Multi-cancer detection

Abstract

A method of detecting whether a subject has cancer (regardless of the stage or type of cancer) includes subjecting a blood sample from the subject to infrared spectroscopic analysis including a wavelength of 400 to 4000 cm-1 to produce a spectroscopic characterization characteristic of the blood sample wherein the spectroscopic characterization characteristic of the blood sample is compared to a representative characteristic from a previous subject having or not having cancer. The spectral signature of the blood sample is analyzed, where subjects having cancer include subjects having different cancer types and different cancer stages, thereby detecting whether the subject has cancer based on the spectral signature obtained from the subject.

Description

Technical Field

[0001] The present disclosure relates to methods of detecting cancer using spectroscopic techniques. Background Art

[0002] Early detection of cancer is crucial to improving patient outcomes and reducing mortality. In 2020, cancer caused 10 million deaths and is one of the leading causes of death worldwide (1). Early diagnosis can inhibit disease progression before tumors spread to more advanced stages. Therefore, with earlier detection, many patients can be cured with surgical intervention alone, meaning they do not have to endure aggressive systemic treatments such as radiation or chemotherapy—a reported 70% of early-stage tumors (stage I) are treated with surgery, while only about 13% of stage IV cancers undergo surgical resection (2). If the disease cannot be identified quickly and the cancerous tumor is allowed to grow and metastasize to distant secondary sites, the patient's outlook is often bleak, as current treatments are rarely effective for advanced disease (3). Current screening approaches often focus on a single target organ, such as breast and colorectal cancer. This approach is currently not economically feasible for many cancer types, and in fact more than two-thirds of fatal cancers do not have any active screening options, making earlier detection extremely difficult (4).

[0003] There has recently been a surge in research investigating liquid biopsy technologies that have the potential to transform cancer diagnostics (5). Many of these tests are based on genomic approaches that utilize genetic material such as circulating tumor DNA (ctDNA) and / or cell-free DNA (cfDNA). Notably, there are approximately 150,000 scientific papers documenting thousands of biomarkers with clear clinical utility, but only 1% of known markers are routinely used in clinical applications (6). Currently, only a few commercial liquid biopsies are available, which are primarily targeted at detecting a single cancer. For example, the SelectMDx test for prostate cancer is a urine-based test that has been successfully launched in the United States and Europe by MDxHealth (7). The ExoDx test (Exosome Diagnostics), also for prostate cancer, provides patients with an individual risk score that determines whether they should undergo a biopsy (8). The commercial use of these single-cancer liquid biopsies demonstrates the potential for early detection strategies in healthcare systems.

[0004] Promising cancer detection methods using liquid biopsies are based on spectral signatures obtained using attenuated total reflectance infrared (IR) spectroscopy to distinguish cancer from non-cancer in a subject (see WO2017 / 221027; Baker et al.). However, this approach does not address the need in the art to develop methods that can detect general cancer signatures to determine cancer status across a variety of cancers, including the type of cancer in a subject, and, for example, to detect early-stage cancer. Therefore, the development of robust tests that can detect multiple cancer types would be transformative in the field of diagnostics. More than 70% of cancer deaths occur in low- and middle-income countries (9), so there is an urgent need to develop low-cost, highly sensitive multi-cancer tests.

[0005] One of the goals of the present disclosure is to develop an early detection diagnostic system that will alleviate one or more of the above-mentioned disadvantages of existing screening methods. Summary of the invention

[0006] The present disclosure is based in part on research that uses methods based on infrared (IR) spectroscopy to identify proliferative diseases (such as cancer) using liquid blood samples. This method is unique because it does not look for individual biomarkers, but rather probes a wide range of biological features and produces a distinctive signature that represents a significant portion or the entire biochemical profile of the sample. The distinctive spectral features contain molecular information from the tumor as well as host responses (such as immune responses). Infrared spectroscopy methods can employ attenuated total reflectance-Fourier transform infrared (ATR-FTIR) spectroscopy, which is particularly suitable for clinical use because the method is minimally invasive, cost-effective, requires little or no sample preparation, and because the analysis is rapid, reproducible results can be generated in a matter of minutes.

[0007] In a first aspect, there is provided a method of detecting whether a subject has cancer, regardless of the stage or type of cancer, the method comprising:

[0008] Blood samples from subjects were analyzed for -1 Infrared spectroscopy (such as ATR-infrared spectroscopy) of a certain wavelength is used to generate spectral characteristic of a blood sample, wherein the spectral characteristic of the blood sample is analyzed against representative characteristics from previous subjects with or without cancer, wherein the previous subjects with cancer include subjects with different cancer types and different cancer stages, thereby detecting whether the subject has cancer based on the spectral characteristic obtained from the subject.

[0009] Therefore, in this document, the term "subject" refers to an individual on whom testing is performed, and the term "previous subject" refers to an individual from whom a reference spectral signature, such as a database or data set of reference spectral signatures, is obtained and which can be used for subsequent comparison and / or correlation with the spectral signature of a blood sample of the "subject" for testing, as explained in more detail below.

[0010] In this article, the term "blood" refers to whole blood or its fractions. In one embodiment, the sample for infrared analysis from the subject is obtained from whole blood or its fractions (such as serum or plasma). In one embodiment, the sample for infrared analysis from the subject is serum.

[0011] The term "cancer" refers to a physiological condition in which cells exhibit abnormal and unregulated growth. Unless otherwise indicated, "cancer" as used herein generally refers to any type of cancer, regardless of the stage of the cancer. Examples of cancer include, but are not limited to, bile duct cancer, bladder cancer, brain and central nervous system cancer, breast cancer, cervical cancer, colorectal cancer, kidney cancer, gallbladder cancer, leukemia, liver cancer, lung cancer, gastric cancer, Hodgkin and non-Hodgkin lymphoma, melanoma, multiple myeloma, mesothelioma, osteosarcoma, oral cancer (including lip cancer and salivary gland cancer), larynx and oropharynx, nasopharynx, hypopharyngeal cancer (laryngeal andoro-, naso-, hypopharyngeal cancer), ovarian cancer, pancreatic cancer, prostate cancer, sarcoma, thyroid cancer, uterine and vaginal cancer, and testicular and penile cancer. In one embodiment, cancer is selected from the group consisting of brain cancer, breast cancer, colorectal cancer, kidney cancer, lung cancer, ovarian cancer, pancreatic cancer, and prostate cancer.

[0012] As used herein, "cancer stage" refers to the size of the tumor and the indication of cancer spreading from the tissue of cancer origin, which enables clinicians to assess cancer progression. According to the overall stage grouping (Overall Stage Grouping, also referred to as Roman Numeral Staging), most cancers (including tumors) can be divided into four stages. Phase I refers to the period when cancer is very small and confined to the organ where cancer begins. Phase II refers to the period when cancer has grown and is larger than Phase I, but has not yet spread to surrounding tissues. In some types of cancer, Phase II may include the situation that cancer has spread to adjacent lymph nodes. Early stage cancer typically refers to Phase I or Phase II cancer. Phase III refers to the period when cancer is larger and may have spread to surrounding tissues (such as adjacent tissues and / or lymph nodes). Phase IV refers to the period when cancer has spread to other organs or has developed to a considerable volume of organ-limited tumors (such as brain cancer) (also referred to as late or metastatic cancer). In one embodiment, the method described herein is particularly suitable for detecting early stage cancer (typically Phase I-III or Phase I-II). In another embodiment, the methods described herein are particularly suitable for detecting cancer regardless of the stage of the cancer, and / or the methods described herein are particularly suitable for detecting cancer at any stage I-IV.

[0013] As used herein, the term "detection" is used in a broad sense and can be used in terms of facilitating the diagnosis and / or prognosis of a subject. Detecting cancer in a subject can include analyzing against representative features obtained from (the) previous subject. This can involve comparing and / or correlating the spectral features of a blood sample (or its components) with one or more reference spectral features previously obtained by statistical analysis from (the) previous subjects with and without cancer and at different cancer stages. Suitable computational methods or programs can be used for comparison and / or correlation. The method can include using machine learning or deep learning processes for comparison and / or correlation.

[0014] The machine learning or deep learning process may be or may include a neural network. The neural network may be or may include a convolutional neural network or a recurrent neural network. The convolutional neural network may include one or more of the following: at least one one-dimensional convolutional layer, at least one down sampling layer, and / or at least one batch normalization layer. The recurrent neural network may include one or more long short-term memory (LSTM) layers. Other exemplary algorithms that may be used include: (oblique) random forests together with partial least squares (PLS), logistic regression, support vector machines or other machine learning algorithms (at each node in each decision tree); other linear models, such as LDA. The machine learning or deep learning model may be a single model, or an integration of multiple models. The integrated model may be constructed on different samples of the data using one or more bootstrap aggregations ("bagging") or random sampling.

