the status of a given disease, where the result is either "presence of the disease" or "absence of the disease", the method of logistic regression is usually used. Other techniques involve the use of, for example, genetic algorithms. The predictive power of these methods is highly dependent on the constituent analytes chosen for the method. Those skilled in the art will recognize that many analytes and parameters that are likely to have predictive power do not improve diagnostic and analytical power in practice.

[0005] Regression methods involve using trends in the explanatory variables to find a correlation with the output results. The linear method is based on linear trends, while the logistic regression is based on logarithmic trends. When predicting biological diseases, logistic regression is most often used to determine the output results.

[0006] The spatial proximity group method allows you to explore a variable correlation topology for a group of similar results. The spatial proximity method has the advantage that it can find correlations where the trends are not related, but have local inverse changes in the topologies in the trends. However, this method is highly non-linear and sensitive to very locally variable results with small measurement errors, which can provide more accurate prediction in biological applications. In addition, both methods discussed here can be combined with the spatial proximity method applied on a small scale to create a consolidated general regression method.

[0007] However, some independent variables, which, from the point of view of logic, should have a correlation, in practice, there is no predictive trend. Therefore, there is a need to develop an approach that improves diagnostic accuracy by using patient-specific and population-specific variables that have not been found to date to provide useful information for diagnosing various disease statuses.

[0008] Many studies have been conducted to find biomarkers that, alone or in combination, can predict the status of various diseases with sufficient reproducibility and predictive power for use in a clinical setting. This study had limited or no success. High frequency proteins (HAPs) have been extensively investigated in search of a single protein that may allow such a prediction. Many examples have been found, but none have low enough false negative rates to allow patients to be screened for disease with such a marker.

[0009] As a result, such individual biomarkers are used only for therapeutic tracking, with the exception of PSA in the case of prostate cancer. This test requires that the concentration that indicates that a biopsy would be appropriate has a high deviation from normal to reduce false negatives, resulting in very high levels of false positives. Up to 80% of men who are shown to need a biopsy are actually negative for prostate cancer. It has also been found that in some cases, DNA markers are very good at detecting a cancer subtype, but again not suitable for screening for the same reasons as HAP as noted above.

[0010] Proteomic approaches using a variety of proteins have also been explored. This work was again focused on HAP or high level effector proteins. This work has been dominated by multiplex protein measurement methods such as immunoassays, arrays, and mass spectrophotometry. There was some success in very early work in the case of ovarian cancer. However, a problem with all of these methods is that many of the selected proteins do not have a strong correlation with progression from healthy to diseased status (and many have no known biological association with disease status, for example, as is usually the case with mass spectrometry). Moreover, mass spectrometry suffers from a serious disadvantage in that the spectrophotometer examines the whole serum sample for protein levels, and therefore training the correlation search algorithm is difficult. In the case of mass spectrometry, a whole whey sample can contain more than 200 protein varieties and have 10,000 mass spectrometric peaks.

[0011] In the field of diagnostics, there is also a need for techniques that involve the use of lower frequency proteins that are more suitable for diagnostic purposes than HAPS, as well as analytical techniques that involve the analysis of low frequency biomarkers.

[0012] The field of diagnostic medicine has long sought a simple and accurate serum blood test to detect cancer and to detect whether the cancer is severe or latent. For example, the current prostate-specific antigen (PSA) test for prostate cancer suffers from a very high false positive rate, with a true false negative rate as high as 1 in 10 men. This test has a predictive power of approximately 57%. What's more, men who are diagnosed with low-grade prostate cancer may not need treatment for many years or until the end of their lives. Today, this diagnosis can only be accurately obtained with a PCa biopsy. The current PSA test sends everyone without exception (90%, one in 10 missed) with a PSA level above 4.0 ng/ml to a biopsy, and only approximately 20% of them have some form of PCa, regardless of the Gleason score. In addition, men with low-severity PCa are at risk of progressing to higher severity later in life, and the only reliable way to accurately diagnose this is with more biopsies. Additional tracking biopsies are not acceptable to the medical community due to cost and are not acceptable to the patient due to pain and side effects. Thus, ongoing monitoring of men with low-grade PCa is done with an intermittent PSA test, followed by digital rectal examinations (DRE) and sometimes CT examinations. In many cases, prophylactic treatment is carried out, removal of the prostate gland, even when this is not medically necessary. This patent discloses a new serum test that distinguishes men without PCa from men with high PCa and detects men with low PCa who may worsen later. In addition, it discloses a blood test that can recognize cancers with a solid tumor at an early stage, such as lung cancer or breast cancer, or the stage of cancer.

[0013] The current screening test for PSA was approved in the mid-1980s and is not currently patented. A new so-called 4K test offered by OPKO under the name "Lab developed Test" does not have regulatory approval. It is designed to detect men with high-severity PCa while separating the condition from low-severity PCa. Generally, PCa with high severity is considered to correspond to a Gleason score (obtained from biopsy) of 7(4+3) or higher (8, 9, or 10), while a score for low severity is considered to be 7(3+ 4) or below. The PSA test for detecting men with PCa of any severity has approximately 57% predictive power, or for a sensitivity of 90%, the false positive rate is approximately 80% (1 in 4 positives is actually negative). This 4K evaluation test has a predictive power of approximately 64%. Thus, for 1 in 10 false negatives, there are approximately 50% of false positives, or approximately 5 out of 10 are actually negative. This is the current state of testing for the diagnosis of PCa in medicine today.

[0014] Currently, there are no regulatory approved methods for detecting diseases such as lung cancer and breast cancer using a simple blood test. Moreover, these diseases can only be assessed in terms of severity by biopsy. The authors of the present invention also proposed additional tests to assess the stage of tumor development using, again, active cytokines in the tumor microenvironment using blood serum as a substitute for these proteins.

[0015] To alleviate these and other shortcomings of the prior art, this document illustratively describes a new test using active cytokines in the tumor microenvironment, with blood serum acting as a substitute for these proteins.

Brief description of the drawings

[0016] To a greater extent, the present invention and many of its attendant advantages will be readily understood as it is more fully understood in connection with the following detailed description when considered in connection with the accompanying figures, where

[0017] in FIG. 1 is a graph showing fluctuations in biomarker concentrations as assessed by the Gleason score for prostate cancer;

[0018] in FIG. 2 is a graph showing fluctuations in biomarker concentrations as assessed by the Gleason scale for lung cancer;

[0019] in FIG. 3 is a graph showing the average increase in the concentrations of biomarkers corresponding to the stages of development of breast cancer;

[0020] in FIG. 4 is a graph plotting sensitivity versus false positive rate ("ROC") of VEGF for aggressive prostate cancer versus no cancer;

[0021] in FIG. 5 is a graph showing the TNFα ROC curve for aggressive prostate cancer versus no cancer;

[0022] in FIG. 6 is a graph showing the ROC PSA curve for aggressive prostate cancer versus no cancer;

[0023] in FIG. 7 is a graph showing the ROC IL 6 curve for aggressive prostate cancer versus no cancer;

[0024] in FIG. 8 is a graph showing the ROC IL 10 curve for advanced lung cancer versus early lung cancer;

[0025] in FIG. 9 is a graph showing the ROC curve of IL 6 for advanced lung cancer versus early lung cancer;

[0026] in FIG. 10 is a graph showing the VEGF ROC curve for advanced lung cancer versus early stage lung cancer;

[0027] in FIG. 11 is a graph showing blind test results with two samples that failed the topology instability test and were corrected with a non-congruent algorithm in accordance with an embodiment of the disclosed diagnostic method;

[0028] in FIG. 12 is a graph showing the results of a breast cancer clinical study, in this case, the training set cancer score values are shown for the training group model I using 10 two-marker planes in accordance with an embodiment of the disclosed diagnostic method;

[0029] in FIG. 13 is a graph showing the results of a breast cancer clinical study, in this case, the training set cancer score values are shown for the training group model II using 105 two-marker planes in accordance with an embodiment of the disclosed diagnostic method;

[0030] in FIG. 14 is a chart showing results with actual diagnosis for blind specimens involved in a clinical study according to an embodiment of the disclosed diagnostic method;

[0031] in FIG. 15 is a diagram showing a two-marker plane for one of ten such planes, showing proximity score values of two biomarkers used in accordance with an embodiment of the disclosed diagnostic method;

[0032] in FIG. 16 is a diagram showing a two-marker plane with training sample data points according to an embodiment of the disclosed diagnostic method;

[0033] in FIG. 17 is a diagram showing a two-marker plane without training sample data points according to an embodiment of the disclosed diagnostic method;

[0034] in FIG. 18 is a diagram showing a two-marker plane with a shaded area where the effect of immune system response is reduced according to an embodiment of the disclosed diagnostic method;

[0035] in FIG. 19 is a diagram showing a two-marker plane with a shaded area where the impact of topology stability issues is reduced in accordance with an embodiment of the disclosed diagnostic method;

[0036] in FIG. 20 is a diagram showing a two-marker plane with a shaded area, where the impact of unreliable measurement results by known analysis is reduced in accordance with an embodiment of the disclosed diagnostic method;

[0037] in FIG. 21 is a graph showing the results of blind tests with two samples that failed the topology instability test and were corrected with a non-congruent algorithm in accordance with an embodiment of the disclosed diagnostic method;

[0038] in FIG. 22 is a flowchart showing the general logical path followed by the software of the present invention, in accordance with an exemplary embodiment;

[0039] in FIG. 23 is a flowchart showing the process of building a training sample (or diagnostic model) and then processing blinded diagnostic score values to evaluate disease risk or disease free status;

[0040] in FIG. 24 shows a typical population distribution, in this case for the cytokine, interleukin 6 (IL 6);

[0041] in FIG. 25 is a chart showing the conversion of biomarker concentration to proximity score (one type of pseudo-concentration); and

[0042] in FIG. 26 shows a typical hardware diagram used in implementing the software of the present invention, in accordance with an illustrative embodiment.

