Surface-enhanced molecular fingerprint analysis for predictive diagnosis and disease monitoring

Abstract

The present disclosure relates generally to methods, devices, and systems for non-invasive, early-stage disease detection and ongoing health monitoring. A method for analyzing a sample is disclosed. The method includes preparing a solution with nanoparticles, mixing the solution with nanoparticles with a sample to create a reaction mixture, applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array, performing fourier-transform infrared (FTIR) spectroscopy on the dried thin-layer of the reaction mixture to produce at least one IR spectra, and analyzing at least one IR spectra with one or more machine learning models or algorithms.

Description

Attorney Docket No. 19178.0007WOU1SURFACE-ENHANCED MOLECULAR FINGERPRINT ANALYSIS FOR PREDICTIVE DIAGNOSIS AND DISEASE MONITORINGCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the priority benefit of U.S. Provisional Application Serial No. 63 / 642,469 filed on May 3, 2024, the disclosure of which is incorporated herein by reference in its entirety. To the extent appropriate, a claim of priority is made to the abovedisclosed application.FIELD

[0002] The present disclosure relates generally to methods, devices, and systems for non- invasive, early-stage disease detection and ongoing health monitoring. In some examples, the disclosure relates to the development and application of nanoparticle-enhanced infrared (“IR”) molecular spectral fingerprint analysis materials and techniques, combined with advanced machine learning and computational algorithms, implemented on a specialized infrared reflective optical sampling plate (“OIR”). In examples, this integrated approach utilizes resonant surface-enhanced infrared absorption (“rSEIRA”) technology to achieve highly sensitive and reproducible detection with superior signal-to-noise ratios. In embodiments, the platform is designed for accurate and precise analysis of subtle molecular-level changes occurring within biological fluids such as blood, serum, plasma, saliva, or urine. By leveraging these capabilities, the disclosure provides a powerful and scalable method for predictive diagnostics, early disease detection, and ongoing monitoring across a broad spectrum of health conditions, including but not limited to various forms of cancer, metabolic disorders, inflammatory diseases, infectious diseases, cardiovascular conditions, and neurological disorders.BACKGROUND

[0003] Early detection of diseases such as cancer, metabolic disorders, and infectious conditions is pivotal for improving patient outcomes, reducing treatment complexity, and lowering healthcare costs. For instance, early-stage breast cancer diagnosis can yield five-year survival rates of approximately 99%, whereas late-stage detection significantly reduces survival rates (American Cancer Society, ACS). According to the World Health Organization (WHO), over 2.3 million new cases of breast cancer are diagnosed globally each year, making it the most prevalent cancer among women.Attorney Docket No. 19178.0007WOU1

[0004] Traditional screening tools and approaches, such as imaging modalities (e.g., Mammography, CT, MRI), face multiple challenges that hinder their widespread and routine use in clinical settings. Mammography, commonly used for breast cancer screening, has notable limitations, including discomfort, radiation exposure, and reduced sensitivity in populations with dense breast tissue. Similarly, imaging techniques like computed tomography (CT) and magnetic resonance imaging (MRI), used for lung and colorectal cancers, respectively, can be expensive, resource-intensive, and associated with risks such as radiation exposure from CT or the need for contrast agents with MRI. High rates of false positives frequently result in unnecessary follow-up procedures and anxiety, while false negatives delay critical treatment, compromising patient outcomes (ACS; WHO).

[0005] There are limitations to other current approaches. Although accurate, procedures such as biopsies are invasive, costly, and unsuitable for widespread population-level screening or regular monitoring. Their invasive nature also limits the repeated sampling required for ongoing health assessments or disease progression monitoring. Molecular diagnostics, including genomic sequencing technologies, provide valuable information but often require extensive laboratory infrastructure, specialized personnel, and substantial operational costs. These methods frequently target predefined biomarkers, potentially overlooking broader molecular signatures indicative of early disease states or subtle biological alterations.

[0006] Due to its label-free, non-destructive, and highly informative nature, Infrared (“IR”) spectroscopy has found broad applicability in chemical, pharmaceutical, and biological fields, including disease diagnostics and biomarker identification (Baker et al., 2014). However, despite its utility, conventional IR spectroscopy faces significant limitations, particularly concerning sensitivity and specificity. Standard IR spectroscopy techniques, such as Fourier-transform infrared (“FTIR”) spectroscopy, generally exhibit insufficient sensitivity for detecting low-concentration molecular species and subtle biochemical alterations characteristic of early-stage disease or subtle pathological processes (Baker et al., 2016). These sensitivity limitations stem primarily from inherently weak absorption signals and limitations related to path length, sample preparation, and instrumental resolution (Kazarian & Chan, 2006).

[0007] Moreover, traditional IR spectroscopic approaches encounter substantial challenges related to reproducibility and consistency across different laboratories and experimental setups. Variations in sample handling, preparation methods, and environmental conditions can significantly impact spectral outcomes, thereby complicating clinical translationAttorney Docket No. 19178.0007WOU1 and implementation (Petibois & Desbat, 2010). Additionally, achieving adequate signal-to- noise ratios remains problematic, especially for samples with inherently weak molecular vibrations or those with complex biochemical compositions, such as bodily fluids or heterogeneous biological tissues (Movasaghi et al., 2008).

[0008] Addressing these limitations necessitates developing non-invasive, scalable, and cost-effective technologies capable of reliably detecting subtle molecular-level changes at early disease stages for sensitive and accurate early disease detection.SUMMARY

[0009] The present disclosure provides several embodiments for analyzing samples and detecting diseases using nanoparticle-enhanced infrared spectroscopy and machine learning.

[0010] In one embodiment, a method for analyzing a sample is disclosed. The method includes (a) preparing a solution with nanoparticles; (b) mixing the solution with nanoparticles with a sample to create a reaction mixture; (c) applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array; (d) performing fourier-transform infrared (FTIR) spectroscopy on the reaction mixture to produce at least one IR spectra; and (e) analyzing the at least one IR spectra with one or more machine learning models or algorithms. In embodiments, the method further includes after step (b) incubating the reaction mixture at room temperature. In embodiments, the method further includes after step (c) allowing the reaction mixture to air dry. In embodiments, the nanoparticles are AuNPs. In embodiments, the nanoparticles are nanorods or nanospheres. In embodiments, the nanoparticles comprise nanoshells or magnetic cores. In embodiments, the nanoparticles have a diameter between 30 nm to 250 nm. In embodiments, the nanoparticles have a diameter between 80 and 200nm. In embodiments, the nanoparticles are bare citrate nanospheres. In embodiments, the nanoparticles have surface chemistries in solution. In embodiments, the nanoparticles are magnetic and functionalized with specific capture molecules such as antibodies, aptamers, or receptor ligands. In embodiments, the nanoparticles are gold nanoparticles with a magnetic material. In embodiments, the solution with nanoparticles has a concentration between IxlO3to IxlO6particles per mm2. In embodiments, the reaction mixture further includes a molecular dye. In embodiments, the reaction mixture further includes a molecular dye selected from H&E, cresyl violet, brilliant cresyl blue, coomassie brilliant blue, bromophenol blue, trypan blue, toluidine blue, or aniline blue. In embodiments, the one or more machine learning models or algorithms are trained with an entire infrared transmittance or absorbance spectrum obtained from SEIRA- or rSEIRA-FTIR analyses as the input data. In embodiments, the one or moreAttorney Docket No. 19178.0007WOU1 machine learning models or algorithms are provided the entire spectral range from 4000 to 400 cm1or 8000 to 400 cm In embodiments, the one or more machine learning models or algorithms include a multilayer perceptron (MLP), convolutional neural networks (CNNs), gradient boosting methods such as LightGBM and XGBoost, support vector machines (SVMs), or a weighted ensemble of MLP, LightGBM and / or XGBoost model. In embodiments, the method further includes a series of data preprocessing steps such as spectral quality control, normalization, feature engineering, and feature selection or importance, applied to the entire input spectrum for each FTIR measurement prior to the step of analyzing the at least one IR spectra with one or more machine learning models or algorithms.

[0011] In another embodiment, a method for analyzing a biological sample is disclosed. The method includes (a) preparing an optical molecular fingerprinting solution (OFS); (b) mixing nanoparticles with a biological sample; (c) combining the OFS with the biological sample with the nanoparticles to create a reaction mixture (d) applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array; (e) performing fourier- transform infrared (FTIR) spectroscopy on the reaction mixture to produce at least one IR spectra; and (f) analyzing the at least one IR spectra. In embodiments, the OFS is an aqueous solution comprising about 0.33% methylene blue (MB), about 0.50% brilliant cresyl blue (BCV), and about 0.50% alcian blue (AB) molecular dyes. In embodiments, the OFS is an aqueous solution comprising 0.33% methylene blue (MB), 0.50% brilliant cresyl blue (BCV), and 0.50% alcian blue (AB) molecular dyes.

[0012] In yet another embodiment, a method for analyzing a biological sample is disclosed. The method includes: (a) preparing an optical molecular fingerprinting solution (OFS); (b) combining the OFS with a biological sample to create a reaction mixture; (c) applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array that has been precoated with nanoparticles; (d) performing fourier-transform infrared (FTIR) spectroscopy on the reaction mixture to produce at least one IR spectra; and (e) analyzing the at least one IR spectra. In embodiments, the OFS is an aqueous solution comprising about 0.33% methylene blue (MB), about 0.50% brilliant cresyl blue (BCV), and about 0.50% alcian blue (AB) molecular dyes. In embodiments, the OFS is an aqueous solution comprising 0.33% methylene blue (MB), 0.50% brilliant cresyl blue (BCV), and 0.50% alcian blue (AB) molecular dyes.