[0015] The data may be, may include, may be derived from, or may represent directly measurable spectral data including or resulting from spectral measurements collected from a blood sample. The data may be or may include measurements collected across a defined wavelength region, or may be or may include measurements obtained from two or more sub-regions, as will be discussed further herein.

[0016] According to an example of the present disclosure, a computer-readable medium carrying a computer program is provided, wherein the program includes computer-readable instructions configured to cause a computer to execute the method described herein.

[0017] According to one example of the present disclosure, a computer device is provided, such as a computer device incorporated in or connected to a spectrometer, which includes or is configured to access: a memory storing processor-readable instructions, and a processor configured to read and execute instructions stored in the memory, wherein the processor-readable instructions include instructions configured to control a computer to perform the method described herein.

[0018] The examples of the subject matter and functional operations described in this specification may be implemented in digital electronic circuits, tangibly embodied computer software or firmware, computer hardware (including the structures disclosed herein and their structural equivalents), or a combination of one or more of them. The examples of the subject matter described in this specification may be implemented in the form of one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-temporary program carrier, for execution by a data processing device or for execution to control the operation of a data processing device. Alternatively or additionally, the program instructions may be encoded on an artificially generated propagation signal (e.g., a machine-generated electrical, optical, or electromagnetic signal), which is generated to encode information for transmission to a suitable receiving device for execution by a data processing device. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access storage device, or a combination of one or more of them.

[0019] The term "data processing apparatus" covers all types of apparatus, devices and machines for processing data, including, for example, a programmable processor, a computer or multiple processors or computers. The apparatus may include special purpose logic circuits, such as an FPGA (field programmable gate array) or an ASIC (application specific integrated circuit). In addition to hardware, the apparatus may also include code that creates an execution environment for the computer program in question, such as code constituting processor firmware, a protocol stack, a database management system, an operating system, or a combination of one or more of them.

[0020] A computer program (which may also be referred to or described as a program, software, software application, module, software module, script, or code) may be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment. A computer program may (but need not) correspond to a file in a file system. A program may be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, subroutines, or portions of code). A computer program may be deployed to execute on one computer or on multiple computers located at one site or distributed across multiple sites and interconnected by a communications network.

[0021] The processes and logic flows described in this specification can be performed by one or more programmable computers that execute one or more computer programs to perform functions by operating input data and generating output. The processes and logic flows can also be performed by special-purpose logic circuits, and the apparatus can also be implemented as special-purpose logic circuits, such as FPGAs (field programmable gate arrays) or ASICs (application-specific integrated circuits).

[0022] Computers suitable for executing computer programs include (e.g., can be based on) general or special microprocessors or both, or any other type of central processing unit. Typically, the central processing unit will receive instructions and data from a read-only memory or a random access memory or both. The essential components of a computer are a central processing unit for performing or executing instructions and one or more storage devices for storing instructions and data. Typically, a computer will also include one or more large-capacity storage devices (e.g., disks, magneto-optical disks, or optical disks) for storing data, or be operably coupled to receive data from the large-capacity storage device or to transmit data to the large-capacity storage device, or both. However, a computer does not need to have such a device. In addition, a computer can be embedded in another device, for example, to name a few, a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device (e.g., a universal serial bus (USB) flash drive).

[0023] Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and storage devices, including, for example, semiconductor memory devices, such as EPROM, EEPROM and flash memory devices; magnetic disks, such as internal hard disks or removable disks; magneto-optical disks; and CD-ROM and DVD-ROM disks. The processor and memory may be supplemented by or incorporated in special purpose logic circuits.

[0024] To provide interaction with a user, examples of the subject matter described in this specification may be implemented on a computer having a display device, such as a CRT (cathode ray tube) or LCD (liquid crystal display) monitor, for displaying information to the user, and a keyboard and pointing device, such as a mouse or trackball, through which the user can provide input to the computer. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback, such as visual feedback, auditory feedback, or tactile feedback; and the input from the user may be received in any form, including sound, voice, or tactile input. In addition, the computer may interact with the user by sending documents to and receiving documents from a device used by the user; for example, in response to a request received from a web browser on a user's client device, sending a web page to the web browser.

[0025] Embodiments of the subject matter described in this specification may be implemented in a computing system that includes a back-end component, such as a data server, or includes a middleware component, such as an application server, or includes a front-end component, such as a client computer with a graphical user interface or a web browser, through which a user can interact with the embodiment of the subject matter described in this specification, or includes any combination of one or more such back-end, middleware, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication, such as a communication network. Examples of communication networks include local area networks (LANs) and wide area networks (WANs), such as the Internet.

[0026] A computing system may include clients and servers. Clients and servers are generally remote from each other and typically interact through a communication network. The relationship of client and server arises from computer programs running on respective computers and having a client-server relationship with each other.

[0027] Reference is made herein to conditional parameter curves, but it should be understood that such reference is to the evolution of decay parameters and may include linear segments and / or arcuate segments, such as segments that may be described by polynomial, exponential, power and other functions, and may vary smoothly or gradually, or may include sharp, angular transitions between regions.

[0028] Although this specification contains many specific implementation details, these details should not be interpreted as limitations on any invention scope or content for protection, but should be interpreted as descriptions of features that may be specific to a specific embodiment of a specific invention. Certain features described in the context of a separate embodiment in this specification may also be implemented in combination in a single embodiment. On the contrary, the various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments. In addition, although features may be described as acting in certain combinations as described above and even initially claimed, in some cases, one or more features from the combination may be deleted from the combination for protection, and the combination for protection may involve a variant of a sub-combination or a sub-combination.

[0029] The methods of the present invention typically use infrared (IR) spectroscopy. Various infrared techniques known in the art can be used in the methods of the present disclosure. One method can utilize Fourier transform infrared (FTIR) spectroscopy. In FTIR, the wavelengths of the wavelengths of light can be measured at 400 to 4000 wavenumbers (cm -1 ) area. Usually, the infrared spectrum can have a 10 cm -1 Less than 5 cm -1 Below or about 4 cm -1 The FTIR spectroscopy may employ at least 10 scans, at least 15 scans, or at least 30 scans. The FTIR spectroscopy may employ up to 100 scans, up to 50 scans, or up to 40 scans. For example, 16 scans may be used. The scans may be co-added. As will be appreciated by the skilled artisan, the number of scans may be selected to optimize data content and data acquisition time.

[0030] Before spectral analysis, a background spectrum can be obtained. Such a background spectrum can provide a correction for the background environment. For example, a background spectrum of air can be obtained to provide an atmospheric correction. Alternatively or additionally, a background spectrum of a solution (e.g., an aqueous solution, such as phosphate buffered saline (PBS)) can be obtained. In one embodiment, a background spectrum is obtained to provide a correction for the background environment. In one embodiment, spectral analysis can also include normalization (e.g., standard normal variables), noise reduction (e.g., based on principal component analysis (PCA)) and derivative (e.g., first order or second order) as pre-processing steps and / or Fourier transform infrared spectroscopy.

[0031] The claimed method utilizes infrared spectroscopy, such as attenuated total reflectance (ATR)-infrared spectroscopy, using a blood sample from a subject. In some embodiments, the spectroscopy used to obtain the spectral signature characteristics of the blood sample can be ATR-FTIR.

[0032] In a typical ATR-IR spectroscopic analysis, a subject's blood sample can be loaded onto an internal reflection element (IRE), and infrared light can pass through the IRE and reflect (e.g., via total internal reflection) at least once on the inner surface of the IRE in contact with the sample. Such reflections can form an "evanescent wave," the extent to which it penetrates the blood sample depends on the wavelength of the light, the angle of incidence, and the refractive index of the IRE, as well as the blood sample itself. By changing the angle of incidence and / or wavelength of the incident light, the penetration depth and the path of the reflected light can be changed. When the light beam leaves the internal reflection element, it can be received by an infrared detector. The IRE is typically an optical material with a higher refractive index than the blood sample to achieve the evanescent wave effect.

[0033] The infrared spectral characteristics of a blood sample (which may be referred to as a signature region or fingerprint region) may typically be part or all of the 400 to 4000 cm -1 Related infrared spectra.

[0034] The spectral feature characteristics of the blood sample are compared with a database of representative features previously obtained from samples from previous cancer subjects or healthy subjects to detect whether the subject has cancer by comparing the respective features. The database includes representative features from previous subjects with different cancer types and different cancer stages. In addition, the spectral features of the blood sample can be compared with a database of representative features obtained from samples from previous subjects with a specific type of cancer to diagnose the type of cancer. Such comparisons can be performed using pattern recognition software and / or machine learning analysis known in the art and / or described herein.

[0035] In one embodiment, the analysis of representative features from previous subjects includes applying a training model to spectral features. The training model may include a trained machine learning model, and optionally a neural network, a support vector machine (SVM) or a random forest (RF) decision tree. The training model may include a classifier or act as a classifier by applying the probability threshold or a certain probability threshold to the probability value output by the training model. The method may additionally include selecting and / or changing the probability threshold, thereby selecting and / or changing the specificity and / or sensitivity of the spectral feature analysis. In addition, the method may involve selecting a probability threshold based on a receiver operating characteristic (ROC) curve for the training model, and / or selecting a training model from a training model set based on multiple ROCs for the training model set, for example, to obtain desired specificity and / or sensitivity.