Brief summary of the present invention

[0043] Without limiting the foregoing, in accordance with a preferred embodiment, the present invention relates to improving the predictive power and diagnostic accuracy of methods for predicting disease status using multivariate (multivariate) correlation methods. Such methods include proteomic, metabolic and other methods that involve the determination of the levels of various biomarkers found in body fluids and tissue samples.

[0044] Various embodiments provided by the inventors and discussed in this application include the use of metavariables, in particular by using methods that correct for the influence of analytes of measured biomarkers on the correlation index. Such metavariables can be determined based on special knowledge about the response of the immune system and knowledge about possible measurement errors. Such methods can be applied either to the design of the training sample model or to the blind samples being diagnosed.

[0045] According to one aspect, the present invention relates to a method for diagnosing a disease, comprising the steps of a) determining concentrations of at least three predetermined analytes in a blind sample from a subject; b) selecting one or more subject-related metavariables that vary in the subject-related population among members of that population who are known to either have the disease or not; c) converting analyte concentrations as a function of one or more characteristics of the distribution of the population and one or more metavariables to calculate the closeness index to which each analyte is represented; d) comparison of proximity measures with a training sample model for proximity measures determined from members of the population who are known to either have the disease or not; and e) determining whether the results of the comparison indicate that the subject has a disease. It is contemplated that step (a) of determining concentrations (or levels) of predetermined analytes may be performed at a different time and location from the rest of the method steps. Likewise, one or more of the other steps of the method may be wholly or partially practiced at other times and locations. Therefore, the authors of the present invention consider as their invention a method that only involves steps (b) to (f).

Detailed disclosure of the present invention

[0046] In describing the preferred embodiments of the present invention illustrated in the drawings, specific terminology will be used for clarity. However, the present invention is not intended to be limited to the specific terms chosen, and each specific term is to be understood to include all technical equivalents that function in a similar manner in a similar purpose. For purposes of illustration, several preferred embodiments of the present invention have been described, it being understood that the present invention may be embodied in other forms not specifically shown in the drawings.

[0047] In the context of the present application, in order to better describe the preferred embodiments of the present invention, specific terminology is used, which is defined below.

[0048] The term "analytical sensitivity" is defined as three standard deviations above a zero calibration standard. Diagnostic representations are not considered accurate for concentrations below this level. Therefore, clinically relevant concentrations below this level are not considered accurate and are not used for diagnostic purposes in the clinical laboratory.

[0049] "Analyte measurement at baseline in an individual" means the measurement of a set of biomarkers of interest to convert an individual patient from disease free to disease status, which is measured in the same individual several times over a period of time. A baseline analyte measurement by disease free status is obtained if the individual patient does not have a disease, and alternatively a baseline analyte measurement by disease status is obtained if the individual patient has a disease. Such baseline measurements are considered representative of the individual patient and may be useful in diagnosing the transition from disease-free to disease status of that individual patient. The measurement of an analyte at baseline in the case of disease status may be useful in diagnosing disease in a second or more occurrence of the disease in such an individual.

[0050] "Biological sample" means a tissue or biological fluid, such as blood or plasma, that is taken from a subject and in which concentrations or levels of diagnostically informative analytes (also called markers or biomarkers) can be determined.

[0051] "Biomarker" or "marker" means a biological component of a biological sample of a subject, which is usually a protein or metabolic analyte measured in a body fluid, such as a serum protein. Examples include cytokines, tumor markers, and the like. The present invention also contemplates other traits as "biomarkers" and "markers" including, but not limited to, height, eye color, geographic factor, environmental factors, etc. In general, such traits will include any measurements or attributes that vary collectively and remain measurable, determinable or observable.

[0052] A "blind sample" is a biological sample taken from a subject without a known diagnosis of a given disease, and for which a prediction of the presence or absence of such a disease is needed.

[0053] "Disease-associated functionality" is a characteristic of a biomarker that either has an effect, continuing or worsening a disease, or has an effect on an organism, stopping the progression of a disease. In the case of cancer, the tumor will act on the body to continue and increase the growth of blood vessels, and the immune system will increase pro-inflammatory actions to destroy the tumor. These biomarkers are distinct from tumor markers, which have no disease-related functionality but enter the circulatory system and can therefore be measured. Examples of functional biomarkers include interleukin 6, which enhances the action of the immune system, or VEGF, which is secreted by the tumor, causing local growth of blood vessels. Whereas CA 125 is a non-functional example. It is a structural protein located in the eye and female reproductive tract that is not used by the body to kill the tumor or is used by the tumor to help the tumor grow.

[0054] The "limit of detection" (LOD) is defined as 2 standard deviations from a concentration value greater than the concentration value of a "zero" calibration standard. Typically, the zero calibration standard is run 20 or more times to get an accurate representation of the standard deviation of the measurement results. The results of determining the concentration below this level are considered as zero or absent, for example, when viruses or bacteria are detected. In the context of the present invention, 1.5 standard deviations can be used if the samples are run in duplicate, although 20 replicates are preferred. Diagnostic views that require a single concentration number are generally not displayed below this level. The measurement results at the detection limit level are statistically at the 95% confidence level. Prediction of disease status using the methods discussed herein is not based on a single concentration, and it has been shown that prediction is possible at measurement levels that are below the concentration LOD.

[0055] "Proteins with low frequency of occurrence" are proteins in serum that are present at very low levels. The definition of this level in the literature is not clearly defined, but when used in this description, the level will be below about 1 picogram/milliliter in serum or plasma and other body fluids from which samples are taken.

[0056] "Metavariable" means information that is specific to a given subject, with the exception of concentrations or levels of analytes and biomarkers, but which is not necessarily individualized or unique to such a subject. Examples of such metavariables include, without limitation, the subject's age, menopausal status (pre-, peri- and post-), and other conditions and characteristics such as puberty, body weight, geographic location or region of residence of the patient, geographic source of the biological sample, percentage of body fat, age, race or racial origin or era of time.

[0057] "Population distribution" means the range of concentrations of a particular analyte in biological samples of a given population of subjects. Specific "population" means, without limitation, individuals selected from a geographic region, a particular race, or a particular gender. And the characterization of the distribution of the population selected for use, as described herein, further provides for the use of two separate subpopulations within this larger defined population, which are members of the population and which have been diagnosed with the presence of the specified disease (the subset of the presence of the disease) and the absence of the disease ( subset of absence of disease). The population can be any group for which a disease status prediction is desired. Moreover, appropriate populations are contemplated to include those subjects who have a disease that has progressed to a particular clinical stage as compared to other stages of disease progression.

[0058] "Population distribution characteristics" can be defined within the population distribution of a biomarker, such as the mean concentration of a particular analyte, or its median concentration, or the dynamic range of concentration, or how the population distribution falls into groups that can be recognized in distinct peaks, since the degree of up- or down-regulation of the various biomarkers and metavariables of interest depends on the onset and progression of the disease as the patient experiences a biological transition or progression from disease-free to disease-positive status.

[0059] "Predictive power" means the mean value of sensitivity and specificity for a diagnostic assay or test, or one minus the total number of false predictions (both false negatives and false positives) divided by the total number of samples.

[0060] "Proximity Score" means a surrogate or substitute value for the concentration of the measured biomarker and, in fact, is a new independent variable that can be used in a diagnostic correlation analysis. Proximity score is related to and calculated from the concentration of measurable biomarker analytes, such analytes having predictive power for a given disease status. Proximity score is calculated using a metavariable-adjusted population distribution characteristic of interest to convert the actual measured concentration of a predictive biomarker in a given patient to be diagnosed as disclosed in International Publication No. WO 2017/127822 and International Publication No. WO 2014/158287. "Proximity score" and "pseudo concentration" have the same definition and can be used interchangeably.

[0061] "Specificity" is a true/false positive test result. Mathematically, this is one minus the number of false positives measured by the test divided by the total number of true negatives measured.

[0062] A "non-congruent training sample model" (or "secondary algorithm") is a secondary training sample model that uses a different phenomenological data reduction method so that individual points on two-marker plane grids have a low probability of an unstable state in the primary correlation model training sample and this secondary algorithm.

[0063] A "proximity correlation method" (or cluster search or cluster analysis) is a method for determining a correlation between independent variables and a binary output with the independent variables plotted along orthogonal axes. Prediction for blind samples is based on proximity to the number (3, 4, 5, or more) of the so-called "training set" data points for which the output is known. The binary output score is based on the total distance calculated from the blind point in the multidimensional space to the points of the opposite training sample output. The smallest distance determines the estimate of an individual data blind point. The same analysis can be carried out on two-marker planes marked with a multidimensional grid, while the individual score of the two-marker plane is combined with the score of other planes to obtain an overall score. This use of slices or 2D orthogonal projections in space can reduce computation time.

[0064] "Training sample" refers to a group of patients (typically 200 or more to achieve statistical significance) with known biomarker concentrations, known metavariable values, and known diagnosis. The training sample is used to determine the values of the "proximity" axes of the "two-marker" planes, as well as the evaluation grid points obtained from the results of the spatial proximity analysis, which will be used to evaluate individual blind samples.