[0013] In another embodiment, a method for disease detection is disclosed. The method includes: (a) depositing a biological sample onto an infrared (IR) reflective sampling arrayAttorney Docket No. 19178.0007WOU1 plate configured for nanoparticle application to produce resonant surface-enhanced infrared absorption (rSEIRA), the plate or the biological sample containing nanoparticles; (b) illuminating the biological sample with IR radiation and collecting the resulting reflected IR spectra; (c) processing and analyzing the collected IR spectra with a machine learning model trained to identify molecular spectral fingerprints indicative of disease states; and (d) outputting a diagnostic result correlating to the presence or absence of disease. In embodiments, the disease being detected comprises: a cancer, including but not limited to breast cancer, lung cancer, colorectal cancer, prostate cancer, or other cancer types; a neurodegenerative disease; or a pre-malignant condition. In embodiments, the machine learning model comprises advanced computational models, including convolutional neural networks (CNNs), deep learning frameworks, gradient boosting methods (LightGBM, XGBoost), support vector machines (SVMs), or combinations thereof, trained on extensive and diverse patient-derived spectral datasets. In embodiments, the method further comprises integrating of spectral data of biomarkers, including circulating RNA, protein signatures, or other metabolites to enhance diagnostic accuracy. In embodiments, the nanoparticles are nanoparticle arrays which comprise gold, silver, or composite nanostructures specifically engineered and optimized to yield maximum electromagnetic field enhancement and spectral amplification at disease-relevant IR frequencies.

[0014] In yet another embodiment, a system for detecting early-stage diseases is disclosed. The method includes: (a) an IR reflective sampling plate engineered with one or more nanoparticle arrays that are either pre-coated onto a surface of the plate or provided in a solution mixture with a sample or an analyte, wherein the plate is configured to produce resonant surface-enhanced infrared absorption (rSEIRA); (b) an IR illumination source capable of emitting radiation across a relevant IR spectral range and a detector configured to capture reflected, enhanced IR spectral data; (c) a processing unit operatively linked to the detector, configured to execute machine learning models trained to analyze IR spectra and accurately predict the likelihood of early-stage disease presence based on molecular fingerprint analysis; and (d) a user interface configured to display information. In embodiments, the IR reflective sampling plate is specifically tailored for the detection of early-stage cancers, particularly breast cancer, and adaptable to other disease conditions by retraining the machine learning models with corresponding disease-specific datasets or reoptimizing the nanoparticle conditions. In embodiments, the processing unit employs advanced pattern recognition, spectral feature extraction, and predictive modeling algorithms to reliably identify and quantifyAttorney Docket No. 19178.0007WOU1 subtle molecular alterations within biological samples, thereby facilitating predictive diagnostics, preventive interventions, and ongoing disease monitoring. In embodiments, the method further includes automated or semi-automated sample preparation components designed to streamline sample handling, deposition, and drying processes, ensuring reproducibility and ease of integration into routine clinical workflows.BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings, which are incorporated in and constitute a part of the description, illustrate several aspects of the present disclosure. A brief description of the drawings is as follows:

[0016] FIG. 1 is an example of an assembly of an infrared (“IR”) reflective sample plate, according to embodiments of the present disclosure.

[0017] FIG. 2 is a flowchart of an example method of sample preparation, IR data acquisition, machine learning analysis, and diagnostic output, according to embodiments of the present disclosure.

[0018] FIG. 3 is a flowchart of an example method of sample preparation, IR data acquisition, and machine learning analysis, according to embodiments of the present disclosure.

[0019] FIG. 4A is a flowchart of an example experimental set up of a predictive model training through retrospective modeling, according to embodiments of the present disclosure.

[0020] FIG. 4B is a flowchart of an example experimental set up of a predictive model evaluating prospective test samples, according to embodiments of the present disclosure.

[0021] FIG. 5A is an example receiver operating characteristic (ROC) analysis curve of the retrospective model trained with rSEIRA spectral data showing superior area under the curve (AUC) compared to conventional methods, according to an embodiment of the present disclosure.

[0022] FIG. 5B is an example plot of the sensitivity and specificity as of function of threshold of the retrospective training model, according to an embodiment of the present disclosure.

[0023] FIG. 5C is an example receiver operating characteristic (ROC) curve of the prospective model, according to an embodiment of the present disclosure.

[0024] FIG. 5D is an example plot of the sensitivity and specificity as of function of threshold of the prospective training model, according to an embodiment of the present disclosure.Attorney Docket No. 19178.0007WOU1

[0025] FIG. 6A is a graph illustrating the sensitivity of the predictive threshold of the retrospective modeling, according to an embodiment of the present disclosure.

[0026] FIG. 6B is a graph illustrating the specificity of the predictive threshold of the retrospective modeling, according to an embodiment of the present disclosure.

[0027] FIG. 7A is a heatmap generated using guided backpropagation with Gradient- weighted Class Activation Mapping (Guided Grad-CAM) for breast cancer samples from the prospective test set, sorted by predicted probabilities, according to an embodiment of the present disclosure.

[0028] FIG. 7B is a heatmap generated using guided backpropagation with Gradient- weighted Class Activation Mapping (Guided Grad-CAM) for normal samples from the prospective test set, sorted by predicted probabilities, according to an embodiment of the present disclosure.

[0029] FIG. 8 A is an example unsupervised clustering analysis by t-SNE of breast cancer samples and normal samples using all spectral features (or wavenumbers) from IR spectroscopy without the use of nanoparticles, according to an embodiment of the disclosure.

[0030] FIG. 8B is an example unsupervised clustering analysis by t-SNE of breast cancer samples and normal samples using the top 100 spectral features (or wavenumbers) from IR spectroscopy as ranked by analysis of variance (ANOVA) score without the use of nanoparticles, according to an embodiment of the disclosure. FIG. 8C is an example unsupervised clustering analysis by t-SNE of breast cancer samples and normal samples using all spectral features (or wavenumbers) from IR spectroscopy with the use of nanoparticles, according to an embodiment of the disclosure.

[0031] FIG. 8D is an example unsupervised clustering analysis by t-SNE of breast cancer samples and normal samples using the top 100 spectral features (or wavenumbers) from IR spectroscopy as ranked by analysis of variance (ANOVA) score with the use of nanoparticles, according to an embodiment of the disclosure.

[0032] FIG. 9A is an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with a machine learning model without a prediction threshold set, according to an embodiment of the disclosure.

[0033] FIG. 9B is an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with a machine learning model with a prediction threshold set of greater or equal to 90%, according to an embodiment of the disclosure.Attorney Docket No. 19178.0007WOU1

[0034] FIG. 10A is an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure.

[0035] FIG. 10B is an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure.

[0036] FIG. 10C is an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure.

[0037] FIG. 11A is an example plot of colorectal cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure.

[0038] FIG. 1 IB is an example plot of lung cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure.

[0039] FIG. 11C is an example plot of prostate cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure.DETAILED DESCRIPTION

[0040] Embodiments of the present disclosure generally provide methods, systems, and devices that integrate nanoparticle-enhanced infrared molecular fingerprint analysis with advanced machine learning algorithms, utilizing a specially designed IR reflective sampling plate. This integration generates resonant surface-enhanced infrared absorption signals, significantly enhancing the sensitivity and accuracy for detecting subtle molecular changes within small-volume biological samples, such as serum, plasma, saliva, urine, or other bodily fluids.

[0041] For purposes of this specification, the following terms are specifically defined as follows:

[0042] “Infrared absorption spectrum” refers to a spectrum that is proportional to the wavelength dependence of the infrared absorption coefficient, absorbance, or similar indication of IR absorption properties of a sample. An example of an infrared absorption spectrum is the absorption measurement produced by a Fourier Transform Infrared (FTIR) spectrometer, i.e. an FTIR absorption spectrum. In general, infrared light will either be absorbed (i.e., a part ofAttorney Docket No. 19178.0007WOU1 the infrared absorption spectrum), transmitted (i.e., a part of the infrared transmission spectrum), or reflected. Reflected or transmitted spectra of a collected probe light can have a different intensity at each wavelength as compared to the intensity at that wavelength in the probe light source. It is noted that IR measurements are often plotted showing the amount of transmitted light as an alternative to showing the amount of light absorbed. For the purposes of this definition, IR transmission spectra and IR absorption spectra are considered equivalent as the two data sets as there is a simple relationship between the two measurements.

[0043] “Infrared source” and “source of infrared radiation” refer to one or more optical sources that generates or emits radiation in the infrared wavelength range, generally between 2-25 microns. The radiation source may be one of a large number of sources, including thermal or Globar sources, supercontinuum laser sources, frequency combs, difference frequency generators, sum frequency generators, harmonic generators, optical parametric oscillators (OPOs), optical parametric generators (OPGs), quantum cascade lasers (QCLs), interband cavity lasers (ICLs), synchrotron infrared radiation sources, nanosecond, picosecond, femtosecond and attosecond laser systems, CO2 lasers, microscopic heaters, electrically or chemically generated sparks, and / or any other source that produces emission of infrared radiation. The source emits infrared radiation in a preferred embodiment, but it can also emit in other wavelength ranges, for example from ultraviolet to THz. The source may be narrowband, for example with a spectral width of <10 cm-1 or <1 cm-1 less, or may be broadband, for example with a spectral width of >10 cm-1, >100 cm-1 or greater than 500 cm- 1. Broadband sources can be made narrow band with filters, monochromators and other devices. The infrared source can also be made up of one of discrete emission lines, e.g. tuned to specific absorption bands of target species.

[0044] “Interacting” in the context of interacting with a sample means that light illuminating a sample is at least one of scattered, refracted, absorbed, aberrated, diverted, diffracted, transmitted, and reflected by, through and / or from the sample.

[0045] “Signal indicative of’ refers to a signal that is mathematically related to a property of interest. The signal may be an analog signal, a digital signal, and / or one or more numbers stored in a computer or other digital electronics. The signal may be a voltage, a current, or any other signal that may be readily transduced and recorded. The signal may be mathematically identical to the property being measured, for example explicitly an absolute phase signal or an absorption coefficient. It may also be a signal that is mathematically related to one or moreAttorney Docket No. 19178.0007WOU1 properties of interest, for example including linear or other scaling, offsets, inversion, or even complex mathematical manipulations.