[0036] As used herein, the term "sensitivity" is understood to be the proportion of diseased subjects correctly identified as "positive" by a test. Thus, sensitivity can be defined as the percentage of true positives expected by a test. A test with 100% sensitivity will correctly detect all patients with a given disease. In other words, high sensitivity means a low incidence of false negatives.

[0037] As used herein, the term "specificity" is understood as the proportion of non-diseased subjects that are correctly identified as "negative" by a test. Thus, specificity can be defined as the percentage of true negatives expected by a test. A test with 100% specificity will correctly detect all patients who do not have a given disease. In other words, high specificity means a low incidence of false positives.

[0038] An advantageous feature of the methods described herein is that cancer can be detected in patients, regardless of the stage of cancer. In one embodiment, the method can detect early cancer, because early medical intervention can improve the possibility of patient survival. In order to enable it to be used as a first-line diagnostic test, the high sensitivity of the test is crucial to ensure that all cancer patients are correctly identified. The sensitivity of existing genomic cancer diagnostic tests (such as ctDNA liquid biopsy tests based on next-generation sequencing (NGS) and methylation-based tests) is not enough to detect early cancer. The spectral liquid biopsy method disclosed herein is different from other tests, because the probability threshold of machine learning analysis can be adjusted according to clinical needs to maximize sensitivity or specificity. In one embodiment, a method for spectral liquid biopsy including the use of machine learning analysis is provided, wherein a blood sample from a patient is used, and the probability threshold of the analysis is adjusted to maximize the sensitivity of cancer detection, regardless of the stage of cancer. In a particular embodiment, the spectral analysis based on liquid biopsy can distinguish patients with early cancer (such as I phase or II phase cancer) from subjects without cancer by maximizing the sensitivity of the analysis.

[0039] In certain applications of diagnostic analysis, such as when subjects with cancer are stratified from subjects without cancer in a triage setting, a classifier with high sensitivity and moderate to low specificity may be preferable to a model with low sensitivity and high specificity. In other diagnostic applications, such as general cancer detection for population-level screening tests, a classifier with high specificity and moderate to low sensitivity may be preferable. The required performance level is typically selected based on a trade-off between the number of false positives and false negatives that must be tolerated by a specific diagnostic application. Such a trade-off typically depends on the medical consequences of an error. In one embodiment, the method described herein includes adjusting a probability threshold of a machine learning algorithm to determine the sensitivity and / or specificity of a spectral analysis.

[0040] High sensitivity or specificity typically refers to a value greater than 70%. Moderate sensitivity or specificity can typically be greater than 50% but less than 70%. In one embodiment, the sensitivity of the analysis is greater than 70%, and the specificity of the analysis is greater than 40%. In another embodiment, the sensitivity of the analysis is greater than 40%, and the specificity of the analysis is greater than 70%. In one embodiment, the sensitivity of the analysis is at least 80%, and the specificity of the analysis is at least 40%. In one embodiment, the sensitivity of the analysis is at least 40%, and the specificity of the analysis is at least 80%. In a specific embodiment, the sensitivity is at least 80%, and the specificity is at least 40%, 45%, 50%, 55%, 60%, 65%, 70% or 75%. In a specific embodiment, the specificity is at least 80%, and the sensitivity is at least 40%, 45%, 50%, 55%, 60%, 65%, 70% or 75%.

[0041] After identifying a patient with cancer, the multi-cancer diagnostic method of the present disclosure can be adjusted to maximize specificity to correctly identify the patient's cancer type. The ability to adjust sensitivity and / or specificity in this way highlights the potential of the method disclosed herein as a cost-effective test to triage subjects in the clinic and enables transformation to adapt to current cancer pathways, as some cancers require high sensitivity while others require high specificity for first-line triage based on next steps.

[0042] In one embodiment, the spectroscopic analysis of the present disclosure includes performing a spectroscopic analysis with high sensitivity to identify patients with cancer, followed by performing a spectroscopic analysis with high specificity to identify patients without cancer.

[0043] In alternative embodiments, the methods described herein can be adjusted to high sensitivity to identify patients with cancer, who can then be directed to alternative tests for further analysis, such as other blood-based tests, NGS-based tests, protein markers, and / or medical imaging (e.g., CT (computed tomography) magnetic resonance imaging (MRI), ultrasound, colonoscopy).

[0044] The present disclosure also provides a method of using spectral features of a blood sample, the method comprising detection of peaks at specific molecular vibrational modes and / or specific wavenumber regions to detect cancer. Detection of peaks at specific molecular vibrational modes and specific wavenumber regions of spectral features can also be used to distinguish cancer types in patients with cancer. Detection of peaks at specific wavenumber regions is associated with molecular vibrational modes, which can be used to identify cancer types in subjects, regardless of the stage of the cancer.

[0045] In one embodiment, the methods described herein can be used to provide an indication of the type of cancer in a subject, such as brain cancer, breast cancer, colorectal cancer, kidney cancer, lung cancer, ovarian cancer, pancreatic cancer, or prostate cancer.

[0046] Analysis against representative signatures from previous subjects may include applying the trained model to the spectral signatures.

[0047] The training model may include a trained machine learning model, optionally a neural network, a support vector machine (SVM), a random forest (RF) decision tree, an ordered random forest (ORF) model. Alternatively or additionally, any other suitable model or model features may be used, such as ORF-PLS, ORF-SVM and / or a bagging model, a shrinkage discriminant analysis or a distance weighted discriminant (DWD) linear model.

[0048] By applying the probability threshold or a certain probability threshold to the probability value output by the training model, the training model may include a classifier or function as a classifier.

[0049] The method may also include selecting and / or changing a probability threshold, thereby selecting and / or changing the specificity and / or sensitivity of the spectral feature analysis.

[0050] The method may also include selecting a probability threshold based on a receiver operating characteristic (ROC) curve for the training model, and / or selecting a training model from the training model set based on multiple ROCs for the training model set, for example, to obtain a desired specificity and / or sensitivity.

[0051] One method of detecting the presence of cancer, and optionally the type of cancer, in a subject using the methods described in the present disclosure is by analyzing the molecular vibrational patterns obtained by analyzing a blood sample from the subject using ATR-IR or similar methods.

[0052] In one embodiment, the method of detecting the presence of cancer in a subject and optionally detecting a specific cancer type comprises analyzing a molecular vibrational mode selected from the group consisting of: NH (in-plane) bending / deformation, CN stretching, CH stretching / deformation, CH2 stretching, CO stretching, CC stretching, C-OH deformation, CH2 wobble, C=O stretching, asymmetric PO Telescopic and / or symmetrical PO Telescopic.

[0053] In one embodiment, the method for detecting whether a subject has cancer is based on analyzing one or more vibrational modes of a spectral feature selected from the group consisting of CO stretching, CC stretching, CH deformation, NH bending, CN stretching, and / or C=O stretching. In another embodiment, the method for detecting whether a subject has cancer is based on analyzing vibrational modes of a spectral feature including CO stretching, CC stretching, CH deformation, NH bending, CN stretching, and C=O stretching.

[0054] In one embodiment, the method of detecting brain cancer in a subject is based on analyzing one or more vibrational modes of a spectral feature selected from: NH bending, CN stretching, CH stretching and / or CH2 stretching. In another embodiment, the method of detecting brain cancer in a subject is based on analyzing vibrational modes of a spectral feature including NH bending, CN stretching, CH stretching and CH2 stretching.

[0055] In one embodiment, the method for detecting breast cancer in a subject is based on analyzing one or more vibrational modes of a spectral feature selected from the group consisting of: CO stretching, CC stretching, C-OH deformation, NH in-plane bending, CN stretching, asymmetric PO In another embodiment, the method of detecting breast cancer in a subject is based on analyzing the vibrational modes of the spectral features, which include CO stretching, CC stretching, C-OH deformation, NH in-plane bending, CN stretching, asymmetric PO Telescopic and swinging CH2.

[0056] In one embodiment, the method for detecting colorectal cancer and / or renal cancer in a subject is based on analyzing one or more vibrational modes of a spectral feature selected from the group consisting of: C=O stretching, CN stretching, NH bending, CO stretching, CC stretching, and / or CH deformation. In another embodiment, the method for detecting colorectal cancer and / or renal cancer in a subject is based on analyzing vibrational modes of a spectral feature including C=O stretching, CN stretching, NH bending, CO stretching, CC stretching, and CH deformation.