[0065] A "training sample model" is an algorithm or group of algorithms generated from a training sample that makes it possible to evaluate blind samples for a predictive output regarding the likelihood that a subject (or patient) has or does not have a disease. The "training sample model" is then used to calculate scores for clinical and diagnostic blinds. For this purpose, an arbitrary range score is obtained that indicates the percentage probability of having a disease or not having a disease, or some other predetermined indicator preferred by the medical professional who is diagnosing the patient.

[0066] "Sensitivity versus False Positive Rate (ROC) Curve" is a graphical way to represent the effectiveness of the signal transmission method used to decide whether there is a trade-off between false positives, false negatives and detection signal strength. In this graphical representation, the ordinate of the graph represents the sensitivity of the testing method, and the abscissa represents the false positive rate. For biomarkers (or signals) with an upward effect to the point of origin of the disease, the curve will be above the 45° zero line, starting at the origin (0,0) of the graph and going to the upper right corner of the graph (1.0,1.0). The area under the curve shows how well the biomarker makes predictions.

[0067] "'Area under the curve' (AUC) for the ROC curve" is the area between the characteristic curve of the biomarker and the abscissa. For a completely useless biomarker, the AUC would be 0.5 and its area under the 45° baseline would be as above. The ideal test has an AUC of 1.0 and the curve runs from the origin on the ordinate to the 100% sensitivity point and then on the ROC curve to the 1.0,1.0 point in the upper right corner.

[0068] The "tumor microenvironment" is bathed in tumor interstitial fluid (TIF) and is the cellular environment in which the tumor exists, including surrounding blood vessels, immunocytes, fibroblasts, inflammatory cells from the bone marrow, lymphocytes, signaling molecules, and extracellular matrix.

[0069] A "tumor marker" is a protein marker that enters the TME or blood supply that has no apparent function, is a marker of either tumor growth due to tumor secretions, or a marker of tumor suppression by the immune system.

[0070] In recent years, there has been increasing interest in cancer immunotherapy research in the tumor microenvironment (TME), which provides an ideal research platform for the development and improvement of new therapeutics and represents a potentially huge treasure trove of diagnostic content. The TME that is bathed in tumor interstitial fluid (TIF) is the cellular environment in which the tumor exists, including surrounding blood vessels, immunocytes, fibroblasts, inflammatory cells from the bone marrow, lymphocytes, signaling molecules, and extracellular matrix.

[0071] TIF is also the transport fluid that links the tumor (and TME) to the blood supply and is important because it is the "battlefield go-between" for active proteins that the immune system uses in an attempt to suppress the tumor, or that the tumor expresses to help your growth. These competing proteins or cytokines, which are constantly at war with each other, fall into several functional categories of low-level signaling proteins: pro- and anti-inflammatory, anti-tumor (or causing cell apoptosis), angiogenesis and vascularization.

[0072] Despite being recognized as a potential source of rich diagnostic information, the development of the TIF assay as a cancer screening method is not moving forward because sampling this fluid is very difficult, and to do so means that the location of the tumor, which means whether the tumor itself exists. Detection of the presence of TME/TIF, and thus malignancy, is quite difficult without this information. This requires a more readily available clinical diagnostic fluid, such as blood serum, coupled with the analysis of a variety of proteins, known as proteomics, that could conceivably correlate with the presence or absence of disease. There are some difficulties with serum in this regard, as it is more of a reflection of a combination of conditions in the patient's body than a direct way to detect the presence of active TME (and therefore a tumor).

[0073] The present disclosure discusses a method for analyzing specific cytokines present in blood serum as an accurate substitute for proteins active in TME and TIF. The method involves several steps, including two proprietary methods called proteomic noise suppression and multivariate (or spatial) correlation. With the method described, a precise substituent can be made for the actions of proteins occurring in TIF and thus the method is useful for detecting the presence of active TME in the body and hence the tumor. Essentially, this method isolates the TME signature in serum and indicates the presence (or absence) of active TME, indicative of the presence of an active tumor. In addition, this method allows the modulation of these proteins to be measured, which provides valuable information about the status of the tumor, the degree of aggressiveness and stage, as well as information about the progress of the immune system in suppressing the tumor.

[0074] Biomarkers of interest

[0075] Biomarkers of interest in the present disclosure are pro-inflammatory (interleukin 6, IL 6, or others); anti-inflammatory (interleukin 10, IL 10, or others.) antitumor or tumor-lysing cytokines (tumor necrosis factor alpha, TNFα, or others) and circulating growth factors such as those that stimulate angiogenesis (interleukin 8, IL 8, or others) and stimulate vascularization cytokines (vascular endothelial growth factor, VEGF, or others). These are cytokines with a directly related functionality to the response of the immune system to a tumor or the effect of a tumor on the body. Vascularization stimulating factors, VEGF, are the action of a tumor to stimulate the growth of the circulatory system into a growing tumor mass. The tumor inhibitory factors, TNFα, are the action of the immune system to promote tumor cytolysis (apoptosis), and the pro-inflammatory factor IL 6 mediates the action of the entire immune system. Anti-inflammatory IL 10 is secreted by the tumor into the interstitial fluid of the tumor to suppress the immune system. Finally, angiogenesis stimulating factors such as IL 8 are secreted by the tumor to increase vascularity in the surrounding tissue.

[0076] In general, cancer is a pro-inflammatory disease in which factors such as IL-6 are up-regulated. However, in the three cases described herein, the tumor at a later stage secretes an anti-inflammatory cytokine into the interstitial fluid of the tumor (and hence into the blood). This has been shown to occur in advanced cancers, stage 3 or 4, in the lung and breast, and at higher Gleason scores (8, 9, or 10 Gleason). At this point, the anti-inflammatory action tends to suppress the pro-inflammatory response of the body's immune system. In some cases, in advanced breast cancer, the response in the form of angiogenesis is also suppressed. Finally, the vascularization-stimulating effect of the tumor increases, as would be expected, as the tumor increases in size in stage 3 or 4 breast or lung cancer and in prostate cancer with a Gleason score of 8, 9, and 10. All of these activities occurring in the tumor microenvironment can be ascertained by taking serum samples from the body and applying the methods described in International Publications No. WO 2017/127822 and No. WO 2014/158287.

[0077] In the particular case of cancer, a high degree of interest in recent therapeutic research has been focused on the so-called "tumor microenvironment" (TME) for the development of treatments. It is believed that a fruitful way to develop these treatments is the suppression or upregulation of proteins active in tumor interstitial fluid (TIF) and occurring in TME. The proteins in TIF that have been found to be good indicators generally belong to the five functional groups of cytokines: pro-inflammatory or anti-inflammatory, anti-tumor (or cell apoptotic), angiogenesis and vascularity.

[0078] Measuring the activity of these proteins can give a better idea of tumor activity and therapeutic effect. For example, treatments that stimulate or inhibit the activity of proteins can be tracked in TIF to determine effectiveness. If appropriate for therapeutic purposes, and when cancer is known to exist, do not collect TIF samples for diagnosis. Since the presence of TIF (and the presence of TME) by definition means that the patient has an active tumor with a known location, its use as a diagnostic tool becomes irrelevant. In addition, the use of these proteins for diagnostics in their presence in other body fluids such as serum or urine has not been considered, since the problem of proteomic noise has made them unusable so far.

[0079] The present document describes systems and methods for obtaining an accurate substitute for the TME activity that active proteins in TIF use with a readily available replacement fluid, in this case serum (but other fluids such as urine are possible). It should be noted that serum is a combination of whole body conditions (referred to as "proteomic noise") and is not tumor specific. The methodology we propose also eliminates proteomic noise, making an accurate assessment of the patient's condition possible.

[0080] In general, the systems and methods disclosed herein include: 1) selecting active TIF proteins that are indicative of conditions in the TME, 2) measuring these proteins in a serum substituent, 3) suppressing proteomic noise to accurately identify cancer-associated activity proteins, 4) then performing a correlation method that enhances the actions of these proteins in a multidimensional matrix, and 5) evaluating the activity of the protein indicative of the presence or absence of cancer and, if present, the stage of its development. This is done primarily to create a training sample representing the population as a whole, which serves as a reference against which individual samples are then compared to determine their status: either affected by the disease or not having the disease.

[0081] Combined actions of biomarkers

[0082] These cytokine biomarkers are very active in high-grade prostate cancer and, when compared to levels in "healthy" men, are highly up- or down-regulated and therefore very good indicators of disease status. It should also be noted that they are active in lung and breast cancer. In FIG. 1, 2 and 3 show such an effect as the tumor progresses. It should be noted that in non-small cell lung and prostate cancer, as seen in FIG. 1 and 2, IL 6 is down-regulated in advanced cancer or high Gleason scores of 8, 9, or 10. It should also be noted that in both cases, when going from low grade lung cancer or low Gleason prostate cancer, increased amounts of interleukin 10 secreted by the tumor lead to downregulation of IL 6 levels. It should also be noted that the secretion of IL 10 in the interstitial fluid of the tumor and therefore into the blood is associated with a poor prognosis for the patient. This usually means that there is an advanced stage of breast cancer. Thus, the combination of IL 6 and IL 10 in the correlation analysis of disease status is improved by the combination of pro-inflammatory and anti-inflammatory cytokines. Moreover, it should be noted that vascularity-influencing cytokines continue to be generally upregulated as the tumor progresses to a later stage or becomes more aggressive.

[0083] Three of these biomarkers have unique ROC curve characteristics that are not common to tumor biomarkers. They have a flat part with 100% sensitivity for certain lower levels of biomarker concentrations. They are also characterized by rather large areas under the curve (AUC), indicating that they are very good biomarkers for this disease, ie. prostate cancer with high severity (PCa) compared with non-PCa disease. One of them has a straight vertical section going up the ordinate from [0,0], indicating that samples in this signal range should have PCa with zero false positives.