[0046] “Spectrum” refers to a measurement of one or more properties of a sample as a function of wavelength or equivalently (and more commonly) as a function of wavenumber.

[0047] The terms “about” or “approximate” and the like are synonymous and are used to indicate that the value modified by the term has an understood range associated with it, where the range can be ±20%, ±15%, ±10%, ±5%, or ±1%.

[0048] The term “substantially” is used to indicate that a result (e.g., measurement value) is close to a targeted value, where close can mean, for example, the result is within 80% of the value, within 90% of the value, within 95% of the value, or within 99% of the value.

[0049] The infrared (“IR”) reflective device disclosed herein utilizes digital signal processing techniques to identify and cancel out reflected signals and alters the physical design of the system to minimize points of reflection.

[0050] Disclosed herein is an infrared (“IR”) reflective sampling device for FTIR-based molecular fingerprint analysis for detecting cancer and other diseases. In embodiments, the sampling device includes a reflective high-throughput biopsy plate. In some examples, the reflective element includes a concave mirror, or an N x N array of concave mirrors, for example with an N=1 to 400.

[0051] As used throughout this disclosure, the term “reflective” means substantially more reflective than absorptive or transmissive. For example, a reflective material may be 99% or more reflective, or 95% or more reflective, or 90% or more reflective. Additionally, reflectivity is often a function of wavelength. Water, for example, is highly absorptive in the ultraviolet regime while nearly transparent in the visible regime. The methods, systems, and devices described herein are used for testing of infrared absorption by biological samples, and thus this disclosure generally describes materials that are reflective in the infrared regime. However, it may be that the methods, systems, and devices described herein can be used for other kinds of tests in which visible or ultraviolet light is of importance. In any case, there may be a predetermined wavelength range of interest (such as, for example, 500nm-1500nm, or 780nm- lOOOnm, etc.) for which the methods, systems, and devices described herein are best suited. Depending upon the range of interest, different materials or thicknesses of those materials can be used that will provide more reflection than combined transmission and absorption. While not explicitly described herein, those of skill in the art will understand how to substituteAttorney Docket No. 19178.0007WOU1 materials that are reflective in other wavelengths to accomplish the results described herein for other predetermined wavelength ranges.

[0052] Referring now to FIG. 1, an example of the assembly 100 of an infrared (IR) reflective sample plate 116 embodying the subject matter of the present disclosure is shown. Assembly 100 of the IR reflective sample plate 116 may comprise creation or acquisition of a suitable base plate 102, application of an IR reflective layer 104, and ionization of the IR reflective wells 106.

[0053] At 102, assembly 100 begins with a suitable base plate 108. A suitable base plate 108 may be designed or selected according to considerations such as throughput, sample size, and use setting.

[0054] Base plate 108 may be made of any suitable material, with consideration to the plate’s use in FTIR analysis. Some nonlimiting examples of suitable materials for base plate 108 include acrylonitrile butadiene styrene (ABS), polystyrene, polypropylene, polycarbonate, polyetherimide, glass, quartz, and other thermally stable materials. In embodiments, base plate 108 may be made of disposable, reusable, or other recyclable material. Base plate 108 may be formed by a number of different processes, such as injection molding or vacuum forming, e.g., vacuum metallization.

[0055] In examples, base plate 108 is designed according to experimental and sample parameters. Some nonlimiting examples of suitable base plate designs include single sample well plates and multi -well plates, such as 1-, 12-, 24-, 48-, 96-, or 384-well formats. In some examples, a standard design that is readily accommodated by a standard plate reader, such as a plate reader for existing designs of FTIR spectroscopy instruments.

[0056] Each well 110 of one or more wells of base plate 108 may be configured to produce a focal point 112 in a placement to eliminate reflection interferences and achieve a desired signal amplitude. In examples, the desired signal amplitude may be an optimal signal amplitude. In embodiments, wells having a shallow depth, and a curved cross-section may be preferred. For example, each well may have a parabolic cross-section with a depth (d), which may also be understood as a height of the well, and a radius (r) configured to optimize the location of a focal point 112. In an example referring to a 96-well format, well depth (d) may range from 0.2 to 2 mm and radius (r) may range from 1.5 to 5 mm, values which are optimized to produce the optimal focal point to eliminate reflection interferences and achieve the optimal signal amplitude.Attorney Docket No. 19178.0007WOU1

[0057] At 104, an IR reflective layer 114 is applied to base plate 108. In examples, IR reflective layer 114 may be a coating, a film, or a foil, such as aluminum, copper, nickel, silver, gold, dielectric mirror, multilayer optical films, etc.

[0058] IR reflective layer 114 can be applied to cover an individual well, such as well 110, or across the entire plate, such as base plate 108. In examples, IR reflective layer 114 may be applied over the full surface of the base plate, such that all plate surfaces are covered, or may be applied only to one or more wells, such that non-well surfaces and one or more wells may not be covered. To ensure smooth and consistent IR reflective coating or layer, an underlying base coat can be applied on the base plate prior to IR reflective coating. The undercoat, or base coat, can be materials compatible with or serves as primer to enhance the adhesion of IR reflective upper coat. The base coat can be an ultra-violet (UV)-curable base coat, epoxy base, polyurethane base coat, and others. In one embodiment, primers under the PARYLENE® brand can be used, which are materials providing barrier layers that prevent moisture, corrosion, and solvent transpiration, while also providing smoothing for an adjacent layer.

[0059] Reflective properties of materials such as aluminum, copper, nickel, chromium, silver, and gold in the visible light spectrum are well-known, but their behavior in the infrared (IR) spectrum is particularly important for various applications. These metals have electrons in the conduction band that can oscillate in response to electromagnetic radiation, such as IR light.

[0060] In the IR spectrum, materials like gold and silver reflect IR radiation effectively due to their high reflectance of longer wavelengths. Gold reflects up to 98% of IR radiation, making it extremely efficient for applications requiring insulation from heat. Silver, although it tarnishes, also provides excellent IR reflectivity when clean. Aluminum, while less reflective than gold and silver, still reflects a significant amount of IR radiation and is more cost-effective. Copper and nickel have more moderate reflectivity in the IR range but are still useful due to their other physical properties.

[0061] The use of these materials in the IR spectrum is multifaceted. Gold’s excellent reflectivity is utilized in satellite and space telescope components to protect against the sun’s heat. Silver’s high IR reflectivity is often used in thermal insulation and in coatings to improve energy efficiency. Aluminum, due to its cost-effectiveness and good IR reflectivity, is widely used in thermal rescue blankets and in architectural designs to reflect IR and reduce heating. Copper finds its use in heat exchangers due to its ability to reflect IR and its excellent thermal conductivity. Nickel, with its moderate IR reflectivity, is often used as a coating on other materials to add protection against IR radiation while providing corrosion resistance.Attorney Docket No. 19178.0007WOU1

[0062] Multiple methods of applying IR reflective layer 114 to base plate 108 are envisioned. IR reflective layer 114, which may be, for example, aluminum, may be vacuum metallized to base plate 108. This method may be preferred in some instances, such as those where cost effectiveness is a significant consideration, due to its simplicity and low cost. In vacuum metallization, also referred to as physical vapor deposition, metal is evaporated in a vacuum environment and then allowed to condense on the substrate’s surface to form a thin film. This method is widely used for creating reflective surfaces on items like mirrors, automotive parts, and decorative items. It provides a uniform coating and can be used with a variety of metals like aluminum, which is often chosen for its reflective properties. Vacuum metallization of the plates with, for example, aluminum to make the wells and / or plates reflective, and it is an easy and low-cost process, being more cost effective than adding a reflective metallized film or aluminum foil. Other aspects of embodiments of the curved optical IR-reflective array plate referenced herein are further described in PCT / US2024 / 014511 filed on February 5, 2024, which is incorporated by reference in its entirety. The optical IR-reflective array plate or device will be referred to as “OIR plate” throughout the remainder of this application and should be understood to mean the plate or device describe herein or any plate or device substantially similar with the same attributes and functionalities as the base plate 108 described above.

[0063] FIG. 2 depicts a flowchart of an example method 200 of sample preparation, high- throughput sampling, IR data acquisition, fingerprint selection, and machine learning analysis and diagnostic output, according to embodiments of the present disclosure.

[0064] In operation 202, a sample is prepared for IR analysis such as Surface-Enhanced Infrared Absorption (“SEIRA”; use of the OIR plate without nanoparticles), Resonant Surface- Enhanced Infrared Absorption (“rSEIRA”; use of the OIR plate with nanoparticles), or Fourier- transform infrared spectroscopy (“FTIR”) analysis. In the embodiment depicted in operation 202 of FIG. 2, the sample to be analyzed is serum. In other embodiments, the sample could be plasma, saliva, biofluids or reaction products therefore, isolate analytes, another type of biological sample or other substance suitable for analysis by FTIR analysis.

[0065] As shown in the example illustrated in FIG. 2 at operation 202, the IR signal from the serum can be enhanced by using a solution with nanoparticles. SEIRA leverages plasmonic nanostructures, typically noble metal nanoparticles such as gold (Au) and silver (Ag), to significantly enhance IR absorption signals through the amplification of local electromagnetic fields. These enhancements occur due to the excitation of localized surface plasmon resonancesAttorney Docket No. 19178.0007WOU1(LSPRs), which concentrate electromagnetic energy near the nanostructure surfaces, dramatically increasing the effective IR cross-section of molecules adsorbed or positioned nearby (Neubrech et al., 2017). The efficiency of SEIRA primarily depends on the morphology, material composition, and arrangement of the plasmonic nanoparticles.