[0057] In one embodiment, the method of detecting lung cancer in a subject is based on analyzing one or more vibrational modes of a spectral signature selected from: symmetric / asymmetric PO In another embodiment, the method of detecting lung cancer in a subject is based on analyzing the vibrational modes of the spectral features, which include symmetric / asymmetric PO stretching, CO stretching, NH in-plane bending and CN stretching.

[0058] In one embodiment, the method of detecting ovarian cancer in a subject is based on analyzing one or more vibrational modes of a spectral signature selected from the group consisting of: NH bending / deformation, CN stretching, CO stretching, CH deformation, and / or CC stretching. In another embodiment, the method of detecting ovarian cancer in a subject is based on analyzing vibrational modes of a spectral signature comprising NH bending / deformation, CN stretching, CO stretching, CH deformation, and CC stretching.

[0059] In one embodiment, the method of detecting pancreatic cancer in a subject is based on analyzing one or more vibrational modes of a spectral signature selected from the group consisting of: NH bending / in-plane bending, CN stretching, symmetric PO In another embodiment, the method of detecting pancreatic cancer in a subject is based on analyzing the vibrational modes of the spectral features, which include NH bending / in-plane bending, CN stretching, symmetric PO Telescopic and CO telescopic.

[0060] In one embodiment, the method of detecting prostate cancer in a subject is based on analyzing one or more vibrational modes of a spectral feature selected from the group consisting of: CO stretching, CH deformation, NH bending / deformation, CN stretching, and / or C=O stretching. In another embodiment, the method of detecting prostate cancer in a subject is based on analyzing vibrational modes of a spectral feature including CO stretching, CH deformation, NH bending / deformation, CN stretching, and C=O stretching.

[0061] An alternative spectral signature that can be used to detect the presence of cancer in a subject, and optionally the type of cancer in a subject having cancer, is by identifying peaks at different wavenumber regions in the spectral signature of a sample from the subject.

[0062] In one embodiment, the method for detecting whether a subject has cancer and detecting the type of cancer if the subject has cancer is based on the presence of a spectral distribution of the ... -1 In an alternative embodiment, the method for detecting whether a subject has cancer and the type of cancer if the subject has cancer is based on the identification of peaks in the sample spectral characteristics at one or more wavenumber regions within one or more portions thereof. -1 Or identification of peaks in a sample's spectral signature at one or more wavenumber regions within one or more portions thereof.

[0063] In one embodiment, the method of detecting whether a subject has cancer is based on the identification of peaks at one or more wavenumbers: 1163 cm -1 、1578 cm -1 and / or 1682 cm -1In another embodiment, the method for detecting whether a subject has cancer is based on 1163 cm -1 、1578 cm -1 and 1682 cm -1 Identification of the peak at .

[0064] In one embodiment, the method of detecting whether a subject has brain cancer is based on the identification of peaks at one or more wavenumbers: 1595 cm -1 、2930 cm -1 and / or 1525 cm -1 In another embodiment, the method for detecting whether a subject has brain cancer is based on 1595 cm -1 、2930 cm -1 and 1525 cm -1 Identification of the peak at .

[0065] In one embodiment, the method of detecting whether a subject has breast cancer is based on the identification of peaks at one or more wavenumbers: 1025 cm -1 、1260 cm -1 and / or 1337 cm -1 In another embodiment, the method for detecting whether a subject has breast cancer is based on 1025 cm -1 、1260 cm -1 and 1337 cm -1 Identification of the peak at .

[0066] In one embodiment, the method of detecting whether a subject has colorectal cancer is based on the identification of peaks at one or more wavenumbers: 1682 cm -1 、1575 cm -1 and / or 1165 cm -1 In another embodiment, the method for detecting whether a subject has colorectal cancer is based on 1682 cm -1 、1575 cm -1 and 1165 cm -1 Identification of the peak at .

[0067] In one embodiment, the method of detecting whether a subject has renal cancer is based on the identification of peaks at one or more wavenumbers: 1682 cm -1 、1590 cm -1 and / or 1164 cm -1 In another embodiment, the method for detecting whether a subject has renal cancer is based on 1682 cm -1 、1590 cm -1 and 1164 cm -1Identification of the peak at .

[0068] In one embodiment, the method of detecting whether a subject has lung cancer is based on the identification of peaks at one or more wavenumbers: 1074 cm -1 、1190 cm -1 and / or 1288 cm -1 In another embodiment, the method for detecting whether a subject has lung cancer is based on 1074 cm -1 、1190 cm -1 and 1288 cm -1 Identification of the peak at .

[0069] In one embodiment, the method of detecting whether a subject has ovarian cancer is based on the identification of peaks at one or more wavenumbers: 1569 cm -1 , 1412 cm -1 and / or 1135 cm -1 In another embodiment, the method for detecting whether a subject has ovarian cancer is based on 1569 cm -1 , 1412 cm -1 and 1135 cm -1 Identification of the peak at .

[0070] In one embodiment, the method of detecting whether a subject has pancreatic cancer is based on the identification of peaks at one or more wavenumbers: 1577 cm -1 , 1074 cm -1 and / or 1289 cm -1 In another embodiment, the method for detecting whether a subject has pancreatic cancer is based on 1577 cm -1 , 1074 cm -1 and 1289 cm -1 Identification of the peak at .

[0071] In one embodiment, the method of detecting whether a subject has prostate cancer is based on the identification of peaks at one or more wavenumbers: 1356 cm -1 、1499 cm -1 and / or 1642 cm -1 In another embodiment, the method for detecting whether a subject has prostate cancer is based on 1356 cm -1 、1499 cm -1 and 1642 cm -1 Identification of the peak at .

[0072] In one embodiment, any of the above wavenumber regions may vary by up to ±100 cm -1 、±50 cm -1、±40cm -1 、±30 cm -1 、±20 cm -1 and / or ±10 cm -1 .

[0073] Based on a predictive model developed by "training" (e.g., via pattern recognition and / or machine learning algorithms) a database of pre-correlated analyses, spectral features can be associated with a favorable or unfavorable diagnosis and / or prognosis. To train the model, we provide known cancer and known non-cancer samples from an appropriate patient population to determine features that can be distinguished. Correlation analysis results with favorable or unfavorable diagnosis and / or prognosis can be performed manually (e.g., by a clinician or other suitable analyst) or automatically (e.g., by computational means). The association can be established qualitatively (e.g., via comparison of graphical trajectories or features) or quantitatively (e.g., by reference to a predetermined threshold or statistical limit). The association analysis results can be performed using a predictive model, optionally as defined herein, which may have been developed by "training" a database of pre-correlated assays and / or analyses.

[0074] In one embodiment, a computer program product may be provided, the computer program product comprising computer readable instructions executable to perform the methods described herein. The trained machine learning model may be configured to receive a blood sample from a subject comprising 400 to 4000 cm -1 In one embodiment, the method for training a machine learning model may include receiving a plurality of data sets representing spectral features of a plurality of subjects, the spectral features being obtained from analyzing a blood sample from a subject in the wavelength range of 400 to 4000 cm -1 The invention relates to a method for determining the probability of a patient having cancer based on the spectral characteristics of the patient, wherein some of the subjects have cancer, and the subjects having cancer include subjects with different cancer types and different cancer stages. The training of the model may also include adjusting at least one parameter of the model to optimize the area under the ROC curve and / or provide the desired sensitivity and / or specificity for determining whether the patient has cancer.

[0075] The multi-cancer detection method of the present disclosure provides a method for analyzing macromolecules in trace patient serum by infrared spectroscopy and machine learning algorithms. Since only a small amount of blood is needed for analysis, it is feasible to integrate this rapid liquid biopsy spectrum test with other existing blood-based tests without interfering with clinical practice, such as extracting a small aliquot of blood from a conventional blood test. In one embodiment, a small aliquot extracted from a blood sample for infrared analysis can typically be less than 1mL. In an alternative embodiment, a small aliquot extracted from a blood sample for infrared analysis can be less than 100 μL, 80 μL, 60 μL, 40 μL or 10 μL. In this way, most of the original sample (typically more than 80%, 90%, 95% or 99%) can still be used for subsequent analysis.

[0076] Therefore, the method of the present disclosure can allow for effective triage of patients by speeding up further evaluation of patients at higher risk while excluding cancer diagnosis in other patients. In addition, if provided in combination with other blood-based diagnostic assays or conventional blood tests, it can identify early cancers in asymptomatic patients or patients with non-specific symptoms. In addition, the method of the present disclosure effectively stratifies subjects with cancer from subjects without cancer, regardless of the stage of the cancer or the type of cancer. Optionally, the method can be further effective regardless of the age and / or gender of the subject. In one embodiment, spectral analysis based on liquid biopsy can distinguish subjects with cancer from subjects without cancer independently of the age of the subject. In another embodiment, spectral analysis based on liquid biopsy can distinguish subjects with cancer from subjects without cancer independently of the gender of the subject.

[0077] In one embodiment, the methods of the present disclosure may be performed as a stand-alone test or, optionally, as an additional test in combination with other blood-based tests.