[0084] In the scientific literature there is a limited number of publications on several selected biomarkers mentioned in this description. And there is nothing associated with the detection of PCa with high severity in comparison with the general population, that is, patients without PCa. In the case of VEGF, the literature does mention the upregulation of biomarkers, but nothing is indicated in relation to the stage of the cancer or, in particular, the Gleason score. Most of the literature is limited to the use of VEGF as a prognostic agent for the treatment of men with already diagnosed PCa. TNFα also has nothing to do with distinguishing the actions of biomarkers associated with tumor stage or, in particular, tumor Gleason score. The results of scientific studies of interleukin 6 give similar information. It is known that for PCa with low severity, some literature mentions a slight positive regulation of the level of the biomarker. Our measurements do not support this, as little downregulation is seen in low PCa, but very strong downregulation is seen in high Gleason PCa, making this cytokine, along with others, a strong indicator of the presence of high PCa. gravity. Much of the literature is devoted to the use of these biomarkers as prognostic factors in men with PCa and in vitro protein expression from PCa cell lines, as well as to the study of ways to suppress expression (especially VEGF) for therapeutic purposes.

[0085] ROC curves for prostate cancer

[0086] VEGF

[0087] FIG. 4 shows the ROC curve for VEGF in the aggressive form (Gleason score 7 (4+3), 8, 9 and 10). It should be noted that a large flat part of the ROC passes through the upper boundary, where the sensitivity is 100%. At VEGF concentration levels at or below about 50 pg/mL of VEGF, there is no high severity PCa, and no such case has been identified. The AUC for this biomarker when comparing the presence/absence of this disease status is 0.87. In addition, the unique formulation with no false positives below 50 pg/mL makes it a very good candidate as a biomarker for high-severity "PCa" versus "non-PCa" disease, as levels below approximately 50 pg/mL do do not point to PCa.

[0088] TNFα

[0089] Comments on TNFα are similar to those described regarding the nature of the ROC curve for the aggressive form (Gleason score of 7 (4+3), 8, 9 and 10) as seen in FIG. 5. In this case, the AUC is 0.85, and again, the high and same breakpoint does not produce false negatives below approximately 6.5 pg/mL. For TNFα, a portion of the curve is also visible that has a zero false positive rate (abscissa) for samples greater than approximately 9.85 pg/mL. There are no false positives in this region.

[0090] PSA

[0091] Comments on the PSA are similar to those described regarding the nature of the ROC curve for the aggressive form (Gleason score of 7 (4+3), 8, 9, and 10) as seen in FIG. 6. In this case, the AUC is 0.85, and again, the high and same breakpoint does not produce false negatives below about 2 ng/mL. The ROC curve for overall PSA analysis for PCa with any Gleason score is shown in green for reference (shown labeled "All PCa").

[0092] IL6

[0093] In contrast, IL 6 is strongly downregulated in the aggressive form (Gleason score of 7 (4+3), 8, 9, and 10), with an AUC approximately twice that of the current PSA required for detection of PCa, as seen in FIG. 7, in the general population (the curve must be inverted to account for this negative regulation). It can be speculated that perhaps high-severity PCa is effective in suppressing the immune system, but this is not the subject of this review. The fact is that a strong negative regulation is observed for this biomarker. A limited number of literature sources indicate a small positive regulation in total PCa. According to our measurements for PCa with any Gleason score, a slight negative regulation is visible. However, with a high Gleason score, this cytokine is strongly downregulated. The general population of PCa has approximately 80% of cases of low severity, so when sampling for PCa, the characteristics of the group will be dominated by low severity. This downregulation is likely caused by the secretion of an anti-inflammatory cytokine (IL 10) as the tumor progresses to an aggressive form with a Gleason score of 7 (4+3) or higher.

[0094] ROC curves for lung cancer

[0095] IL 10

[0096] FIG. 8 shows the ROC curve for interleukin 10 in a low grade separation (stage 1 and 2 from later stage 3 and 4 non-small cell lung cancer). It should be noted that there is a positive regulation during the transition from the early stage (1st and 2nd) to later stages (3rd and 4th). This is consistent with the downregulation of interleukin 6 and is caused by the anti-inflammatory action of the tumor secreting IL 10 into the tumor microenvironment and then into the bloodstream.

[0097] IL6

[0098] The ROC curve for IL 6 is shown in FIG. 9 and again for the case of non-small cell lung cancer in the early stages (1st and 2nd) compared with the late stages (3rd and 4th). As shown in FIG. 9, this action of IL 6 is inhibited by the anti-inflammatory action of the tumor.

[0099] VEGF

[0100] The ROC curve for VEGF is shown in FIG. 10, which shows upregulation of a vascularization stimulating factor that occurs in other cancers as the tumor grows and progresses to advanced stages.

[0101] A test for aggressive or late-stage cancer versus non-aggressive or early-stage cancer

[0102] These biomarkers can be taken together to develop a very simple proteomic algorithm to track men with low-severity prostate cancer with a Gleason score of 5, 6, or 7(3+4) for transition to high-severity PCa with a score of Gleason score of 7 (4+3), 8, 9, or 10. In addition, these biomarkers distinguish early-stage cancer, stage 1 or 2, from stage 3 or 4. Combining IL 6 and IL 10 with opposite actions can give (with a simple correlation technique such as logistic regression) 80% predictive power. The addition of proteomic noise suppression and the proximity correlation method will yield predictive power values of 90%. Adding VEGF action to the biomarker panel will improve predictive power to 95% or more.

[0103] Test for an aggressive form of prostate cancer in comparison with men without cancer

[0104] In fact, VEGF by itself will produce a test with 76% predictive power, 100% sensitivity, and 76% specificity (24% false positives). This simple model will simply exclude samples diagnosed as absent of PCa in concentration ranges where the ROC curve excludes them, and will again include samples with PCa in those areas where it is included in the ROC curve. It will then use a simple breakpoint count and the number of positive and negative scores for each biomarker that is not included in the exclusion or inclusion criteria. The number must be greater than 3 out of 4 for cases not previously excluded or included. This simple model generates a 100% representative sample set of 100 PCa-diagnosed specimens with a high Gleason score (defined as 7 (4+3) or higher) and 100 PCa-free specimens. The combination of VEGF, IL 6, TNFα and PSA will give a predictive power of 90%. In addition, this test will be predictive of "no cancer" for men with elevated PSA levels, but not cytokines. These men have benign prostatic hyperplasia or another non-cancerous condition of the prostate and constitute the bulk of the many false positives on the current PSA screening test for prostate cancer. A test that includes these cytokines overcomes this problem of false positives.

[0105] Cancer staging prediction

[0106] The data obtained during the study of breast cancer under the supervision of the Herzen Institute in Moscow made it possible to predict the stage of breast cancer with high accuracy using the equipment and reagents described below. 189 breast tissue samples were obtained with stage information (from 0 to 4). Measurements were made for the tumor marker PSA and the four mentioned cytokines of proinflammatory (IL 6), antitumor genesis (TNFα), angiogenesis (IL 8) and vascularization (VEGF) cytokines. In this case, the goal was to evaluate each sample for possible staging information obtained from the biopsy. All methods of determining correlation are binary in nature and cannot be implemented without some manipulation evaluation of the four different outputs. Therefore, the stage groups were combined into binary groups representing all the stage groups; 1 plus 2, 3, 4; 2 plus 1, 3, 4; 3 plus 1, 2, 4, and 4 plus 1, 2, 3. All four groups were modeled and evaluated using the age normalization, noise suppression, and proximity correlation methods described in International Publication No. WO 2017/127822 and International Publication No. WO 2014/158287. A score for each individual sample was then calculated using each individual group score for each sample, together with a weighting on the contribution of each to that group (1 or 1/3). This model gave 99% accuracy.

[0107] Several methods have been described herein to improve the predictive power of conventional proteomics correlation methods for disease diagnosis. They include: 1) using the metavariable and proximity score values to establish correlation, and 2) using special knowledge about the stability of the topology and measurement characteristics during the analysis to correct for the influence of the two-marker plane in the training sample model. In addition, methods are described for detecting and correcting stability issues for blind samples unique to a particular training set model using a non-congruent training set model. In addition, methods for finding and correcting non-disease related conditions that partially mimic the training set model for the status of a given disease are described. All of these methods are complementary and can be used together. For example, adjusting the training sample model for areas of high probability of instability cannot completely eliminate this issue from blind predictive calculations, and therefore both methods can be used to improve predictive power. The present inventors have found that a combination of these methods can give a predictive power of over 95%, and the breast cancer study discussed in Example 1 below gives a predictive power of over 98% (100% sensitivity, 97.5% specificity).

[0108] EXAMPLE 1: clinical study evaluating a blood test for breast cancer

[0109] The performance of the OTraces BC Sera Dx test kit and the OTraces CDx Immunochemistry Instrument system ( www.otraces.com ) was evaluated in a breast cancer risk experiment. The test kit measured concentrations of five very low level cytokines and tissue markers, and the kit used a training set model that was designed as described above to calculate CS1 and CSq scores for breast cancer risk assessment. The proteins measured were IL-6, IL-8, VEGF, TNFα and PSA. The experiment consisted of measuring approximately 300 patient samples, separated by approximately 50% between biopsy-diagnosed breast cancers and 50% of patients presumed to be disease-free (or in this case free of breast cancer). From this group, the biopsy results for 200 specimens were divided exactly into 50% of cases with no disease status and 50% of cases with breast cancer, and each group was further divided into specific age groups.