[0066] To further boost the sensitivity, nanoparticle arrays can be precisely engineered to achieve resonance matching with specific vibrational frequencies of molecules, a technique known as rSEIRA. By carefully adjusting nanoparticle geometry (shape, size, aspect ratio), spacing, and periodic arrangement, plasmon resonances can be tuned to match molecular vibrations, leading to pronounced signal amplification (Rodrigo et al., 2015; Neumann et al., 2013). The resonant enhancement in rSEIRA originates from the strong coupling between molecular vibrations and plasmon resonances, yielding sharper and more intense absorption features than non-resonant SEIRA setups (Brown et al., 2013).

[0067] In the embodiment illustrated at operation 202 in FIG. 2, a serum sample is mixed with a solution with nanoparticles can be made by preparing a stock suspension of gold nanoparticles (“AuNPs”) at the desired final concentration of IxlO3to IxlO6particles per mm2. In other embodiments, the stock suspension may be of silver nanoparticles (“AgNPs”). In some embodiments, the nanoparticles can be nanorods, nanospheres, nanostars, other tailored geometries, or a combination thereof. Shapes such as nanospheres, nanorods, and nanostars can be tailored to optimize localized surface plasmon resonances (LSPRs). In embodiments, the nanoparticles can have compositions such as composite nanoshells or magnetic cores. For example, in embodiments, magnetic gold nanoparticles (“magnetic AuNPs)”, comprise a gold coating of magnetic nanoparticle core made of materials like iron oxide, provide additional functionality of magnetically driven separation. These magnetic AuNPs with broad or targeted surface functionalization can be effectively utilized to isolate, concentrate, and enrich target biomolecules or analytes from complex biological samples, significantly improving assay sensitivity and reducing background interference. When coupled with resonant surface- enhanced infrared absorption Fourier-transform infrared spectroscopy (rSEIRA-FTIR), magnetic AuNPs enhance the analytical power and robustness of the platform, enabling highly sensitive detection and precise molecular characterization essential for clinical diagnostics and biomarker discovery.

[0068] In various embodiments, the nanoparticles can be between approximately 5 nm to 250 nm. In other embodiments, the nanoparticles can between 30 nm to 250 nm in size. Gold nanospheres of diameter in the range of 30 nm to 200 nm can be used in embodiments. In otherAttorney Docket No. 19178.0007WOU1 embodiments, the optimal size range for gold nanospheres and analytes of interest using FTIR in the mid-IR region is between 80 nm and 200 nm.

[0069] In some embodiments, the nanoparticles have surface chemistries in solution, such as when mixed with analytes. For example, AuNPs can be functionalized further to specifically target a broad range of biomolecules and analytes, including proteins, peptides, nucleic acids, metabolites, and even whole cells or pathogens. Functionalization strategies commonly involve attaching biomolecular ligands such as antibodies, aptamers, peptides, nucleic acid sequences, and receptor ligands onto the nanoparticle surfaces, enabling highly specific molecular recognition and binding interactions. Such tailored functionalization enhances the selectivity, sensitivity, and specificity of detection assays. In embodiments, the nanoparticle surface chemistries, include bare citrate, citrate-capped, polyethylene glycol (PEG)-coated, amine- functionalized such as branched polyethyleneimine (BPEI), carboxyl-functionalized (e.g., lipoic acid, etc.), polyvinylpyrrolidone (PVP), and streptavidin-conjugated surfaces, or combinations thereof. These example surface chemistries facilitate diverse functionalization strategies and specific targeting of a broad class of analytes for molecular fingerprint analysis. In embodiments, nanoparticles can be engineered to achieve resonance matching with specific vibrational frequencies of target molecules through varying things such as the geometry, coating, spacing, and periodic arrangement of the nanoparticles to maximize signal amplification.

[0070] In a particular embodiment, bare citrate AuNPs provide a robust and broad rSEIRA-FTIR application for the detection and classification of breast, lung, colorectal, and prostate cancer relative to the normal / healthy control serum samples for unsupervised clustering analysis. Detection sensitivity and specificity of other cancer types, including multiple myeloma, have also been found to be improved in said embodiment using bare citrate AuNPs. In one embodiment, IR analysis of 10 pL of mixture per plate well of 100 nm Au nanospheres in citrate used with serum samples in a ratio of (10: 10) between serum and Au nanospheres in citrate illustrated better separation for discriminating molecular spectral fingerprints between breast cancer samples and normal samples when compared with IR analysis of 10 pL of mixture per plate well of 100 nm Au nanorods (about 12nm - 20 nm in diameter) in citrate used with serum samples in a ratio of (10: 10) between serum and Au nanorods in citrate.

[0071] In operation 202 depicted in FIG. 2, the solution with nanoparticles can include an optical molecular fingerprinting solution (“OFS”) for molecular structural enrichment. In suchAttorney Docket No. 19178.0007WOU1 an embodiment, the stock suspension of nanoparticles can be premixed directly into an OFS to create a nanoparticle solution prior to being combined with the sample, such as a serum sample. In another embodiment, the sample may be mixed with the OFS and then the solution with nanoparticles is added to the mixture. In embodiments, mixing a biological sample with nanoparticles facilitates effective adsorption and structural modification of the analytes in the sample. In some embodiments, the OFS is an Oncodea™ Optical Fingerprinting Solution. In embodiments, the OFS is a mixture of molecular biology grade water and at least one dye.

[0072] In other embodiments, OFS is a mixture of molecular biology grade water, various blue dyes, and potassium thiocyanate. In some embodiments, the OFS includes molecular binding reagents that facilitate the structural modification and enrichment of cancer-associated biomolecules, such as glycans, glycosylated proteins, and nucleic acids — particularly microRNAs. In some embodiments, the OFS includes a mixture of molecular binding reagents and can include extracellular vesicle lysing or homogenizing agents when needed, as further described in PCT / US2021 / 071334 filed on September 1, 2021, which is incorporated by reference in its entirety.

[0073] In some non-limiting examples, the use of OFS modifies the overall structural configurations and corresponding vibrational frequencies of biomolecules, effectively enhancing their unique, disease-specific infrared spectral signatures. This structural enhancement process substantially reduces background interference and enhances the distinguishing features essential for accurate disease identification and classification. Consequently, OFS-treated samples yield significantly improved spectral fingerprint specificity and reproducibility compared to untreated samples, thereby substantially boosting the accuracy and reliability of disease classification through rSEIRA-FTIR analysis. In contrast, analyses conducted without OFS demonstrate lower accuracy due to greater variability in spectral fingerprints and increased background noise. Thus, the combination of OFS with nanoparticle-enhanced SEIRA and rSEIRA provides a powerful, robust, and precise analytical framework for early-stage disease detection and precise molecular-level diagnostics.

[0074] In embodiments, colorimetric dyes, such as those commonly used in microscopy, can be added to the OFS to provide reagent visualization during preparation, mixing, and application onto the OIR for analysis. Molecular dyes such as H&E, cresyl violet, brilliant cresyl blue, Coomassie brilliant blue, bromophenol blue, trypan blue, toluidine blue, methylene blue, aniline blue, alcian blue, and others can be used, either alone or in combination with oneAttorney Docket No. 19178.0007WOU1 another. The inclusion of one or more colorimetric dyes in the OFS can minimize technical errors and variations and to improve traceability.

[0075] In one embodiment, the OFS is designed to enable colorimetric visualization for sample preparation traceability and to facilitate targeted molecular binding. In that embodiment, the dual functionality minimizes background noise and enhances the precision of infrared (IR) vibrational molecular fingerprint analysis. In one example, the OFS is an aqueous solution comprising about 0.33% methylene blue (MB), about 0.50% brilliant cresyl blue (BCV), and about 0.50% alcian blue (AB) molecular dyes (shown as % v / v), which are optimized fortheir specific interactions with cancer-associated biomolecules, including nucleic acids, proteins, and glycans.

[0076] In some examples, the concentration of MB in the OFS (% v / v) may be from about 0.01% to about 5%, about 0.01% to about 4.5%, about 0.01% to about 4 %, about 0.01% to about 3.5%, about 0.01% to about 3%, about 0.01% to about 2.5%, about 0.01% to about 2%, about 0.01% to about 1.5%, about 0.01% to about 1%, about 0.01% to about 0.5%, about 0.01% to about 0.1%, about 0.1% to about 5 %, about 0.5% to about 5%, about 1 % to about 5%, about 1.5% to about 5%, about 2% to about 5%, about 2.5% to about 5%, about 3% to about 5%, about 3.5% to about 5%, about 4% to about 5%, or about 4.5% to about 5%.

[0077] In some examples, the concentration of BCV in the OFS (% v / v) may be from about 0.01% to about 5%, about 0.01% to about 4.5%, about 0.01% to about 4 %, about 0.01% to about 3.5%, about 0.01% to about 3%, about 0.01% to about 2.5%, about 0.01% to about 2%, about 0.01% to about 1.5%, about 0.01% to about 1%, about 0.01% to about 0.5%, about 0.01% to about 0.1%, about 0.1% to about 5 %, about 0.5% to about 5%, about 1 % to about 5%, about 1.5% to about 5%, about 2% to about 5%, about 2.5% to about 5%, about 3% to about 5%, about 3.5% to about 5%, about 4% to about 5%, or about 4.5% to about 5%.

[0078] In some examples, the concentration of AB in the OFS (% v / v) may be from about 0.01% to about 5%, about 0.01% to about 4.5%, about 0.01% to about 4 %, about 0.01% to about 3.5%, about 0.01% to about 3%, about 0.01% to about 2.5%, about 0.01% to about 2%, about 0.01% to about 1.5%, about 0.01% to about 1%, about 0.01% to about 0.5%, about 0.01% to about 0.1%, about 0.1% to about 5 %, about 0.5% to about 5%, about 1 % to about 5%, about 1.5% to about 5%, about 2% to about 5%, about 2.5% to about 5%, about 3% to about 5%, about 3.5% to about 5%, about 4% to about 5%, or about 4.5% to about 5%.