[0078] In one embodiment, the method disclosed herein can be carried out in combination with one or more other blood-based tests using a single blood draw, wherein a first aliquot is used to perform infrared spectroscopy as described herein, and a second aliquot comprising the remainder of a blood sample or a portion thereof is used to perform one or more other blood-based tests. In one embodiment, the remainder of the blood draw can be used to perform standard conventional blood tests or other diagnostic tests based on liquid biopsy. In an alternative embodiment, further analysis may include performing subsequent tests, such as liquid biopsy sequencing assays (e.g., ctDNA tests based on NGS, methylation tests, biomarker tests). A single blood draw may be aliquoted into different containers or vessels, or a single blood draw may be provided in a single container or vessel.

[0079] In another aspect, a computer program product is provided, the computer program product comprising computer readable instructions executable to perform the method claimed or described herein.

[0080] In another aspect (which may be provided independently), a training model is provided, the training model being configured to receive a blood sample from a subject comprising a range of 400 to 4000 cm -1 ATR-infrared spectroscopy takes as input the spectral signature of a given wavelength and provides as output whether the subject has cancer, regardless of the stage or type of cancer.

[0081] In another aspect (which may be provided independently), a method for training a model is provided, the method comprising receiving a plurality of data sets representing spectral features of a plurality of subjects, the spectral features being obtained from a blood sample from the subject and comprising spectral characteristics of 400 to 4000 cm -1 The invention relates to a method for determining a probability of whether a patient has cancer based on the spectral features of the patient, wherein some of the subjects have cancer, and the subjects having cancer include subjects with different cancer types and different cancer stages.

[0082] Training of the model may include adjusting at least one parameter of the model to optimize the area under the ROC curve and / or provide a desired sensitivity and / or specificity for determining whether a patient has cancer.

[0083] Features of any aspect may be provided as features of any one or more other aspects. For example, any method, computer program product, model or device feature may be provided as any one or more other method, computer program product, model or device feature. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The present disclosure will now be further described by way of example and with reference to tables and accompanying drawings which show:

[0085] Table 1. Demographic characteristics of patients according to cancer (C) and non-cancer (NC) categories.

[0086] Table 2. Detection rates of sensitivity-adjusted models for each cancer category, by stage, based on a model with a cross-validated lower limit of 45% specificity.

[0087] Table 3. Detection rates in sensitivity- and specificity-adjusted models, by age and sex.

[0088] Table 4. Patient demographic characteristics by organ-specific cancer (C) and non-cancer symptom (NCS) categories.

[0089] Table 5. Patient demographic characteristics of ovarian and prostate cancers classified by organ specificity for sex-specific non-cancer symptom (NCS) groups.

[0090] Table 6. Sensitivity and specificity values ​​for the resampled test set for each organ-specific cancer classification. The results here are based on thresholds selected at a cross-validated sensitivity or specificity of 90%. Non-cancer symptoms (NCS), NCS for women only (NCS-F), and NCS for men only (NCS-M).

[0091] Table 7. Detection rates of organ-specific cancer classification (specificity-adjusted model by stage). The results here are based on a lower limit of 45% for the cross-validated sensitivity.

[0092] Table 8. The top 3 wavenumber regions found to be the most discriminative in each binary classification, along with their corresponding biological assignment and vibrational mode.

[0093] Table 9. The top 5 wavenumber regions found to be the most discriminative in each binary classification, along with their corresponding tentative biological attribution and vibrational pattern.

[0094] Table 10. Overview of brain cancer types by tumor grade.

[0095] Table 11. 95% confidence intervals (CI) of the one-sample Student's t test performed for each selected threshold on the receiver operating characteristic curve presented for each category.

[0096] Figure 1 Results for cancer (C) and non-cancer (NC) classification are shown: (a) average receiver operating characteristic curves showing the trade-off between sensitivity (Sens) and specificity (Spec), where the markers represent the sensitivity-adjusted model (black circles) and the specificity-adjusted model (black diamonds), and AUC represents the area under the curve; (b) detection rate for each cancer stage; and (c) detection rate for each cancer type. Both (b) and (c) are shown for the sensitivity-adjusted model.

[0097] Figure 2Mean receiver operating characteristic curves for organ-specific cancer classifications: (a) brain, breast, colorectal, renal, lung, and pancreatic cancers relative to noncancer symptoms (NCS), and (b) ovarian cancer relative to female-only NCS (NCS-F) and prostate cancer relative to male-only NCS (NCS-M). In all ROC graphs, the markers represent the sensitivity-adjusted model (black circles) and specificity-adjusted model (black diamonds), and AUC represents the area under the curve.

[0098] Figure 3 . The feature importance chart highlights the wavenumber regions found to be the most discriminative for each binary classification.

[0099] Figure 4 Results for cancer (C) and non-cancer (NC) classification are shown: (a) average receiver operating characteristic curves showing the trade-off between sensitivity (Sens) and specificity (Spec), where the markers represent the sensitivity-adjusted model (black circles) and the specificity-adjusted model (black diamonds), and AUC represents the area under the curve; (b) detection rate for each cancer stage; and (c) detection rate for each cancer type. Both (b) and (c) are shown for the specificity-adjusted model.

[0100] Figure 5 Results for the cancer (C) versus asymptomatic non-cancer (NCA) classification (a, c, e) and C versus all non-cancer (NC) classification (b, d, f). Average receiver operating characteristic curves showing the trade-off between sensitivity (Sens) and specificity (Spec) for a) C versus NCA and b) C versus NC, where the markers represent the sensitivity-adjusted model (●) and the specificity-adjusted model (■), and AUC represents the area under the curve. Detection rates for the sensitivity-adjusted model (c, d) and the specificity-adjusted model (e, f) are shown for the respective classifications (differentiated by cancer stage). DETAILED DESCRIPTION OF THE INVENTION

[0102] method

[0103] Patient sample cohort selection

[0104] All samples from patients eligible for inclusion in the study were purchased and / or collected from biobanks, commercial sources, and ongoing clinical studies (Lothian REC 15 / ES / 0094; IRAS 238735). All patients gave written consent for inclusion in the study. Cancer samples were collected from patients with confirmed cancer prior to surgical resection and advanced treatment. The non-cancer group included healthy volunteers and patients with suspicious symptoms, including patients with benign non-cancerous conditions and no evidence of neoplasia - non-cancer patient sera were collected during clinical evaluation.

[0105] Blood samples were obtained by venipuncture using serum collection tubes and were made anonymous. Serum was extracted via centrifugation and stored in a -80°C freezer. Clinical and demographic data were obtained with respect for the confidentiality of the participants.

[0106] Patient sample analysis

[0107] Patient serum samples were analyzed using the Dxcover® Cancer Liquid Biopsy (Dxcover® Ltd., Glasgow, UK) spectroscopic test. This test has been used in a previously conducted clinical study, to which we direct readers for more information (10). In this study, serum samples from the described sources were stored at -80°C until the day of analysis; samples were thawed at room temperature (18-25°C) for up to 30 minutes and inverted three times to ensure adequate mixing and thawing before use. Each patient sample was prepared for analysis by pipetting 3 μL of serum into each of the three sample wells of a Dxcover® specimen slide (Dxcover® Ltd., Glasgow, UK). The prepared slides were placed in a drying unit incubator (ThermoScientific TM Heratherm TM , Waltham, Massachusetts, US) at 35°C for 1 hour to control the dehydration process of the serum droplets (11). Each dried sample slide was then inserted into a Dxcover® autosampler (Dxcover® Ltd., Glasgow, UK) in preparation for spectral acquisition. In this study, the PerkinElmer® Spectrum Two TM FTIR spectrometer (PerkinElmer® Inc., Waltham, Massachusetts, USA) was used to generate spectral data (16 co-scans, resolution 4 cm -1 , data spacing 1 cm -1 ). A total of three spectra are collected per sample well, resulting in nine replicates / patient, which are then submitted to the diagnostic algorithm to generate a disease prediction. Based on the diagnostic algorithm results, the patient sample is reported as positive or negative for cancer.

[0108] Algorithm training

[0109] Machine learning models were developed to identify cancerous features of known patient cohorts, and then predict the presence of cancer in unknown populations. A nested cross-validation strategy was used to develop the model, in which internal cross-validation was used to adjust model hyperparameters, and external cross-validation provided a robust test of model performance. In this method, patients were randomly divided into a training set and a test set (divided by 70:30). The model hyperparameters were adjusted to optimize the area under the receiver operating characteristic curve and provide the required sensitivity or specificity, as estimated by 5 times cross-validation of the training set (70%). The training model was used to predict the spectrum of the test set (30%). Since each patient sample was analyzed nine times, the final diagnosis was made as a consensus prediction (maximum number of votes) of all nine spectra. In order to obtain a robust estimate of classifier performance, the model building process was repeated 51 times using different training sets and test sets, and the mean and standard deviation of the resulting classification indicators were recorded.