[0110] The results of the sample analysis were used to develop a training sample model that predicted disease status. The remaining samples (approximately 110) were then processed as blanks using a training sample model to obtain final numerical estimates of the risk of having cancer and these estimates were transmitted to the main clinical center. These blank scores were then analyzed at the clinical center to evaluate the clinical accuracy of the results.

[0111] Two diagnostic models were developed for this experiment and are referred to herein as Algorithm I and Algorithm II. For both algorithms, the spatial proximity analysis method was used. The age of the participants was not used as an independent variable, but as a metavariable to convert the measured concentrations into new independent variables, referred to herein as proximity scores, which were directly used in the correlation analysis. The difference between Algorithm I and Algorithm II was the number of new independent variables used in determining the correlation. Algorithm I used five variables of proximity measures in a ten-dimensional cluster space. The lower limit of Algorithm I had two dimensions, and it was based not on a specific method, but on the fact that correlation is generally determined. Correlation, in fact, involved more than one dimension. The upper limit of Algorithm I was theoretically infinity, but in practice it was limited by the calculation time and the statistical power of the study. The cluster space could be seen with the naked eye using a projection or slices of such a multidimensional space, allowing to see it in a two-dimensional two-marker plane. According to this exemplary embodiment of Algorithm I, there are ten such planes.

[0112] Algorithm II uses ten times as many explanatory variables as created, so that there are approximately 100 two-marker planes. It was expected that 200 samples would be enough for the training set model to closely model the population. A secondary or incongruent training set model was developed based on the same set of training data of 200 samples. The training sample model is the main estimation method used to describe the results in the present description. An incongruent training sample model was used to arbitrate cancer status scores calculated with the primary training sample model that were considered unstable, i.e. estimates that remained in the region of topological instability. Although the incongruent training set model was somewhat less accurate on blind samples, it could still arbitrate for the underlying training set model and thus improve predictive power.

[0113] The above spatial proximity analysis method has significant advantages over logistic regression in that it can accommodate highly non-linear trends in the explanatory variables used to produce the calculated output. The output is either disease status or disease absence status (in this case, cancer or cancer absence) and is based on proximity scores of the results calculated using the training sample model. The disadvantage of this method is that highly non-linear regions can be associated with very large topology slopes. Thus, an unknown (or blind) sample may be at a sharp peak or in a deep steep trough, leading to amplification of small errors in the calculated proximity metrics. The present inventors evaluated the stability of the calculated scores using their proprietary stability test and then used Algorithm II to arbitrate the results obtained with Algorithm I for samples that were observed to be stable.

[0114] FIG. Figures 11, 12 and 13 show the results obtained with Algorithm I for the training set. The model itself consisted of 10 two-marker planes with 40,000 topology points, each of which was assessed for disease free status and disease status (in this case, breast cancer) using a spatial proximity method. These figures show the ability of the model to divide into two samples: the absence of cancer and the presence of cancer. The model had to be built from very close to or preferably exactly 50% to 50%, or very close to one of the two output statuses. In addition, the method included the use of age as a transformative metavariable. The training samples included samples distributed across all age groups of interest. The model (FIG. 12) for Algorithm I was built from data from 100 healthy women and 98 women with breast cancer. In the summary table in Fig. 12 shows numerical results where N=198 is the number of samples. CI denotes correct samples and FI denotes false samples, and 4 samples were considered indeterminate. The secondary training set model was developed to recognize the four uncertain samples that resulted from the application of the primary training set model. This model is an incongruent training sample model. This secondary model used the same training data as the main model. In FIG. 13 shows the results of calculations by the incongruent model of the training set. Algorithm II gave 100% separation with over 60 split points.

[0115] Blind test results in a breast cancer study

[0116] FIG. 14 shows the results for blind samples evaluated in a clinical study. The results show 100% sensitivity and 97.5% specificity. The oncologists at the Clinical Research Center have determined the value of the diagnostic transition so that all samples positive for breast cancer are correctly identified. So, two samples with the status of the absence of the disease were designated positive for cancer. From a medical point of view, this was justified by the fact that all samples evaluated as positive would be sent to the next stage of diagnosis - mammography. Many women were not referred for mammograms because they lived quite far from facilities with medical equipment. However, their blood could be taken remotely from a clinical lab and sent on ice to a lab in a major city.

[0117] EXAMPLE 2: use of the meta-variable "age" to improve diagnostic accuracy

[0118] Table 1 (below) shows the tabulated results of examining samples from 868 subjects for breast cancer.

Table 1 Summary of diagnostic accuracy in breast cancer

State Group Correctly identified indefinite Misidentified Mammary cancer 495 98.0% 1.0% 1.0% healthy women 373 98.0% 0.5% 1.5%

[0119] Table 2 (below) shows the results of comparing different methods of calculating the correlation. The standard method, logistic regression, gave only 82% predictive power. This improved the standard spatial proximity analysis, giving approximately 88% predictive power in linear form and 90% predictive power in logarithmic form. The methods described herein, using metavariable and weighting approaches, topology stability normalization, immune system response pooling, and analysis weight normalization, combined with blind instability testing and correction with a non-congruent algorithm, yielded predictive power greater than 97%. .

Table 2. Comparative predictive power of the results of calculations of the correlation of diseases

[0120] EXAMPLE 3: application of the metavariable "age" to improve diagnostic accuracy in an ovarian cancer study

[0121] Table 3 (below) shows the results of a study of 107 women with or without ovarian cancer using the metavariable method described in the embodiments of the present invention. This study did not utilize all of the improvements in predictive power described herein, but still achieved a relatively superior predictive power of approximately 95%.

Table 3 Summary of diagnostic accuracy in ovarian cancer

State Group Correctly identified indefinite Misidentified ovarian cancer 51 94.1% 3.9% 0.0% healthy women 56 96.4% 3.6% 0.0%

[0122] EXAMPLE 4: application of the metavariable "age" to improve the accuracy of diagnosis in prostate cancer

[0123] Table 4 (below) shows the results of a study of 259 men suffering from prostate cancer or benign prostatic hyperplasia (BPH) using the metavariable method described herein. This study also did not use all of the improvements in predictive power described herein, but still achieved a relatively superior predictive power of approximately 94%. It should be noted that BPH is the most common pathological condition that gives false positive results in the current PSA test for prostate cancer. Men with BPH have approximately 4 out of five positive results on conventional prostate cancer diagnosis, resulting in the majority of prostate cancer biopsies being negative. The metavariable method allows these misdiagnoses to be corrected, as discussed above.

Table 4 Summary of diagnostic accuracy in prostate cancer

State Group Correctly identified indefinite Misidentified prostate cancer 111 93.70% 0.90% 5.40% Benign prostatic hyperplasia/hypertrophy 148 95.90% 0.00% 4.10%

[0124] For the above results in Examples 3 and 4 (for ovarian cancer and prostate cancer, respectively), no metavariable or influence adjustment methods (LOD, subsets, groups, and instability) were used, nor was the blinded stability method used.

[0125] To further improve predictive power, these age- or group-adjusted concentrations were constrained to normalize them and reduce or eliminate spatial bias (also known as spatial bias) when clustering multivariate clustered marker plots for spatial proximity analysis. See, for example, FIG. 15 showing a two marker plane for IL-6 and VEGF. For the diagnostic panel of tests for breast cancer on five biomarkers, there were ten such planes. In this case, the calculated values of the proximity index were normalized and biased to obtain arbitrary values from zero to twenty, while the outlier values of very high concentrations were subjected to a high degree of compression.

[0126] Each of the two-marker projections of the multidimensional marker planes at the same normalized distance relative to the concentrations derived from the age/group analysis was compressed and normalized to the age-adjusted means as well as age-adjusted (or population) subgroups .

[0127] Improving the Predictive Power of the Training Sample Model with Adjustable Influence Levels on the Two-Marker Plane

[0128] Typically, the two-marker plane was evaluated with binary numbers for disease-free status and disease status (eg, +1 and -1). The proximity metric method described herein can be modified to further improve predictive power by selectively adjusting the influence levels of the two binary numbers. The following methods were implemented in the training sample model and, after their creation, were fixed in this model.

[0129] FIG. 16 and 17 show projections of one two-marker plane for the case of five biomarkers used to predict the presence of a disease, in this case breast cancer, using five markers: IL-6, IL-8, TNFα, VEGF and PSA. In FIG. 16 shows a training sample model with data used to evaluate grid points on a plot using a proximity analysis method. In FIG. 17 shows the training sample model without data. It is a training set model. The training data used to create the model was not needed because each of the 40,000 grid points was evaluated, and the blind sample was evaluated by placing it on the grid. On the topology, one could see a red positive signal in the presence of cancer, and a blue signal meant a negative status for cancer. When calculating the overall score in this case, the points on the grid with the status of the absence of the disease were set equal to +1, and the points on the grids with the status of the presence of the disease (cancer) were set equal to -1. Each dual marker in this five-biomarker example was analyzed in a five-orthogonal space, for which FIG. 16 shows one projection in two dimensions. This graph shows the topology of the various immune system response subsets. In this case, all grid points (2000×2000 or 40000 in this case) were evaluated in the usual way, and the assigned value is -1 for the positive status of the presence of the disease (breast cancer), and for the status of the absence of the disease is +1. This two-marker plane was normalized with distance to proximity score and age metavariable as noted above.