[0079] In some examples, the concentration of molecular dye in the OFS (% v / v) may be from about 0.01% to about 5%, about 0.01% to about 4.5%, about 0.01% to about 4 %, aboutAttorney Docket No. 19178.0007WOU10.01% to about 3.5%, about 0.01% to about 3%, about 0.01% to about 2.5%, about 0.01% to about 2%, about 0.01% to about 1.5%, about 0.01% to about 1%, about 0.01% to about 0.5%, about 0.01% to about 0.1%, about 0.1% to about 5 %, about 0.5% to about 5%, about 1 % to about 5%, about 1.5% to about 5%, about 2% to about 5%, about 2.5% to about 5%, about 3% to about 5%, about 3.5% to about 5%, about 4% to about 5%, or about 4.5% to about 5%.

[0080] In some embodiments, the molecular dye percentages can be empirically determined to generate robust cancer type classification by rSEIRA-FTIR through direct dye mixture without washing or removal of free, unbound dyes. In examples, the final dye content in OFS (% w / v) for MB, BCV, and AB is about 0.005%, about 0.0005%, and about 0.005%, respectively. In some examples the final dye content in OFS is diluted. Some non-limiting example of appropriate dilution ratios of a molecular dye percent weight to the OFS volume are 1 : 1000, 1 :500, 1 : 100, 1 :50, 1 :20, 1 : 10, 1 :5, 1 : 1, 5: 1, 10: 1, 20: 1, 50: 1, 100: 1, or 1000:1. Other non-limiting examples of suitable dilution ratios include 5: 1, 6: 1, 7:1, 8: 1, 9: 1, 10: 1, 11 : 1, 12: 1, 13: 1, 14: 1, and 15: 1. In various embodiments, MB, BCV, and AB are molecular dyes that serve as molecular binding and modifying reagents, as well as provide a bluish solution for visualization and traceability.

[0081] In embodiments, the OFS also contains 0.05% (20 mM) potassium thiocyanate (KSCN). In other embodiments, the OFS also contains 0.1% (20 mM) potassium thiocyanate (KSCN). In some embodiments, the OFS also contains 0.15% (20 mM) potassium thiocyanate (KSCN). In other embodiments, the OFS also contains 0.2% (20 mM) potassium thiocyanate In some embodiments, the OFS also contains 0.25% (20 mM) potassium thiocyanate (KSCN). (KSCN). In various embodiments, the OFS also contains 0.3% (20 mM) potassium thiocyanate (KSCN). In some embodiments, the OFS also contains 0.35% (20 mM) potassium thiocyanate (KSCN). In embodiments, the OFS also contains 0.4% (20 mM) potassium thiocyanate (KSCN). In other embodiments, the OFS also contains 0.45% (20 mM) potassium thiocyanate (KSCN). In some embodiments, the OFS also contains 0.5% (20 mM) potassium thiocyanate (KSCN).

[0082] In embodiments, the KSCN can serve as an IR sample loading control. In embodiments, KSCN provides a distinct and consistent IR absorbance peak, the amplitude of which correlates with the total serological biomolecular content in the reaction mixture. In some embodiments, this correlation ensures accurate normalization of the IR spectra, improving the reliability and reproducibility of molecular fingerprint analysis.Attorney Docket No. 19178.0007WOU1

[0083] In embodiments, the inclusion of these components in precise concentrations ensures optimal performance in detecting subtle molecular changes associated with cancer, while maintaining consistency across samples and experimental conditions. It is understood that the specified concentrations (percentages by either v / v or w / v) are examples used, and their inclusion is possible at various ranges with a standard deviation of ± 50% for each component. In some embodiments, an OFS dye solution is supplemented with the nanoparticles at 8 x 107parti cles / mL to formulate an OFS with nanoparticles ready for a reaction mixture with serum or plasma samples (or other liquid biospecimens) at a 1 :1 ratio as described further below.

[0084] In operation 202 illustrated in FIG. 2, the solution with nanoparticles is mixed with an OFS containing colorimetric molecular dye for visualization in a manner that the nanoparticles are in suspension and uniformly dispersed throughout the solution. In embodiments, equal volumes of the solution with nanoparticles that contains the OFS, and the sample are combined. In some non-limiting examples, the mixture of OFS with a sample results in a single-step reaction that results in the staining of the EVs and lipids, allowing contrast and enhancement with nanoparticle arrays without additional extraction, isolation, or sample processing steps. In embodiments, as a quality control OFS can be tested to reduce the background noise and improve the signal-to-noise ratios in the discriminative cancer-specific spectral fingerprint regions of the rSEIRA spectra.

[0085] In the embodiment depicted in operation 202, the combination of the solution with nanoparticles and the sample is a reaction mixture. In embodiments, the reaction mixture is incubated at room temperature between 5 and 30 minutes, to facilitate effective biomolecule binding, structural modification, and nanoparticle-analyte absorption.

[0086] In operation 204, the reaction mixture is deposited onto the IR reflective arrays of the OIR plate. The reflective surface of the OIR plate is suitable for thin-layer sample presentation and facilitates reflection-based IR measurements that ensure maximum signal recovery by efficiently directing IR radiation back toward the detector, thus improving sensitivity and signal-to-noise ratios. Each well or sampling area on the OIR plate has a specific curvature and surface treatments, facilitating consistent sample drying, uniform thin-layer formation, and reliable signal acquisition.

[0087] In a different embodiment, rather than being in a reaction mixture with the solution with nanoparticles, the sample may be applied to an OIR plate that has been pre-coated with nanoparticles on the IR reflective arrays at various concentrations. By incorporating precisely fabricated nanoparticles onto the IR reflective arrays of the OIR, the system achievesAttorney Docket No. 19178.0007WOU1 reproducible, uniform, and optimized enhancement conditions via rSEIRA. This integration enables highly sensitive detection and characterization of subtle molecular changes at low concentrations. The carefully engineered nanoparticle arrays on these reflective plates promote consistently localized surface plasmon resonances (LSPRs), thereby facilitating robust and uniform enhancement of infrared signals essential for clinical-grade molecular fingerprint analysis and early disease detection. The combined attributes of these sampling plates — high reflectivity, reproducible nanoparticle positioning, and straightforward handling — make them ideal for routine clinical diagnostics and high-throughput optical screening applications, potentially transforming the effectiveness and scalability of early-stage disease detection and ongoing health monitoring.

[0088] In operation 206, the sample deposited on the OIR plate is illuminated with infrared radiation to capture spectral fingerprint data. The reflected IR spectra (transflectance), which are significantly enhanced by the resonant surface-enhanced infrared absorption (rSEIRA) effect facilitated by the integrated nanoparticles, are captured. The signals captured by the IR detector include signals from proteins, nucleic acids, lipids, carbohydrates, and other metabolites within a sample.

[0089] Cancer-specific molecular “fingerprints” can be generated by the complex mixture of biomolecules identified by IR light absorption in the sample, such as a serum. In one such example, by using IR spectroscopy, differences in serum molecular compositions can be documented to generate unique spectrometric signatures, or fingerprints, specific to different health conditions. This technology works based on the principle that blood samples from patients with developing cancer contain biomolecules with different structural compositions than those from healthy individuals. Unlike other techniques that focus specifically on DNA, mRNA, miRNA, or the exosomes, this approach detects the complex mixture of all components present in the serum and within the larger EVs and exosomes. These include differences in the biochemical components, proteins, circulating cell-free DNA / RNA / miRNA, and glycans (carbohydrates) that are both soluble in serum as well as packed inside the lipid encapsulated extracellular vesicles (EVs), such as exosomes, present in the serum. This is beneficial, because cancer-related components are also found in larger microvesicles as well as in the apoptotic bodies of dying cancer cells. By measuring the biochemical and physical properties of the serum and EV components, the differences in the molecular fingerprints can be linked to the presence of different types of cancer.Attorney Docket No. 19178.0007WOU1

[0090] In embodiments, measurements of blood serum for molecular fingerprints can be taken using the visible and near-infrared (VIS / NIR) to mid-infrared (MIR) light absorbance spectra as the base component combined with the modulated NIR / MIR spectra and the predefined biomarker-specific fluorescence spectra. This allows sensing devices embodying the technology to be compact, portable, and energy efficient. In some embodiments, in addition to the base spectrometric signatures and biomarker-specific signals that define the molecular fingerprints, the modulated absorbance spectra using specific chemical compounds, peptides, and antibodies designed to bind to specific sets of cancer-specific proteins, nucleic acids, and carbohydrates to produce broad molecular fingerprints for a breast cancer detection platform and enhance early cancer detection accuracy.

[0091] In operation 208, selected cancer fingerprints are determined. Using artificial intelligence assisted signal processing and feature extraction, a fingerprint can be engineered and selected from the rSEIRA spectra.

[0092] In operation 210, the predictive model performs training and classification using the spectral fingerprint to detect tissue-specific cancer types. In embodiments, the captured IR spectra are processed through a predictive model such as at least one advanced machine learning model or algorithm specifically trained using extensive, annotated datasets containing both diseased and healthy samples. The machine learning models or algorithms are trained with the entire processed infrared transmittance or absorbance spectrum obtained from SEIRA- or rSEIRA-FTIR analyses serving as the input data in embodiments. Rather than preselecting spectral regions or specific features, the entire spectral range (typically 4000 to 400 cm or 8000 to 400 cm ') is provided to the models or algorithms in some embodiments. In embodiments, each model is preceded by data preprocessing steps (spectral QC, normalization, feature engineering, and feature selection or importance) applied to the entire input spectrum for each FTIR measurement. This approach allows the models or algorithms to autonomously determine and optimize the most discriminative spectral features and patterns relevant to disease classification and biomarker discovery. By analyzing the entire spectral dataset, the models or algorithms can identify subtle yet crucial molecular variations that might otherwise be overlooked. Integrating these advanced models or algorithms with spectral data across different molecular spectral regions (e.g., circulating nucleic acids, proteins, lipids, and metabolites) significantly enhances diagnostic accuracy, providing early-stage detection capabilities across a range of conditions, including various cancers (such as breast, lung, colorectal, prostate, others), metabolic disorders, neurodegenerative diseases (such asAttorney Docket No. 19178.0007WOU1Alzheimer’s disease, Parkinson’s disease, Multiple sclerosis (MS), amyotrophic lateral sclerosis), and emerging infectious diseases.