[0110] Any suitable machine learning model (e.g., any suitable deep learning model) can be used according to the embodiment, and can be trained according to any suitable training technique and using any suitable training data set. The model can include a neural network, such as a convolutional neural network or a recurrent neural network. In some embodiments, the convolutional neural network can include one or more of the following: at least one one-dimensional convolution layer, at least one downsampling layer and / or at least one batch normalization layer. The recurrent neural network can include one or more long short-term memory (LSTM) layers. Other exemplary models that can be used include: partial least squares (PLS) model, logistic regression model, state vector machine (SVM) or LDA model, random forest (RF) decision tree, ordered random forest (ORF) model, ORF-PLS, ORF-SVM, bagging model, shrinkage discriminant analysis and / or distance weighted discriminant (DWD) linear model and other algorithms. The machine learning model (e.g., deep learning model) can be a single model, or an integration of multiple models. For example, the integrated model can be constructed on different samples of data using one or more bootstrap aggregation ("bagging") or random sampling. In some embodiments, two or more machine learning algorithms are applied sequentially, where, for example, a cancer prediction from a first algorithm is reclassified by a second algorithm. In this case, any suitable machine learning algorithm can be used for the first or second algorithm.

[0111] By applying the probability threshold or a certain probability threshold to the probability value output by the training model, the training model may include a classifier or function as a classifier. In some embodiments, the probability threshold is selected and / or changed, thereby selecting and / or changing the specificity and / or sensitivity of the spectral feature analysis.

[0112] The method may also include selecting a probability threshold based on a receiver operating characteristic (ROC) curve for the training model, and / or selecting a training model from the training model set based on multiple ROCs for the training model set, for example, to obtain a desired specificity and / or sensitivity.

[0113] Metadata analysis

[0114] Patient metadata such as Figure 1 Those shown) were performed as a post hoc analysis using test set predictions from all 51 training-test splits. The test sets were randomly sampled from the full dataset, so a given patient was expected to appear in approximately 15 of the 51 total test sets.

[0115] For each patient, the predictions for all test sets that the patient was in were collected and the detection rate was calculated as the ratio of correct predictions to the total number of predictions. Then, for each metadata category (e.g., disease stage or actual disease, e.g., Figure 1 The detection rates for all patients in the test set (shown) were averaged. It is important to note that the resulting values ​​are not directly comparable to the sensitivity values ​​that appear on the ROC curve, which are calculated as the average of all test sets (not subsets). Example

[0116] Overall cancer classification

[0117] The complete cohort consisted of 2092 patients, of which 1542 were diagnosed with cancer (Table 1). The cancer set included patients with brain cancer, breast cancer, colorectal cancer, kidney cancer, lung cancer, ovarian cancer, pancreatic cancer, or prostate cancer. 91 of the non-cancer group were healthy volunteers, and the remaining 459 were patients with general presenting symptoms (such as headache and stroke) and other benign conditions (e.g., non-malignant cysts and polyps). A fairly balanced male and female participant with a wide age range has been included.

[0118] Initially, the algorithm was tuned to a threshold selected to give 90% sensitivity or specificity on the cross validation (CV) set. The mean receiver operating characteristic (ROC) curve for patient prediction calculated on all resampled test sets reported an area under the curve (AUC) value of 0.86, indicating excellent detection ability between cancer and non-cancer (NC) (12). Figure 1 The two points marked on the ROC curve in represent the sensitivity-adjusted model that achieves 90% sensitivity and 61% specificity (black circle) and the specificity-adjusted model that achieves 56% sensitivity and 91% specificity (black diamond) (for the resampled test set). We can stratify the results from this classifier to visualize the number of correct predictions obtained. Figure 1The bars in the figure represent the detection rates of the sensitivity-adjusted models by (b) stage and (c) cancer type. Figure 4 Equivalent diagrams for the specificity-adjusted model are shown.

[0119] The most important feature of a multi-cancer test is that it must be able to detect early-stage cancers. Figure 1 It is also shown that when adjusted to 90% CV sensitivity, Dxcover® Cancer Liquid Biopsy detected 93% (214 / 231) of stage I cancers and 84% (431 / 516) of stage II cancers. Similarly, the detection rate of cancers at later stages was also very high, with 92% (379 / 410) and 95% (359 / 377) for stage III and IV, respectively. For all cancers combined, 90.1% (1391 / 1543) were correctly predicted. Regarding cancer type, Figure 1 The chart in the figure demonstrates the ability of the Dxcover® Cancer Liquid Biopsy Test to detect a wide variety of cancers—brain cancer (89%), breast cancer (76%), colorectal cancer (96%), kidney cancer (99%), lung cancer (99%), ovarian cancer (88%), pancreatic cancer (84%), and prostate cancer (87%).

[0120] Next, we adjusted for maximum sensitivity or specificity, while the other statistic was fixed at a minimum of 45% above the CV result. This produced 94% sensitivity and 47% specificity, and 94% specificity and 48% sensitivity. By increasing the sensitivity to 94%, the test will detect more cancers, but this will increase the number of false positives, with specificity dropping from 61% to 47%. Additionally, patient metadata factors were explored to assess any impact on multi-cancer liquid biopsy predictions. Patient age had no significant effect on the sensitivity-adjusted model or the specificity-adjusted model (Table 3). Similarly, the detection rates of the two models when divided by sex did not indicate any problems with potential confounding factors.

[0121] Figure 5 Results are shown for the alternative cancer (C) versus asymptomatic non-cancer (NCA) classification (a, c, e) and C versus all non-cancer (NC) classifications (b, d, f).

[0122] This C vs NCA (C v NCA) algorithm was tuned to the selected threshold to obtain 98% sensitivity or specificity for the CV set. ROC analysis of C vs NCA reported an AUC value of 0.94, indicating excellent detection ability ( Figure 5a). This yields 98% sensitivity (59% specificity) or 99% specificity (57% sensitivity). Liquid biopsies targeting asymptomatic screening populations are currently designed to be highly specific to minimize false positives. However, in symptomatic populations, triage testing may be more appropriate, so the selected probability thresholds for the overall C vs. NCA classifier (which combines both asymptomatic and symptomatic non-cancer patients) were chosen to be 90% sensitivity and 95% specificity for the CV set. For the C vs. NC dataset ( Figure 5 b), the sensitivity-adjusted model achieved 90% sensitivity and 60% specificity, and when customized for a higher specificity (95%), the sensitivity was 40%. The ROC curve generated an AUC of 0.85.

[0123] Figure 5 The bars in the graph represent the detection rates by stage: (c), (d) sensitivity-adjusted results and (e), (f) specificity-adjusted results for the C vs. NCA and C vs. NC classifiers, respectively. When exploring C vs. NCA, the sensitivity-adjusted model correctly predicted 98% of all cancers. The detection rates were consistent across all stages: stage I, 99%; stage II, 96%; stage III, 99%; stage IV, 99%. On the other hand, the high specificity (99%) model was still able to detect 64% of stage I cancers and identify 51% of stage II cancers. Therefore, 55% of stage I-II cancers were correctly predicted, highlighting the great potential of Dxcover® Cancer Liquid Biopsy in early cancer detection. For the overall C vs. NC classification, Figure 5 d shows that when adjusted for higher sensitivity, 92% (213 / 231) of stage I cancers and 85% (438 / 516) of stage II cancers were detected. Similarly, detection rates for advanced cancers were also very high—91% (375 / 410) and 95% (359 / 377) for stage III and IV, respectively. For a model with 95% specificity, detection rates were fairly consistent across stages: stage I, 39%; stage II, 32%; stage III, 42%; stage IV, 49% ( Figure 5 f).

[0124] Organ-specific cancers

[0125] Next, the ability of liquid biopsy to distinguish organ-specific cancer from non-cancer patients (under binary classification) with performance symptoms (NCS) is examined. Symptomatic cohorts are more suitable for real-world triage tests and are therefore more relevant comparators than healthy participants. The NCS set covers several conditions, such as stroke, inflammation and epilepsy as well as benign polyps and cysts. Brain cancer, breast cancer, colorectal cancer, kidney cancer, lung cancer and pancreatic cancer groups were tested for the complete NCS data set, which includes balanced male and female patients (Table 4). The ovarian cancer set is compared for female participants (NCS-F) in the NCS group, and the prostate cancer group is examined for male NCS patients (NCS-M) only, as shown in Table 5.