[0130] In FIG. 18 shows the same two-marker model and, in addition, immune response groups (see FIG. 24) within the gray areas. The effect of the gray areas was corrected for the fact that each gray area occupied had a slightly different effect on the likelihood that a patient had disease free status or disease status. This adjustment could be made either by assessing the person with a training sample check, or by rigorous computer multivariate incremental analysis. These adjustments improved the training sample model. For the two outputs, which were disease status or disease absence status, two separate two-marker planes were created. In this case, blind data points in immunoresponse group IV were more likely to indicate disease status, and the impact increased slightly (absolute value) (eg, when the score changed from -1 to -1.1). The actual value of this increment is preferably determined by computer analysis or possibly by rigorous manual methods. This method is suitable for spatial proximity (also known as pseudo-concentration) correlation analysis, but other means can be used to obtain a similar result. Such methods of weighting impact on disease can provide an improvement in predictive power of approximately 1%. This is very important for predictive power values greater than 95%.

[0131] In FIG. 19 again shows the same two-marker plane with the gray area circled in the complex area of the non-linear, rapidly changing topology of disease status relative to disease-free status. Such areas could be identified by plugging the values of a blind test sample with noise included (say, +/- 10%) into the model, and then plugging in the measured amount of noise. Most of these blind spots did little to change the assessment of disease status (in this case, cancer). However, it was possible to find some grid points that, after this kind of noise adjustment, abruptly changed from an assessment of the absence of disease to the presence of disease. These were areas where most or all of the 2-marker planes had a rapidly changing topology that overlapped the high-dimensional common 2-marker planes. By carefully reducing the influence in these areas, you can increase the weight in several relevant two-marker planes that have noisy data on the wide plane, without coming close to changing the boundaries of the output results. It has been shown that erroneous predictions can be corrected using this method. In the above case, the influence of the red areas related to cancer status shifted downwards (in absolute value), for example from -1.0 to -0.9. Alternatively, the blue areas related to disease-free status shifted from +1.0 to -0.9. The level of optimal bias could be determined by rigorous computer analysis.

[0132] Analysis noise can affect the accuracy of the correlation analysis. This noise can be especially problematic at levels within or below the detection limit using such an analysis. This noise can also be mitigated by reducing the influence of measured points for individual biomarkers that are in these unstable zones. In FIG. 20 again shows the two-marker plane for PSA and IL-6 for the breast cancer diagnostic panel. Areas within the gray box in the lower left corner of the figure are below the generally accepted limit of detection (LOD) of the assay. The generally accepted LOD is defined as two standard deviations of 20 zero calibration standards plus the average of twenty zero calibration standards. Statistical confidence for such values at this level is 95% within two standard deviations, and of course the measurement confidence decreases as the measured sample goes below the LOD. These data can still provide useful information, but should be applied to the analysis with less impact. In this case, the effect on the base points of the blind sample in the gray area was reduced, for example, from +1.0 to -0.9, for grid points of the training sample model within the gray area. This increased the influence of the base points for this test sample, which exceeded the limit of detection on their two-marker planes. The above methods are complementary and can be used together.

[0133] Methods for improving predictive power by testing blind samples for instability

[0134] Once the training set model was completed and fixed, it was used to calculate scores of cancer status for blind patient samples. The present inventors have used two preferred methods for obtaining cancer status scores. The first, called the linear method (CS1), took the topology location estimate (+1 or -1) multiplied by the predictive power for that two-marker plane. They were then added, scaled and shifted to give a score from 0 to 200. A second score, called the q score (CSq), was calculated using the square root of the sum of squares of the same values. According to this second method, attention was focused on differences in individual two-marker scores and is useful in making a final diagnosis by the physician.

[0135] Topology instability still persisted in the two-marker planes due to the highly non-linear nature of the proximity correlation method, and could not be completely eliminated. In accordance with other aspects of the present invention, a stability test and techniques can be applied to a blind dataset that take into account included noise. And an incongruent training sample model can be used to make an arbitral award or adjust estimates of cancer status. For this aspect of the present invention, a fixed noise level (eg, plus or minus 10%) is included for each set of blind data from a patient. If a set of blind samples were taken from approximately 100 patients, then the actual computer run to create a training sample model was performed on a set of 300 samples, each performed in triplicate (raw data plus noise and minus noise). The resulting set of triplicate data was then tested for stability (a is -10%, b is +10%, and c is the raw data). Table 5 (below) shows the result of the stability test for the data obtained during the clinical study. It should be noted that three of the samples showed very high instability in cancer status scores. All samples 138, 207, 34 and 29 showed a very high rate. This metric (the lower the better) should capture both the degree of score shift and especially whether or not the score shifts from healthy to cancer status or vice versa during prediction. These blinded datasets have a high risk of being misleading in making a predictive diagnosis.

Table. Topology instability test results

[0136] An incongruent training sample model can be used to arbitrate on datasets from "at risk" patient samples that fail a noise test. These points are at risk due to the inevitable measurement noise, either accidentally or systematically associated with extreme topology instability due to the fact that the blind sample data point lies on a very steep slope on almost all, if not all, two-marker planes. , so small perturbations give large deviations in the result of the estimate. Table 5 shows samples with noise turned on. Each sample has three values: 1) plus noise, 2) minus noise, and 3) raw data without noise. For these samples, cancer status scores are shown that go from having disease to not having disease and back again with +-10% noise included. The data for these samples in this case was assessed as unstable. The level of instability was not precisely defined and adjustments could be made for various levels of noise inclusion. In this case, they were corrected for +-10% noise and stability scores greater than 200 (it should be noted that the stability score and the cancer status score are two completely different numbers with different values).

[0137] Using such a non-congruent second algorithm (algorithm II), an arbitration decision could be made on the noise of the measurement results. The incongruent algorithm used to determine could be used to correct these sets of "at risk" patient samples, even if it had slightly less predictive power than the main algorithm, as it increased the likelihood that the point was correct. In this case, two of them were corrected (see Fig. 21); sample 138 had a missing status score of 85 and was adjusted to 195 with the incongruent algorithm (this point was stable using Algorithm I), sample 34 had a score of 102 (linear) and was adjusted again to 198 with Algorithm II. Samples 29 and 207 did not change when using the incongruent algorithm.

[0138] The incongruent training set model (Algorithm II) used 105 two-marker planes and was incongruent with the main training set model (Algorithm I) in that the same samples appeared to be stable in the stability test with Algorithm II. Testing of the incongruent model of the training sample was performed in the same way as in the case of the main model of the training sample. It should be noted that a logistic regression method could also be used to calculate these sample scores. Algorithm II has a high predictive power, which is why it was used. The training sample arbitrage model can be used if its predictive power is less (but preferably not less than 50% of the predictive power) than that of the main algorithm, provided that it gives a relatively correct result without instability. It should be noted that the correction is important for blind samples that fail the noise test. All of these samples did indeed have cancer status with high scores. Eight of the ten two-marker planes for these samples lay on a topology with very highly unstable grid points. Thus, the scores were at risk and were indeed incorrect (one was incorrect and the other was indeterminate and had scores of 100/120). In this case, one sample was corrected with an improvement in predictive power from 97% to 98%, which was a very significant reduction in error (by 50%). One sample, although of indeterminate status, was changed to cancer status and was also adjusted.

[0139] A method for improving the predictive power of a disease status-adjusted binary output by eliminating an independent status that partially mimics one of the statuses of the output obtained in the main disease status analysis

[0140] The analysis of spatial proximity usually involves the use of three or more independent variables, often the concentration of proteins in the patient's blood serum. The correlation calculation algorithm can only affect the binary output, i.e. no disease status or disease status, but it produces a continuous estimate that is more closely related to the probability that the actual output is two binary states. In some cases, there are other conditions that are nominally classified as disease-free status that partially mimic disease status in the population distributions of the biomarkers used. In some of these cases, this "MIMIC (mimic)" status of the absence of disease can lead to a false positive outcome of the correlation analysis. The solution to eliminate such a false positive result is to create an additional new correlation analysis, completely separate from the analysis of the status of the presence or absence of the disease. This new correlation analysis preferably uses the same measured data for biomarkers as in the case of correlation for the absence or presence of the disease, or it may use some or completely different biomarkers. This new correlation analysis results in a "MIMIC disease free" or "disease present" status, or at least an estimate that allows a judgment to be made about the actual condition of the patient. An indeterminate or near-transitional score in the case of a disease-free or disease status analysis, combined with a very low or high score on the MIMIC correlation of disease-free or disease-positive status, can help the practitioner improve the judgment of disease status and reduce false positives.

[0141] An example of such a situation where the state of the absence of a disease mimics the state of the presence of a disease is a non-malignant condition such as benign prostatic hypertrophy (BPH). In such a condition, high levels of at least one biomarker used to diagnose prostate cancer will typically be observed. For example, a biomarker such as prostate-specific antigen (PSA) will be elevated in men with HPR as well as those with prostate cancer. Table 4 shows that an additional method of correlation analysis can help to recognize men with HRD and men with prostate cancer, and similarly, using the same biomarkers but a different training sample model, men who are presumed to be in the disease-free status can be recognized. , and men with confirmed prostate cancer in the status of having the disease. In a small proportion of men, a false positive result will be obtained using a training sample model of disease-free versus cancer status, but can be recognized by a training sample model of BHP status versus cancer status. In these cases, two scores, one for presumed disease-free versus cancer status and the other for BHP versus cancer status, will help the physician or other healthcare professional decide which next stage of diagnosis to choose. For example, in the case of an overall score (in the case of either CS1 or CSq) of 0 to 200 for both models, a score of 110 for "PROSTATE CANCER FREE STATUS OR PROSTATE CANCER PRESENT STATUS" means a weak score for positive cancer status, but in the case of a second score of 30 for BPH or cancer status, it would also indicate to the practitioner a high probability of having BPH but not cancer. The practitioner will use this additional information, along with other medical information and the patient's medical history, to decide on the next steps in diagnosis.