[0093] In some embodiments, the machine learning models automatically identify and interpret complex spectral fingerprints associated with early-stage disease states. In embodiments, interpreting the complex and multidimensional spectral data generated by SEIRA and rSEIRA-FTIR analyses necessitates advanced computational tools. Machine learning (“ML”) and artificial intelligence (“Al”) models offer powerful methods to extract meaningful molecular signatures and subtle spectral patterns indicative of early disease states, even in the absence of predefined biomarkers. Deep learning approaches, particularly convolutional neural networks (“CNNs”), have demonstrated exceptional proficiency in analyzing spectral data due to their ability to recognize spatially correlated patterns and hierarchical features within spectra (Ghosh et al., 2019; Thrift et al., 2020). Additional neural network frameworks, such as feedforward neural networks (MLP), offer customizable architectures and robust computational efficiency, enabling precise model tuning and efficient learning from complex, high-dimensional data. Ensemble methods such as Light Gradient Boosting Machine (LightGBM) and Extreme Gradient Boosting (XGBoost) provide alternative powerful predictive approaches that efficiently handle large datasets, facilitate interpretability, and consistently achieve high accuracy across diverse biomedical data (Ke et al., 2017; Chen & Guestrin, 2016). Support Vector Machines (SVM) remain a valuable ML approach for spectral classification tasks, particularly useful when dealing with smaller or imbalanced datasets due to their robustness, excellent generalization capability, and effectiveness in highdimensional feature spaces (Noble, 2006).

[0094] Referring now to FIG. 3, a flow chart of an example method 300 of detecting a disease or condition using nanoparticle-analyte interactions on optical infrared reflective arrays and machine learning models according to an embodiment of the disclosure is illustrated. Various approaches for facilitating nanoparticle-analyte interactions on the optical infrared reflective arrays are described as follows.

[0095] At operation 302, a sample is prepared. In one embodiment, a small volume, approximately 40 pL for repeated analyses or approximately 5 pL per individual analysis well, of a patient’s biological sample, such as serum or plasma, is obtained.

[0096] At operation 304, a reaction mixture that contains the sample is created. In one embodiment, nanoparticles are premixed with a sample of serum or analytes. In said embodiment, nanoparticles are directly mixed and incubated with a sample of biological fluidsAttorney Docket No. 19178.0007WOU1 such as serum, plasma, or specific analyte solutions prior to their application onto an OIR plate. Alternatively, a solution with nanoparticles at a desired concentration is prepared in an OFS containing at least one molecular dye or dye mixture. In some embodiments, the solution with nanoparticles is a fresh stock suspension of gold nanoparticles (AuNPs) at the desired final concentration of IxlO3to IxlO6particles per mm2. The solution with nanoparticles and OFS is then mixed with the sample, such as a biological fluid or specific analyte solution, to form a reaction mixture. In some embodiments, equal volumes of the solution with nanoparticle and OFS and a biological sample (e.g., serum, plasma, saliva, or isolated analytes) are combined to form a reaction mixture. The reaction mixture is then incubated. The incubation of the nanoparticle solution with the biospecimens facilitates the effective adsorption of analytes onto the nanoparticle surface while allowing the OFS reagents to bind to and modify the structural compositions of the analytes. This enhances the local electromagnetic fields and significantly amplifies the modified analyte's IR spectral signatures. In embodiments, the incubation time period lasts between 5 and 30 minutes.

[0097] In other embodiments, a reaction mixture consists of a solution with nanoparticles and a sample, such as a biological fluid or specific analyte solution. In embodiments, a reaction mixture consists of nanoparticles and a sample, such as a biological fluid or a specific analyte solution. In other embodiments, a reaction mixture consists of nanoparticles and a sample, such as a biological fluid and a specific analyte solution. In some embodiment, a reaction mixture consists of an OFS and a sample, such as a biological fluid or specific analyte solution. In some embodiments, the method 300 does not include operation 304 such that it proceeds with the sample from operation 302 to operation 304.

[0098] At operation 306, the nanoparticles can be magnetically captured, and the target analyte can be enriched. Operation 306 is an optional step in method 300. In an embodiment, magnetic gold nanoparticles (magnetic AuNPs) — comprising gold nanoparticles integrated with magnetic materials like iron oxide — can be employed for targeted analyte capture and enrichment. These magnetic AuNPs are functionalized with specific capture molecules (e.g., antibodies, aptamers, or receptor ligands) that selectively bind targeted biomolecules within complex biological fluids. Following incubation during operation 304, a magnetic field is applied to efficiently isolate and concentrate analyte-bound magnetic AuNPs. The isolated nanoparticles are washed to remove any nonspecifically bound components. The samples or analytes remain bound to the magnetic AuNPs, facilitating highly selective and sensitive rSEIRA-FTIR analysis with minimal background interference.Attorney Docket No. 19178.0007WOU1

[0099] At operation 308, the sample, the reaction mixture containing the sample combined with a solution with nanoparticles, or the purified nanoparticle-analyte complexes are deposited onto an IR reflective sampling plate. In embodiments, the volume deposited on the IR reflective sampling plate is between 1-10 pL. In some embodiments, purified nanoparticle-analyte complexes are mixed with OFS prior to being deposited onto an IR reflective sampling plate. The sample or reaction mixture is allowed to dry uniformly. In embodiments, the sample or reaction mixture is allowed to dry under ambient or gentle heat, such as between 35 °C and 45°C, for 30 to 60 minutes. Sample thin (micro) layers can be calorimetrically visualized by the naked eye or under the microscope to ensure uniformity as a quality control process.

[0100] In embodiments, the IR reflective sample plate is an OIR plate. In some embodiments, the IR reflective sample plate is a nanoparticle-enhanced IR reflective sampling plate. In embodiments, the reflective arrays of an OIR plate are precoated with nanoparticles to create stable and uniform plasmonic substrates. This approach ensures consistent and reproducible nanoparticle distribution and simplifies the assay workflow. In one such embodiment, analytes or biological samples are subsequently applied directly or mixed with the OFS (with molecular dyes) and applied onto a nanoparticle-coated reflective plate. Upon drying, the analytes or biological samples interact with the nanoparticles' enhanced electromagnetic fields, producing robust rSEIRA signals suitable for detailed molecular fingerprint analysis.

[0101] At operation 310, data is acquired through FTIR analysis. The prepared and dried OIR plate is inserted into the FTIR spectrometer’s reflective autosampler accessory or microscope attachment for a transflectance measurement. The FTIR instrument parameters are set, typically using a spectral range of 4000 to 400 cm1or may be extended up to 8000 cm-1 with a spectral resolution of 4 cm Either IR transmittance or absorbance spectra are acquired, averaging multiple scans (typically 16-128 scans) per measurement to ensure optimal signal- to-noise ratios.

[0102] At operation 312, data analysis is conducted. Baseline correction, normalization, and spectral smoothing are performed as required. Multivariate statistical analysis, data preprocessing, and machine learning are conducted to interpret spectral data for accurate biomarker identification and disease classification. Sophisticated algorithms — including multilayer perceptron (“MLP”), convolutional neural networks (“CNNs”), gradient boosting methods such as LightGBM and XGBoost, support vector machines (“SVMs”), and a weightedAttorney Docket No. 19178.0007WOU1 ensemble of MLP, LightGBM and / or XGBoost models — are utilized to analyze complex, highdimensional IR spectral datasets.

[0103] These algorithms are trained using large and diverse patient-derived datasets, encompassing spectra from healthy controls and various disease conditions. For example, the preliminary studies performed using the nanoparticle-enhanced rSEIRA-FTIR data trained with over 5,000 patient samples demonstrate strong potential for clinical diagnostics in multiple high-prevalent cancer types, including breast, lung, colorectal, and prostate cancers.

[0104] By leveraging full-spectrum data inputs, the machine learning models autonomously identify subtle molecular fingerprint patterns indicative of pathological processes, thereby enabling early and accurate disease detection and risk stratification prior to clinical manifestation. Unexpectedly, CNN models have been shown in this disclosure to provide highly robust and accurate detection of breast cancer down to Stages 0 and I at greater than 96% sensitivity and specificity (healthy samples) in a blinded prospective clinical validation study. Similar success is found in this disclosure with the MLP and weighed ensemble models in multi-class detection of breast, lung, colorectal, and prostate cancer at above 93% sensitivity and greater than 97% specificity (detection of healthy samples) based on independent test sets.

[0105] Operation 312 can conclude by generating a clear diagnostic result indicating the likelihood or probability of early-stage disease presence, thus providing actionable information to clinicians to support timely and informed medical interventions.

[0106] In embodiments, the method 300 is compatible with commercially available automation systems. In some embodiments, automated or semi-automated sample preparation components designed to streamline sample handling, deposition, and drying processes can be used to ensure reproducibility and ease of integration into the method 300. Some such embodiments can include the use of robotic liquid handling / pipetting, sample depositing on the OIR plate, drying, and FTIR scanning.

[0107] Referring now to FIG. 4A, a flow chart illustrating an experimental set up for retrospective machine learning modeling performed by training on a retrospective IR spectra dataset and evaluating performance on an independent test set drawn from the same dataset is shown, according to an embodiment of the disclosure. In an embodiment, 20% of normal (NC) and breast cancer (BC) female serum samples are first randomly selected and left out as the test set, and the remaining serum sample IR spectra data are then subjected to a stratified split into training and validation sets following an 80% / 20% ratio. To prevent data leakage, allAttorney Docket No. 19178.0007WOU1 replicates from the same patient are constrained to be within a single data partition. In embodiments, model training is terminated using an early-stopping criterion, specifically when the validation cross-entropy loss ceases to improve. In some embodiments, to reduce sampling- induced variance, the entire training-validation-test partitioning process is repeated 50 times.