[0126] Not all cancers require the same approach. Optimal test sensitivity and specificity for individual cancers may be influenced by the availability, cost, and risk of follow-up diagnostic investigations or by whether the test would impact current screening programs (13). Similar to overall cancer classification, probability thresholds were selected that yielded 90% sensitivity or specificity for the CV of each organ-specific classifier, and sensitivity and specificity values ​​for the resampled test sets are reported (Table 6). Notably, classifiers for colorectal cancer (91% sensitivity / 72% specificity), lung cancer (91% sensitivity / 77% specificity), and renal cancer (92% sensitivity / 75% specificity) showed real promise for organ-specific applications with well-balanced sensitivity and specificity. We then adjusted for higher sensitivity or specificity by taking a 45% lower limit on the CV results, which was selected based on currently commercially available triage-based liquid biopsy tests. Figure 2 The average ROC curve of the test set of each organ-specific classifier is shown, where the two points on the curve correspond to the sensitivity adjustment model (black circle) and the specificity adjustment model (black diamond), as well as their corresponding AUC values. Brain cancer, colorectal cancer, kidney cancer, and lung cancer reported very promising results relative to the classification of NCS, with AUCs of more than 0.90. Pancreatic cancer reached an AUC of 0.84 relative to the NCS model. Breast cancer reached an AUC of 0.76 relative to the NCS model; however, this still produced a sensitivity of 88% when the specificity was 43%, and a specificity of 87% when the sensitivity was 47%. Both ovarian cancer and prostate cancer models performed extremely well, and both reported an AUC of 0.86. For each organ-specific classification, predictions were checked by cancer stage. The detection rates of both the sensitivity adjustment model (Table 2) and the specificity adjustment model (Table 7) were calculated. For each classification in this study, the 95% confidence interval (CI) for each selected threshold on each ROC curve was calculated, as shown in Table 11.

[0127] The results of sensitivity adjustment are shown in Table 2. The detection rate of brain cancer is as follows: I level is 100% (8 / 8), II level is 85% (23 / 27), III level is 86% (12 / 14), and IV level is 99% (191 / 192) (it is the overwhelming majority that brain cancer concentrates on). The breast cancer group has more early stage samples, and the detection rate is as follows: I phase is 96% (24 / 25), II phase is 87% (79 / 91), III phase is 89% (67 / 75), and IV phase is 100% (9 / 9). The classifiers of colorectal cancer, renal cancer and lung cancer have all reported extremely high detection rates, wherein I phase and II phase are 98-100%. Despite being the smallest subset in this study, ovarian cancer predictions were still very promising: 97% (30 / 31) for stage I; 86% (12 / 14) for stage II; 92% (47 / 51) for stage III; and 100% (29 / 29) for stage IV. Similarly, due to the difficulty in obtaining stage I pancreatic cancer samples, only 8 samples were included in the data set, but the sensitivity-adjusted model was still able to successfully predict 7 (88%) of the stage I tumors. In addition, the detection rates for other pancreatic cancer stages were 94% (II, 61 / 65), 99% (III, 71 / 72), and 95% (IV, 19 / 20). Finally, the prostate cancer results further highlighted the potential for earlier detection: 100% (4 / 4) for stage I; 93% (149 / 160) for stage II; 97% (30 / 31) for stage III; and 75% (3 / 4) for stage IV.

[0128] For the model customized for higher specificity (table 7), more cancer samples are not detected. The brain cancer detection rate is lower for I-III level, but 52% (100 / 192) (under 99% specificity) is reported for IV level. The detection rate of breast cancer group is as follows: I phase is 32% (8 / 25), II phase is 47% (43 / 91), III phase is 59% (44 / 75), and IV phase is 78% (7 / 9). The colorectal cancer sorter reports the highest I phase detection rate of the specific adjustment model, and 18 of 36 are predicted correctly (50%). In addition, colorectal II phase, III phase and IV phase detection rates are respectively 36% (25 / 70), 51% (34 / 67) and 44% (12 / 27). Renal cancer reported the highest detection rate for stage II (66%, 19 / 29), and the remaining stages were: stage I 46% (40 / 87); stage III 38% (13 / 34); stage IV 29% (15 / 51). The detection rates for lung cancer were as follows: stage I 35% (11 / 31), stage II 40% (24 / 60), stage III 60% (39 / 65), and stage IV 44% (20 / 45). Detection rates for ovarian and pancreatic cancers were similar: ovarian cancer—stage I 39% (12 / 31), stage II 36% (5 / 14), stage III 53% (27 / 51), stage IV 45% (13 / 29); pancreatic cancer—stage I 38% (3 / 8), stage II 37% (24 / 65), stage III 44% (32 / 72), stage IV 55% (11 / 20). Finally, prostate cancer performance was quite good for stage II (43%, 68 / 160) and stage III (58%, 18 / 31), but lower for stage I and stage IV, which may be due to the very small number of samples in those groups (n=4).

[0129] Advantageously, it was found that the detection rate was relatively stable in all four stages of the disease. Other detection techniques involving liquid biopsy often show an increase in sensitivity from phase I to phase IV. This is based on the recognized concept in the art that as cancer increases, the concentration of diagnostic information (such as PSA or any genetic marker) will increase, and therefore other liquid biopsies should be observed and improve the detection accuracy in each stage. On the contrary, the data set of the representative features of the present invention includes several stages, preferably all phases I-IV, and the training model can distinguish cancer from non-cancer, regardless of the cancer stage.

[0130] Feature Importance

[0131] When biological samples are illuminated with infrared light, the absorbance of this light causes molecular excitation and transitions between vibrational states, producing an infrared spectrum. The biomolecules in serum vibrate in different planes, such as stretching and bending of bonds between chemical functional groups, and these vibrations are characteristic of these biomolecules. Spectral regions or specific wavenumbers that contribute to classification can be evaluated by feature importance analysis. Wavenumber regions found to be the most discriminative can be visualized in an importance chart ( Figure 3 ), and the top 3 important regions in each binary classification are shown in Tables 8 and 9. Figure 3 The wavenumber regions found to be most discriminative are shown. The importance values ​​were normalized to a maximum of 100 to make them more comparable. The variable importance measure here is based on a weighted sum of the absolute regression coefficients. These are weighted according to the sum of squares over the partial least squares (PLS) components selected for the discriminant model. With respect to C versus NC, the wavenumber region found to be of highest importance is Amide I (~1682 cm -1 ) and Amide II (~1578 cm -1 ) bands. These are the two largest peaks in the serum spectrum and contain information about overlapping bands associated with protein secondary structures (such as α-helices and β-sheets), so changes in these bands are often indicative of disease states (9). They are associated with NH bending vibrations and CN and CO stretching vibrations in protein molecules. In addition, for C relative to the NC model, it was found that the lower wavenumber end of the spectrum (~1163 cm -1 ) have high importance for CO and CC stretching vibrations and CH deformation vibrations associated with proteins and carbohydrates. Similar importance was reported in the amide II region for brain cancer relative to NCS classification, and lipid bands in the high wavenumber region were also shown to be significant in this model, appearing at ~2930 cm -1 The breast cancer model has a broader range of importance values ​​than the other cancers and has many narrow peaks throughout the spectrum. The top three regions of importance are ~1025 cm -1 (CO and CC stretching, COH deformation), protein amide III (~1260 cm -1 )(NH in-plane bending, CN stretching vibration), and the ~1337 cm that produces CH2 sway and CO stretching vibration in phospholipids and collagen -1 The importance trajectories for the colorectal and kidney cancer models are very similar, with the most important regions in both classifications being the amide I and II bands and the 1165 cm -1The importance of lung cancer features relative to NCS appears to be quite unique, as most of the important bands appear at the low end of the spectrum, mainly related to the symmetric (1074 cm -1 ) and asymmetric (1190 cm -1 ) PO stretching vibration and amide III region (1288 cm -1 ) are related to protein-related vibrations in the NCS-F model. The importance values ​​for ovarian and pancreatic cancer are also very similar, as the amide II band has the highest importance for both, but they differ in other top wavenumber regions. For ovarian cancer relative to the NCS-F model, ~1412 cm -1 and ~1135 cm -1 is also significant, while for pancreatic cancer relative to the NCS classifier, 1074 cm -1 (symmetric PO related to nucleic acids and phosphodiesters Telescopic, CO telescopic) and 1289 cm -1 Finally, the significant values ​​in the prostate cancer relative to the NCS-M model are mainly related to protein vibrations (amide I / II) and CO stretching in lipids and proteins together with CH and NH deformations (appearing at about ~1356 cm -1 )related.

[0132] in conclusion

[0133] The development of robust multi-cancer detection tests will be revolutionary in the field of cancer diagnosis, as earlier detection is crucial for inhibiting disease progression, improving patient prognosis and reducing mortality. In this large-scale spectral study, the inventors demonstrated Dxcover® cancer liquid biopsy, which is considered to be the world's first infrared spectroscopy-based blood test for early detection of multiple cancers. For binary classifiers for individual cancer types in symptomatic control patients, promising AUC values ​​for all cancers were reported: brain cancer (0.90), breast cancer (0.74), colorectal cancer (0.91), kidney cancer (0.91), lung cancer (0.90), ovarian cancer (0.85), pancreatic cancer (0.81) and prostate cancer (0.85). The results presented here show the great potential of this spectral method for early detection of multiple cancers. Importantly, the detection rate did not change significantly according to the cancer stage, indicating that the technology is very effective in detecting early tumors, which is a necessary condition for early cancer detection and patient survival.