[0142] Detailed consideration of methods

[0143] Analysis steps / algorithm

[0144] The process of developing an analytical model in accordance with the present invention generally follows the logical path described below, which is illustrated in FIG. 22.

[0145] In step 2200 "Patient Sample Collection", the software collected a large group of samples from patients with known disease free and disease status. Samples were generally not screened for any other unrelated pathological conditions (non-malignant in the case of neoplasms), but were collected in such a way that the sample sets looked statistically like the population.

[0146] In step 2202 "Measuring Biomarker Concentrations", the software measured parameters such as biomarker concentrations using methods and devices known in the art.

[0147] In step 2204 "Calculate Proximity Score for Each Biomarker", the software calculated proximity scoring curves for each biomarker and set zones for each as shown in FIG. 25.

[0148] In step 2206, "Assess Samples by Cancer or No Cancer Status," the software runs a model program to evaluate the samples with the proximity correlation method. The model used compression or renormalization equations unique to each of the 4 zones (see Equation 1 below).

[0149] In step 2208 "Test and Correct Estimation", the software tested the individual samples for topology stability and corrected those that could be calculated correctly with the incongruent algorithm. First, all cancer scores were tested for topology stability in the usual manner, including plus or minus noise at the measured concentration level, calculating biased proximity scores and applying them to a basic spatial proximity model. If these skewed cancer status estimates skewed beyond a predetermined limit, the cancer status estimate calculated by the primary model was rejected. The initial concentration levels of the failed tests were then converted to new proximity measures using a secondary or incongruent model. These new proximity scores for these failed samples were then applied to a spatial proximity correlation analysis model. These new cancer status scores were then tested in a secondary model for stability in the same manner. If these samples passed the stability test, they were recorded as having been analyzed by the incongruent model. If both the primary and secondary models were unstable, the sample was reported as indeterminate.

[0150] Finally, at step 2210, the software outputted the result discussed above in the training sample model, categorizing the status of having a disease or the status of not having a disease.

[0151] Devices and reagents used for breast cancer validation study: test platform description

[0152] OTraces CDx Instrument System

[0153] The test data included for the portion below and for most of the work discussed above were measured on the devices and reagents listed below. Data were processed in the OTraces LEVIS system or, in some cases, calculations were performed using PC software. All computing software has been written and validated by OTraces Inc. One skilled in the art will readily appreciate that other equivalent hardware, devices, and reagents may be used to obtain similar results.

[0154] The CDx Instrument system is based on the MicroLab Starlet system from Hamilton. It is customized with the OTraces immunoassay transfer software for Hamilton's high-speed ELISA robot. Hamilton is a reputable company that sells automated fluid handling systems worldwide, including the MicroLab Starlet system. The device has been customized by Hamilton for OTraces to provide full automation. The OTraces CDx system includes an integrated microplate washer and plate reader. These two additional devices allow the system to perform one complete run of a total of five panel immunoassays in one shift without operator intervention after initial setup. The configured system performed 40 cancer status assessments per day. Improvements include software to analyze one target analyte at a time. This is necessary to be able to re-run a particular test if an error occurs during the full passage of the test.

[0155] BC Sera DX Test Kit

[0156] This test kit includes all reagents and disposable devices for performing 120 test assessments of cancer status, including all buffers, blocking solutions, wash solution, antibodies and calibrators. The improvements needed to fully commercialize this kit include the addition of two controls. These controls provide an independent check that a blinded test item is giving a correct assessment of cancer status. The two controls are designed to produce proximity scores of 50 and 150, respectively. The LIMS quality control program (see below) verified the correctness of these controls, thereby verifying that the individual tests were carried out in the field. The test kits were assembled in a GMP factory and received the CE mark. The microtiter plate was pre-coated at the factory with antibody solutions for capture and protein for blocking.

[0157] Laboratory Information Management System (LIMS)

[0158] All clinical chemistry systems on the market today, for example, Roche and Abbott, include a graphical interface with software sufficient to manage patient data, provide instrument quality control and chemical operations, and facilitate identification of a test sample and its introduction into test system. These menus are integrated into the supplied chemistry system. The OTraces business model is to enable these features on OTraces computer servers located at OTraces facilities in the US and connect the CDx to those servers over the Internet using cloud technology. This provides several significant benefits. 1) The LIMS software includes FDA compliant archiving software so that all test run data from each field deployed CDx system is run on OTraces servers. With feedback from the established base, input from key institutions on patient outcomes allows OTraces to collect FDA-compliant data to help FDA market balance in the US. 2) Preferably, the barcoded reagent packaging allows the instrument and LEVIS to link all QC test results from factory QC tests. This data is available in real time as field tests are performed to further validate the results of field tests. 3) The CDx system only works with validated OTraces reagents and therefore test runs using non-OTraces reagents are not possible. This system is a typical operator user interface with all real-time functions and patient reports are available immediately after the test is completed.

[0159] A step-by-step method for developing a training sample model and calculating a risk score is shown in the flowchart of FIG. 23. This method can be implemented in software in certain embodiments of the present invention. First, a training sample model is built, and its final product provides diagnostic results for unknown patient samples, called blind samples, since the correct diagnosis is not known at the time of analysis of these blind samples. In general, the present invention allows the healthcare professional to determine a risk score, which the healthcare professional then considers that score, along with other patient factors, to make a medical judgment about the presence or absence of a given disease.

[0160] Steps 2302-2318 are a description of the method by which a training sample model is created. In step 2302, the software determined training sample requirements based on diagnostic needs, which are predetermined criteria that can be specified by one of skill in the art. For example, these criteria may be disease status versus disease free status, more specifically, for example, breast cancer, comparing breast cancer positive samples with samples known to be free of breast cancer.

[0161] In step 2304, the program determined the metavariables to be calculated as well as the independent variables (ie, biomarkers) to be measured.

[0162] In step 2306, the software collected training samples according to the parameters set in steps 2302-2304. In step 2308, the software determined the measured explanatory variables and metavariables, as well as the correct disease diagnosis associated with those results, using the appropriate medical equipment for each training sample. In step 2310, the software calculated a two-marker topology for each sample and training set. In step 2312, the software calculated optimal weight or two-marker topology effect corrections for the following: (1) detection limit uncertainties, eg, samples that were determined to be below the classical detection limit; (2) extremal topology instabilities, such as those determined using the methods described in paragraph [0111] and in relation to the topology stability discussed above. At step 2314, the calculations were considered complete and the main training model was fixed for diagnosing a disease (eg, cancer). In step 2316, the software developed a secondary training model using fundamentally incongruent correlation modeling (see, for example, FIG. 10). At step 2318, the calculations were considered complete and fixed, since a secondary training model was created to diagnose the status of the presence of the disease. Thus, a set of training models for diagnosing the disease was created.

[0163] Steps 2320-2338 are a description of how the software of the present invention used the developed training model to diagnose diseases such as cancer. In step 2320, the software measured blind explanatory variables, such as biomarkers, using medical equipment similar to that used to develop the training sample model. In step 2322, the software received or measured and calculated the metavariable data for each blank. In step 2324, the software used this data to calculate an initial disease risk score for the blind sample using the primary training sample model. In step 2326, the software determined the topology stability of the blind sample evaluation from the patient. In step 2328, the software checked whether the evaluation passed the topology stability test. The pass/fail criteria involved determining how large the error is due to instability and, most importantly, whether the score changes from a positive to a negative disease status or vice versa. In step 2330, if the assessment was found to have passed the stability test, a diagnostic report and risk assessment was output and/or published. If the score failed the test, then at step 2332 the software additionally calculated a secondary risk score for the presence of the disease using the incongruent method algorithm (Algorithm II) as described above. At step 2334, the software again checked whether the evaluation passed the topology stability test. In step 2336, if the assessment was found to have passed the stability test, a diagnostic report and risk assessment was output and/or published. If the evaluation still failed the test, at step 2338 the software generated a diagnostic report and output and/or publish the results as indeterminate as to whether a disease is present.

[0164] Proximity scores calculated according to the present invention have several unique properties. In accordance with certain embodiments, mean protein values are included in the log compression as a ratio to the actual measured concentration for a patient of a given age. In essence, this method created a fan of similar equations, each of which is unique, for example, in terms of age in years for a population of patients. Each unknown sample received a unique equation for the age of the sample.

[0165] A ratio could be used that included the age-adjusted mean for disease free status and disease status, and the actual concentration in the patient sample, in the following form.

[0166] EQUATION 1

[0167] Proximity index = (K) * ln ((C _i /C _{(c or h)} )-(C _h /C _c )) ² where:

[0168] K = proportional constant;

[0169] C _i = measured concentration of the patient's actual analyte;

[0170] C( _{C or h} ) = age-adjusted patient analyte concentration of that patient; the value is adjusted depending on whether the patient has the status of the absence of the disease or the presence of the disease;

[0171] C _h = age-adjusted mean analyte concentration of a patient with no disease status; and

[0172] C _c = patient age-adjusted mean analyte concentration of a patient with disease status.

[0173] This Equation 1 is designed to adjust contraction and expansion depending on the zone of the positively regulated group, as shown in FIG. 25. The above formula for the proximity indicator fulfills this requirement; however, many other forms of this equation can be implemented as will be apparent to those skilled in the art. For example, C _i , C _h and C _c can be actual concentrations or concentration distances from the mean, median, or distance from subgroup medians or dynamic range edges, as discussed above. Other variations of this calculation are given below as Equations 2 and 3.

[0174] EQUATION 2

[0175] Proximity index = K * ln (((concentration in unknown sample) / (concentration with average value for cancer status at the age of the patient from whom the unknown sample was taken)) - ((concentration with average value for cancer absence status at age of the patient from whom the unknown sample was taken) / (concentration with the average value for the status of having cancer at the age of the patient from whom the unknown sample was taken))) ² .