[0108] FIG. 4B depicts a flow chart that outlines an experimental set up for an artificial intelligence training procedure where a model is trained on a retrospective dataset using a single randomized 80% / 20% split for training and validation. This trained model is then evaluated across seven independent repeats of a prospective cohort according to an embodiment of the disclosure.

[0109] FIG. 5A displays ROC curves derived from 50 independent realizations of a retrospective cohort (shown in gray), alongside a mean threshold-averaged ROC curve (black), showing an area under the curve (AUC) of 0.993, according to an embodiment of the disclosure. In embodiments, for each test realization, the predicted class probability is determined by taking the median across three technical replicates.

[0110] FIG. 5B shows plots of sensitivity and specificity as functions of the prediction threshold across the same 50 realizations of the retrospective cohort described with reference to FIG. 5 A, according to an embodiment of the disclosure. Each realization’s results are depicted as faded lines, while the mean sensitivity and specificity curves are overlaid in solid lines, respectively. In embodiments, two threshold values, corresponding to 97% mean sensitivity and 97% mean specificity, are selected and applied to a prospective test set across seven independent replicates.[OHl] FIG. 5C illustrates ROC curves for seven independent repeats of a prospective cohort (gray lines), along with the threshold-averaged ROC curve (black), according to an embodiment of the disclosure. In the depicted embodiment, the AUC of the ROC curve achieved 0.994.

[0112] FIG. 5D presents threshold-dependent sensitivity and specificity curves for the prospective cohort across the same seven repeats, according to an embodiment of the disclosure. Individual repeats are shown as faded lines, with the mean sensitivity and specificity plotted in solid black and gray, respectively.

[0113] FIG. 6A depicts the sensitivity across various breast cancer stages within a prospective cohort based upon the optimized threshold that was obtained from a retrospective test set, according to an embodiment of the disclosure. In some embodiments, the optimized threshold obtained from the retrospective test set is the value depicted in FIG. 5B. In theAttorney Docket No. 19178.0007WOU1 embodiment depicted in FIG. 6A, the sensitivity optimized result is shown which uses the threshold based on the 97% mean sensitivity.

[0114] FIG. 6B depicts the specificity across various breast cancer stages within a prospective cohort based upon the optimized threshold that was obtained from a retrospective test set, according to an embodiment of the disclosure. In some embodiments, the optimized threshold obtained from the retrospective test set is the value depicted in FIG. 5D. In the embodiment depicted in FIG. 6B, the specificity optimized result is shown which uses the threshold based on the 97% mean specificity.

[0115] FIG. 7A shows the heatmaps generated using guided backpropagation with Gradient-weighted Class Activation Mapping (Guided Grad-CAM) for breast cancer samples from a prospective test set, sorted by predicted probabilities, according to an embodiment of the disclosure. In the depicted heatmap, 60 breast cancer samples were analyzed and sorted for sensitivity optimization. Only samples from the first prospective repeat are shown in the depicted heatmap. Each row in the heatmap represents a sample, and the color intensity indicates the wavenumber regions that contribute most significantly to the model's predictions. From the figure, the wavenumber range 1050-790 cm1is identified as the most critical region, commonly associated with phosphate groups, carbohydrates, and nucleic acids.

[0116] FIG. 7B shows the heatmaps generated using guided backpropagation with Gradient-weighted Class Activation Mapping (Guided Grad-CAM) for normal samples from a prospective test set, sorted by predicted probabilities, according to an embodiment of the disclosure. In the depicted heatmap, 125 normal samples were analyzed and sorted for sensitivity optimization. Only samples from the first prospective repeat are shown in the depicted heatmap. Each row in the heatmap represents a sample, and the color intensity indicates the wavenumber regions that contribute most significantly to the model's predictions. From the figure, the wavenumber range 1050-790 cm1is identified as the most critical region, commonly associated with phosphate groups, carbohydrates, and nucleic acids.

[0117] FIG. 8A shows an example unsupervised clustering analysis by t-SNE of breast cancer samples and normal samples using all spectral features (or wavenumbers) from IR spectroscopy without the use of nanoparticles, according to an embodiment of the disclosure. In the embodiment depicted in FIG. 8A, the term “all features” refers to the entire absorbance spectra (all wavenumbers measured) from the FTIR that were used as inputs without feature selection. In embodiments, these features include: absorbance peaks and shapes at specific IR wavenumbers; vibrational signatures of biomolecules such as proteins, lipids, nucleic acids,Attorney Docket No. 19178.0007WOU1 and carbohydrates; and enhanced spectral regions amplified by the rSEIRA effect. In some embodiments, prior to machine learning, preprocessed data are accessed and visualized based on unsupervised analysis, such as by t-SNE or principal component analysis, among other techniques.

[0118] FIG. 8B shows an example unsupervised clustering analysis by t-SNE of breast cancer samples and normal samples using the top 100 spectral features (or wavenumbers) from IR spectroscopy as ranked by analysis of variance (ANOVA) score without the use of nanoparticles, according to an embodiment of the disclosure.

[0119] FIG. 8C shows an example unsupervised clustering analysis by t-SNE of breast cancer samples and normal samples using all spectral features (or wavenumbers) from IR spectroscopy with the use of nanoparticles, according to an embodiment of the disclosure. In the example illustrated in FIG. 8C, feature selection was not necessary for the sample classification as normal or having breast cancer as compared to the example illustrated in FIG. 8A because the use of nanoparticles illustrates enhanced discrimination between normal and breast cancer samples. In FIG. 8C, the use of nanoparticles also produced highly consistent molecular fingerprints between the replicates of each sample, as shown by the tight clustering among the replicates. In the embodiment depicted in FIG. 8C, the term “all features” refers to the entire absorbance spectra (all wavenumbers measured) from the FTIR were used as inputs without feature selection. In embodiments, these features include: absorbance peaks and shapes at specific IR wavenumbers; vibrational signatures of biomolecules such as proteins, lipids, nucleic acids, and carbohydrates; and enhanced spectral regions amplified by the rSEIRA effect. In some embodiments, prior to machine learning, preprocessed data are accessed and visualized based on unsupervised analysis, such as by t-SNE or principal component analysis, among other techniques.

[0120] FIG. 8D shows an example unsupervised clustering analysis by t-SNE of breast cancer samples and normal samples using the top 100 spectral features (or wavenumbers) from IR spectroscopy as ranked by analysis of variance (ANOVA) score with the use of nanoparticles, according to an embodiment of the disclosure.

[0121] FIG. 9A shows an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with a machine learning model without a prediction threshold set, according to an embodiment of the disclosure. In the embodiment depicted in FIG. 9A, independent daily testing for breast cancer in samples was conducted across 11 days and the results are depicted as an average across 11 different daily tests, with a standard deviation ofAttorney Docket No. 19178.0007WOU1 about 6.5% (1.3% SEM) for sensitivity and about 3.6% (1.2% SEM) for specificity. In the plot shown in FIG. 9A, 931 patients with breast cancer were tested.

[0122] FIG. 9B shows an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with a machine learning model with a prediction threshold set of greater or equal to 90%, according to an embodiment of the disclosure. In the embodiment depicted in FIG. 9B, independent daily testing for breast cancer in samples using nanoparticles was conducted across 11 days and the results are depicted as an average across 11 different daily tests,

[0123] FIG. 10A shows an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure. In the plot depicted in FIG. 10A, the first replicate of three replicated measurements taken without a prediction probably threshold is illustrated.

[0124] FIG. 10B shows an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure. In the plot depicted in FIG. 10B, the second replicate of three replicated measurements taken without a prediction probably threshold is illustrated.

[0125] FIG. 10C shows an example plot of breast cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure. In the plot depicted in FIG. 10C, the first replicate of three replicated measurements taken without a prediction probably threshold is illustrated. FIG. 10 A- 10C show a strong reproducibility across the three replicate plots depicted with a standard deviation of -0.6% for sensitivity and -0.4% for specificity.

[0126] FIG. 11 A shows an example plot of colorectal cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure.

[0127] FIG. 1 IB shows an example plot of lung cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure.

[0128] FIG. 11C shows an example plot of prostate cancer samples and normal samples analyzed by IR spectroscopy with the use of nanoparticles and a machine learning model, according to an embodiment of the disclosure.Attorney Docket No. 19178.0007WOU1EXAMPLESExample 1

[0129] Unsupervised clustering analysis and machine learning on spectral data was performed to identify the most effective nanoparticle configurations for specific cancer or disease types. IR analysis of 10 pL of mixture per IR plate well of 100 nm Au nanospheres in citrate used with serum samples in a ratio of (10: 10) between serum and Au nanospheres in citrate illustrated better separation for discriminating molecular spectral fingerprints between breast cancer samples and normal samples when compared with IR analysis of 10 pL of mixture per plate well of 100 nm Au nanorods (about 12nm - 20 nm in diameter) in citrate used with serum samples in a ratio of (10: 10) between serum and Au nanorods in citrate.Example 2

[0130] Differentiation of cancer-associated biomarkers and metabolites based upon their molecular structural compositions through vibrational analysis and the process of identifying unique cancer-associated molecular spectral fingerprints was performed with three approaches.Structural Feature Enhancement

[0131] The first approach involved the application of chemical reagents to biological samples to enhance differential structural features of certain biomolecules, particularly RNAs and glycans / glycosylated proteins, to reduce background noise and improve a cancer-specific spectral fingerprint prior to FTIR analysis.Patient samples and inclusion criteria

[0132] Human blood serum samples were sourced from multiple commercial biobank sites located in the United States that are established providers of human biological specimens, and clinical information with large networks of Institutional Review Board (“IRB”) approved collection sites throughout the US. These samples were de-identified and included essential pathological and clinical information to ensure they provide a diverse demographic.