[0134] Table 1

[0135]

[0136] *Total includes C and NC only; NCA patients are included in the NC set.

[0137] **No installment required.

[0138] ***Cancer stage and / or staging information was not recorded.

[0139] Table 2

[0140]

[0141] *No installment required.

[0142] **Cancer stage and / or staging information was not recorded.

[0143] ***Brain cancer is classified by grade as shown in Table 10.

[0144] Table 3

[0145]

[0146] Table 4

[0147]

[0148] *No installment required.

[0149] **Cancer stage and / or staging information was not recorded.

[0150] ***Brain cancer is classified by grade as shown in Table 10.

[0151] Table 5

[0152]

[0153] *No installment required.

[0154] Table 6

[0155]

[0156] Table 7

[0157]

[0158] *No installment required.

[0159] **Cancer stage and / or staging information was not recorded.

[0160] ***Brain cancer is classified by grade as shown in Table 10.

[0161] Table 8

[0162]

[0163] Table 9

[0164]

[0165]

[0166] Table 10

[0167]

[0168] Table 11

[0169]

[0170] References

[0171] 1. World Health Organization, Cancer. Accessed August 2021 (2021),(available at https: / / www.who.int / news-room / fact-sheets / detail / cancer).

[0172] 2. S. Testori, Public Health England, Cancer Research UK, What’s the Most Common Treatment For Cancer? Cancer patients diagnosed at an earlier stage are more likely to have surgery than chemotherapy (2017), (available at https: / / news.cancerresearchuk.org / 2017 / 10 / 26 / cancer-patients-diagnosed-at-an-earlier-stage-are-more-likely-to-have-surgery-than-chemotherapy / ).

[0173] 3.M. Arruebo, N. Vilaboa, B. Sáez-Gutierrez, J. Lambea, A. Tres, M.Valladares, Á. González-Fernández, Assessment of the Evolution of CancerTreatment Therapies. Cancers. 3, 3279–3330 (2011).

[0174] 4.G. D. Braunstein, Cedars-Sinai Medical Center and the David GeffenSchool of Medicine at UCLA, Los Angeles, CA, USA, J. J. Ofman, GRAIL, Inc.,Menlo Park, CA, USA, Criteria for Evaluating Multi-cancer Early DetectionTests. Oncology & Haematology. 17, 3 (2021).

[0175] 5.E. Crowley, F. Di Nicolantonio, F. Loupakis, A. Bardelli, Liquidbiopsy: monitoring cancer-genetics in the blood. Nat Rev Clin Oncol. 10, 472–484 (2013).

[0176] 6.G. Poste, Bring on the biomarkers. Nature. 469, 156–157 (2011).

[0177] 7.MDx Health, MDxHealth Select MDx Test. Accessed April 2021,(available at https: / / mdxhealth.com / tests / #selectMdx).

[0178] 8.Exosome Diagnostics GmbH, ExoDx Test. Accessed April 2021,(available at https: / / www.exosomedx.com / europe / our-technology).

[0179] 9.American Cancer Society, The Global Cancer Burden; Why GlobalCancer Rates are Rising. Accessed November 2021, (available at https: / / www.cancer.org / health-care-professionals / our-global-health-work / global-cancer-burden.html).

[0180] 10.H. J. Butler, P. M. Brennan, J. M. Cameron, D. Finlayson, M. G.Hegarty, M. D. Jenkinson, D. S. Palmer, B. R. Smith, M. J. Baker, Developmentof high-throughput ATR-FTIR technology for rapid triage of brain cancer. NatCommun. 10, 4501 (2019).

[0181] 11.J. M. Cameron, H. J. Butler, D. S. Palmer, M. J. Baker, Biofluidspectroscopic disease diagnostics: A review on the processes and spectralimpact of drying. Journal of Biophotonics. 11, e201700299 (2018).

[0182] 12.K. Hajian-Tilaki, Receiver Operating Characteristic (ROC) CurveAnalysis for Medical Diagnostic Test Evaluation. Caspian J Intern Med. 4,627–635 (2013).

[0183] 13.G. Putcha, A. Gutierrez, S. Skates, Multicancer Screening: OneSize Does Not Fit All. JCO Precision Oncology, 574–576 (2021).

Claims

1. A method for detecting whether a subject has cancer, regardless of the stage or type of cancer, the method comprising: A blood sample from the subject is subjected to a 400 to 4000 cm -1 Infrared spectral analysis of a certain wavelength to generate spectral characteristic features of the blood sample, wherein the spectral features of the blood sample are analyzed against representative features from previous subjects with or without cancer, wherein the previous subjects with cancer include subjects with different cancer types and different cancer stages, thereby detecting whether the subject has cancer based on the spectral features obtained from the subject.

2. The method according to claim 1, wherein the infrared spectroscopy analysis comprises ATR-infrared spectroscopy analysis. The method of claim 1 , wherein a background spectrum is acquired to provide correction for background environment.

4. The method according to claims 1 to 3, wherein the spectral analysis further comprises normalization, noise reduction and / or derivative as one or more pre-processing steps and / or Fourier Transform Infrared (FTIR) spectral analysis.

5. The method according to any preceding claim, wherein the infrared spectroscopy analysis of the blood sample further comprises detecting a type of cancer in a subject suffering from cancer based on the spectral signature obtained from the subject.

6. A method according to any preceding claim, wherein analysis of the spectral signature can be used to provide an indication of brain cancer, breast cancer, colorectal cancer, kidney cancer, lung cancer, ovarian cancer, pancreatic cancer and / or prostate cancer.

7. A method according to any preceding claim, wherein analysis against representative signatures from previous subjects comprises applying a trained model to the spectral signature.

8. The method according to claim 7, wherein the training model comprises a trained machine learning model, optionally a neural network, a support vector machine (SVM), a random forest (RF) decision tree, an ordered random forest (ORF) model.

9. The method according to claim 7 or 8, wherein the training model comprises a classifier or functions as a classifier by applying the probability threshold or a certain probability threshold to the probability value output by the training model.

10. The method according to claim 9 further comprises selecting and / or changing the probability threshold, thereby selecting and / or changing the specificity and / or sensitivity of the spectral feature analysis.

11. The method according to claim 9 or 10 further comprises selecting the probability threshold based on a receiver operating characteristic (ROC) curve for the training model, and / or selecting the training model from the training model set based on multiple ROCs for the training model set, for example, to obtain the desired specificity and / or sensitivity.

12. The method according to any preceding claim, wherein the analysis comprises (i) detecting whether the subject has cancer and / or (ii) detecting the type of cancer in the subject.

13. A method according to any preceding claim, wherein the spectral analysis comprises detection of vibrational modes and / or wavenumber regions.

14. The method according to claims 12 to 13, for detecting whether a subject has cancer and for detecting the type of cancer if the subject has cancer, the method being based on spectral features comprising information on one or more molecular vibrational modes.

15. The method of claim 14, wherein the molecular vibration mode is selected from the group consisting of: NH (in-plane) bending / deformation, CN stretching, CH stretching / deformation, CH2 stretching, CO stretching, CC stretching, C-OH deformation, asymmetric / symmetric PO stretching, CH2 swing and / or C=O stretching.

16. A method according to any preceding claim for detecting whether a subject has cancer and, if the subject has cancer, for detecting the type of cancer, said method being based on the presence of a wavelength in the range of 400 to 4000 cm -1 Identification of peaks at one or more wavenumber regions within a .

17. A method according to any preceding claim, wherein the method is performed in combination with other blood-based tests using a single blood draw, wherein a first aliquot is used to perform ATR-infrared spectroscopy analysis and a second aliquot comprising the remainder of the blood sample or a portion thereof is used to perform the other blood-based test.

18. A computer program product comprising computer readable instructions executable to perform the method according to any one of claims 1 to 17.

19. A training model configured to receive a blood sample from a subject comprising a range of 400 to 4000 cm -1 The ATR-IR spectroscopic analysis takes as input a spectral signature of a wavelength and provides an output representing whether the subject has cancer, regardless of the stage or type of cancer.

20. A method for training a model, the method comprising: A plurality of data sets representing spectral features of a plurality of subjects are received, the spectral features being derived from spectral measurements of blood samples from the subjects in the range of 400 to 4000 cm -1 wavelength obtained from ATR-infrared spectroscopy; receiving cancer data indicating, for at least some of the subjects, whether the subjects have cancer; and A model is trained to determine the probability of whether a patient has cancer based on the patient's spectral features, wherein Some subjects have cancer, and the subjects having cancer include subjects with different cancer types and different cancer stages.

21. The method of claim 20, wherein training the model comprises adjusting at least one parameter of the model to optimize the area under the ROC curve and / or provide a desired sensitivity and / or specificity for determining whether a patient has cancer.

Images