[0176] This equation yields negative infinity (the natural logarithm of zero) if the value of the unknown sample is equal to the mean of the cancer-free status with the value of the age of the patients from whom the unknown samples were taken. It is replaced in the actual equation with a given value, such as 2, as shown in FIG. 25. In other words, the software tested values outside the preset range and reset to a value at the limit of the preset range.

[0177] EQUATION 3

[0178] Proximity index = K * ln (((concentration in unknown sample) / (concentration with mean value of cancer-free status at the age of the patient from whom the unknown sample was taken)) - ((concentration with the mean value of cancer-free status at age of the patient from whom the unknown sample was taken) / (concentration with the average value for the cancer-free status at the age of the patient from whom the unknown sample was taken))) ² .

[0179] Equation 3 gives the natural logarithm of zero, i.e. negative infinity if the value of the unknown sample is equal to the mean of the cancer status for the age of the patient from whom the unknown sample was taken. This embodiment of the equation is used if the unknown sample has a value greater than the average concentration between the means of the statuses of the presence or absence of cancer, at the age of the patient from whom the unknown sample was taken (presumed to have cancer). In this situation, the whole equation thus goes to positive infinity if the unknown sample has an average cancer status value for its age value. This infinity is replaced in the actual equation by a given value, eg 18. In the graph of FIG. 25 shows a family of equations that lead to the age range of interest. The equations operate independently on each of the four zones shown in FIG. 25. These areas are: 1) below the population mean with no disease status; (2) above the mean of disease free status and below the derived midpoint between the mean of disease free status and disease present status (transition from disease free/to disease status); (3) between the derivative of the midpoint between the mean disease free status/disease status and the population mean disease status; and (4) above average for disease status. It should be noted that these zones do not indicate that the samples located within the zone have the status of the presence of the disease or the status of the absence of the disease. The true diagnosis of an individual sample can be anything, and its position, if "incorrect", may be due to another condition that affects this biomarker. We call this proteomic noise. The zones only indicate how the individual samples relate to the averages, and are a framework for compression or renormalization, carried out according to Equation 1, for example. It should be noted that each equation represents only one age value, and that the total set is a set of equations, each of which represents a single age value. The general set of equations is designed to set proximity score values to the same predetermined value for all ages when the actual concentration is exactly equal to the mean values. The ages shown are 35, 50, and 65. The full set looks like a fan, with one equation for each patient's age from which the unknown sample was taken.

[0180] Proximity scores (without units and therefore not concentrations or levels), for example, are illustratively calculated as described above and then used on a multivariate spatial proximity correlation plot for analysis. In addition, all graphs are normalized to the general characteristics of the population distribution; mean age of disease-free status and disease status (adjusted for age or not), median, or subgroup dynamic range. These methods can improve predictive power by 5 percentage points or more.

[0181] The above exemplary embodiments of the systems and methods of the present invention can be implemented in software running on both networked and non-networked hardware. An exemplary embodiment of the hardware used to implement the present invention is described in connection with FIG. 26. In an exemplary system 2600, one or more peripheral devices 2610 are connected to one or more computers 2620 via a network 2630. Examples of peripheral devices 2610 include smartphones, smartwatches, tablets, wearable electronic devices, medical devices such as ECG and blood pressure monitors, and any other devices that collect data on biomarkers, which are known in the present field of technology. Network 2630 may be a wide area network, such as the Internet, or a local area network, such as an intranet. Because of the network 2630, the physical location of the peripherals 2610 and computers 2620 does not affect the functionality of the present invention. Both implementations are described herein, and unless otherwise noted, assume that peripherals 2610 and computers 2620 may be in the same or different physical locations. Communication between the hardware components of the system can take place in a variety of known ways, for example, using network connection components such as a modem or an Ethernet adapter. Both 2610 peripherals and 2620 computers will include or be connected to communications equipment. It is assumed that communication occurs through standardized production network protocols such as HTTP.

[0182] Each computer 2620 consists of a central processing unit 2622, a storage medium 2624, a user input device 2626, and a display 2628. Examples of computers that can be used are commercially available personal computers, open source computing devices (e.g., Raspberry Pi), commercially available servers; and commercially available portable devices (eg smartphones, smartwatches, tablets). In accordance with one embodiment, each of the peripherals 2610 and each of the computers 2620 of the system may have software associated with the system installed thereon. In accordance with such an embodiment, the biomarker data may be stored locally on the networked computers 2620, or alternatively, on one or more remote servers 2640 that are accessible to any of the networked computers 2620 via the network 2630. In accordance with alternative embodiments, , the software runs as an application on 2610 peripherals.

[0183] While certain features of the described embodiments have been illustrated, many modifications, substitutions, alterations, and equivalents will now be apparent to those skilled in the art. Therefore, it is to be understood that the present invention is not limited to the specific embodiments or schemes disclosed, but rather is intended to cover any changes, adaptations or modifications that fall within the scope and spirit of the present invention as defined in the appended claims. All sources cited herein, including patents and applications, are expressly incorporated in their entirety.

Claims (31)

1. A computerized method for diagnosing a disease, according to which the computer on which the method is implemented contains one or more processors containing memory functionally associated with said one or more processors, and the method includes the following steps: (a) obtaining a first set of one or more concentration values of a first biomarker from a first patient sample, wherein the first patient sample is diagnosed with a disease-free status, wherein the first biomarker is an active tumor interstitial fluid protein that is selected in serum as a proxy for activity in the tumor microenvironment ; (b) obtaining a second set of one or more first biomarker concentration values from a second patient sample, wherein the second patient sample is diagnosed with disease status, wherein the first biomarker is an active tumor interstitial fluid protein that is selected for serum as a proxy for activity in the tumor microenvironment ; (c) calculating a first set of proximity metrics from the first set of concentration values and a second set of proximity metrics from the second set of concentration values; and (d) calculating a correlation for a first biomarker diagnosed with a disease from the first and second set of concentration values and the first and second set of proximity score values, the correlation being one of simple regression, ROC area maximization, topology stabilization, or spatial proximity analysis; and (e) diagnosing the disease of the patient using the calculated correlation for the first biomarker. 2. The computerized method according to claim 1, wherein steps (a) to (d) are repeated for a maximum of five biomarkers. 3. The computerized method of claim 1, wherein the correlation combines two or more of simple regression, area under the ROC curve maximization, topology stabilization, and spatial proximity analysis. 4. The computerized method of claim 1, wherein the first and second patient samples include at least one of blood samples, urine samples, or tissue samples. 5. The computerized method of claim 1, wherein the disease being diagnosed is one of prostate cancer, breast cancer, lung cancer, or ovarian cancer. 6. The computerized method of claim 5, wherein the disease being diagnosed is the stage of prostate cancer, breast cancer, lung cancer, or ovarian cancer based on the Gleason score. 7. The computerized method of claim 6, wherein the first and second patient samples comprise cancer staging data, and wherein the cancer staging data is categorized into a plurality of binary groups. 8. Computerized method according to claim 7, in which each binary group is evaluated. 9. The computerized method of claim 1, wherein the biomarkers are selected from the functional group of cytokines, and wherein the functions of the cytokines are at least three of pro-inflammatory, anti-inflammatory, antitumor genesis, cell apoptosis, and vascularization. 10. The computerized method of claim 1, wherein the first biomarker is VEGF. 11. Server for diagnosing a disease, containing: a central processing unit and a storage medium containing instructions stored on it, which, when executed by the processor, causes the server to: (a) obtain a first set of one or more concentration values of a first biomarker from a first patient sample, wherein the first patient sample is diagnosed with a disease-free status, wherein the first biomarker is an active tumor interstitial fluid protein that is selected in serum as a proxy for activity in the tumor microenvironment ; (b) obtain a second set of one or more first biomarker concentration values from a second patient sample, wherein the second patient sample is diagnosed with a disease status, wherein the first biomarker is an active tumor interstitial fluid protein that is selected for serum as a proxy for activity in the tumor microenvironment ; (c) calculate a first set of proximity metrics from the first set of concentration values and a second set of proximity metrics from the second set of concentration values; (d) calculate correlations for a first biomarker diagnosed with a disease from the first and second set of concentration values and the first and second set of proximity score values, the correlation being one of simple regression, ROC area maximization, topology stabilization, or spatial proximity analysis; and (e) diagnose the patient's disease using the calculated correlation for the first biomarker. 12. Server according to claim 11, wherein steps (a) to (d) are repeated for a maximum of five biomarkers. 13. The server of claim 11, wherein the correlation combines two or more of simple regression, area under the ROC curve maximization, topology stabilization, and spatial proximity analysis. 14. The server of claim 11, wherein the first and second patient samples include at least one of blood samples, urine samples, or tissue samples. 15. The server of claim 11, wherein the disease being diagnosed is one of prostate cancer, breast cancer, lung cancer, or ovarian cancer. 16. Server according to claim 15, wherein the diagnosed disease is the stage of prostate cancer, breast cancer, lung cancer, or ovarian cancer based on the Gleason score. 17. The server of claim 16, wherein the first and second patient samples contain cancer staging data, and wherein the cancer staging data is categorized into a plurality of binary groups. 18. Server according to claim 17, each binary group being evaluated. 19. Server according to claim 11, wherein the biomarkers are selected from the functional group of cytokines, and wherein the functions of the cytokines are at least three of pro-inflammatory, anti-inflammatory, antitumor genesis, cell apoptosis, and vascularization. 20. Server according to claim 11, wherein the first biomarker is VEGF.

Images