[0133] The samples included cancer patients at stages I to IV in the categories of breast (invasive ductal, invasive lobular), lung (non-small cell and small cell), prostate and colorectal. Breast cancer samples also included ones at stage 0 ductal carcinoma in situ (DCIS). Other cancers not included in those four categories were placed in a separate category. Other cancers included melanoma, lymphoma, bladder, head and neck, ovarian, pancreatic, myeloma, kidney, endometrial, cervical, bone, leukemia, etc. Only cancer patients who had not undergone treatment were included to eliminate potential confounding spectral changes due to treatment.Attorney Docket No. 19178.0007WOU1The non-cancer donors (normal; no history of cancer; approximately 1 : 1 male / female ratio) included those with other underlying conditions, such as obesity, hypertension, diabetes, inflammatory disease, chronic obstructive pulmonary disease, chronic bronchitis, bronchial asthma, heart failure, etc. The racial distribution of the samples consisted of approximately 80% Caucasian, 17% African American, 1% Hispanic American, 1% Asian American and 1% Native American and other races.Sample preparation

[0134] Each sample was analyzed at least in triplicates. Only samples that had not undergone more than three freeze-thaw cycles were used for to minimize degradation of serum components.

[0135] Blood serum samples were prepared by taking a whole blood sample from a patient in a 2 mL microtube and centrifuging it at 1,500 x g for 10 minutes to separate out the red blood cells. The samples were then shipped from the supplier on dry ice. The serums were then aliquoted into 0.2 mL tubes and stored in a -80°C freezer for future use. Prior to sample preparation, patient serum samples were thawed on ice for 10 minutes.Oncodea™ Optical Fingerprinting Solution (“OOFS”)

[0136] OOFS was aliquoted into 1.5 mL borosilicate glass vials. A 1 mL pipette was used to transfer 1.4 mL of OOFS to each vial. The vials were then labeled with “Dye soln” and technician’s initials and the date. The aliquots were stored in a 4°C fridge for future use. During sample analysis, one plate needed approximately 0.62 mL of OOFS, therefore one dye aliquot was enough for two IR plates. A fresh solution with nanoparticles was prepared by adding 0.02 mL of nanoparticles into one OOFS dye aliquot. The vial was then marked with “NP” to indicate the addition of nanoparticles.

[0137] Quality control for OOFS with and without nanoparticles involved obtaining a spectral fingerprint using FTIR. During sample preparation, a reference background was placed on two wells, one with the dye mixture only and the other with OOFS containing nanoparticles. If the fingerprint was abnormal, the plate was repeated with new OOFS.

[0138] OIR plates were treated with plasmonic energy before sample loading. This was done using the Henniker Plasma Vacuum System HPT- 100 made by Henniker Scientific (Runcorn, UK). The surface of the OIR plates was activated with plasma to increase surface energy of the wells, which allowed higher adhesion of liquids being dispensed onto the surface and allowed the mixture with the sample to evenly spread across the well of the plate.Attorney Docket No. 19178.0007WOU1

[0139] In a 96-well PCR plate, 20 pL of OOFS containing nanoparticles was mixed with 20 pL of blood serum at room temperature in a ratio of 1 : 1. 10 pL of the mixture was then loaded onto the plate wells in triplicate. After the plate was loaded, it was placed into a dehydration chamber set to 42°C and dried for 1 hour. Plates with dried serum samples waiting for analysis were stored in a dehydration chamber set to 20% RH and 23°C. Alternatively, they were stored in an airtight box containing desiccant at room temperature.Intentional Sampling

[0140] Fourier-transform IR spectroscopy was employed to obtain the spectral molecular fingerprint for each sample, with the methodology enhanced by resonant surface-enhanced IR absorption (rSEIRA). Spectra were obtained via Invenio-S (Bruker Optics, Billerica, MA) or Spectrum 3 (PerkinElmer, Shelton, CT) FTIR spectrometer with an XY Autosampler attachment (Pike Technologies®, USA). Each patient sample was replicated on three wells. Spectra for each sample were measured between 8000 and 400 cm’1(4cm'1resolution, 32 coadded scans). Background spectra were collected (64 co-added scans) at the start and the midpoint of each analysis for atmospheric correction.

[0141] Spectra were quality checked by ensuring the absorbance was in an expected range at key wavelengths. Pre-processing of spectra signatures included a second derivative (Savitsky-Golay filter window size of 9 and second order polynomial) followed by Standard Normal Variate (SNV) normalization. Second derivates can reveal additional peaks, while SNV normalizes the spectra to have a mean of 0 and standard deviation of 1.Spectral preprocessing, background subtraction, and feature selectionMachine Learning Modeling

[0142] For model development, serum samples were collected from retrospective cohorts (N=4,120) consisting of healthy controls (n=2 ,263), breast cancer patients from stages 0 to IV (n=752), and a mixture of various cancer types (n=l,105) and used to train a deep learningbased ML algorithm to identify molecular fingerprints indicative of breast cancer. For the breast cancer versus normal sample classification, the available retrospective dataset, which comprised preprocessed spectral signatures, served as the input for training the deep learningbased predictive model. The main objective of this modeling was to identify spectral signatures indicative of breast cancer. This was formulated as a binary classification problem setting, where the goal was to discriminate between Non-Cancerous (“NC”) and Breast Cancer (“BC”) cases based on the input spectral features.Attorney Docket No. 19178.0007WOU1

[0143] The available dataset included spectral features from multiple cancer types, which presented a challenge for modeling the binary classification task. A simple approach would have been to select only the NC and BC examples for model training, but this would have significantly reduced the size of the training set, resulting in diminishing model's performance due to the loss of spectral examples. To address this challenge, a two-step transfer learning strategy was implemented.

[0144] The first step was multiclass classification. In this first step, the model was pretrained to perform a multiclass classification task to discriminate between NC and various cancer types (lung, colorectal, prostate, and other cancer types). The second step was binary classification. Once the model pre-training in the first step was completed, the trained model's final weights were transferred and used as the initial weights for the binary classification task. In this step, the model was fine-tuned to specifically classify NC versus BC. This transfer learning strategy allowed the feature representations learned in Step 1 to improve classification performance in Step 2. By pre-training the model on a multiclass classification task, it captured discriminative spectral signatures associated with various cancer types, which could then be fine-tuned to improve the performance of NC versus BC classification. This approach ensured full utilization of the available dataset, including spectral examples of different cancer types, to enhance the performance of the final binary classification model.Predictive Model Performance Evaluation

[0145] The Receiver Operating Characteristic (ROC) curve was selected to evaluate the overall classification performance of the predictive model. The threshold averaging ROC was used to compute the average curve across multiple repeats. Sensitivity and specificity were used to evaluate the model at specific classification threshold, which are defined as:TPSensitivity =TP + FNTNSpecificity =FP + TN

[0146] Where TP, TN, FP, and FN denoted true positives, true negatives, false positives, and false negatives, respectively, of the confusion matrix. Since each patient sample was analyzed in three replicates, the final prediction of the sample was determined using majority voting based on replicate predictions.

[0147] The NC versus BC modeling was evaluated on two experimental setups using independent test sets from the retrospective cohort and the prospective cohort, respectively.Retrospective Cohort ExperimentAttorney Docket No. 19178.0007WOU1

[0148] In the first experiment, modeling performance was evaluated on an independent test set from the retrospective dataset. The full cohort comprised of 4120 patients, of whom 1857 were confirmed cancer cases, either in the pre-treatment phase or actively undergoing treatment status, while 2263 were classified as Non-Cancer (NC). Within the cancer group, patients were further categorized based on their cancer type: breast (n=752), prostate (n=405), colorectal (n=254), lung (n=247), and other cancers (n=199).

[0149] Specifically, 20% of NC and BC female cases were first randomly selected as the test set, while the remaining data was partitioned randomly into training and validation sets using an 80% / 20% split. All replicates from a single patient sample were constrained to only belong in one of the sets to prevent data leakage into the test set. The optimal model parameterization was selected based on the lowest validation cross-entropy loss. To reduce the variability in train / validation / test partitioning, this process was repeated 50 times, each realization with a different partition. For each realization, the test performance metrics were recorded, and the final reported values were the averages of these metrics along with their 95% confidence intervals.

[0150] FIG. 5A displays the test ROC curves from 50 independent realizations (shown in gray color), and its overall threshold-averaged ROC curve (shown in black color) with an area under the ROC curve (AUC) of 0.993. The prediction of each test sample was determined based on the median of prediction probabilities of three replicates, reflecting a near-perfect discrimination between the NC and BC test patient samples.

[0151] FIG. 5B illustrates the selection of the prediction threshold based on maximizing sensitivity while maintaining specificity at 97% on the validation set (specificity-optimized), or maximizing specificity while maintaining sensitivity at 97% on the validation set (sensitivity-optimized). Two threshold values, corresponding to 97% mean sensitivity and 97% mean specificity, were selected and applied to the prospective test set.Prospective Cohort Experiment

[0152] In the second experiment, modeling performance was evaluated on a prospective cohort through a blinded study. During model training, the whole retrospective cohort was randomly split into 80% / 20% training / validation sets, and optimal model parameterization was determined based on the same condition as the retrospective cohort experiment. The trained model was then deployed to identify the unknown samples from a blinded, prospectively collected sera from an independent multi-center cohort (n=185) that included breast cancer patients at various stages (n=60, Stages 0-III) and healthy controls (n=125). Within BC, 14Attorney Docket No. 19178.0007WOU1 were stage 0, 35 were stage I, 6 were stage II, 3 were stage III, and 2 had unknown stages. All patients in the prospective cohort were female. The average performance metric across 7 repeats of the prospective cohort was reported to ensure statistical significance.

[0153] FIG. 5C illustrates the ROC curves from 7 independent repeats (in gray) alongside the overall threshold-averaged ROC curves (in black) of the prospective cohort. Similar to FIG. 5A, the ROC curves shown in FIG. 5C are derived based on the median of prediction probabilities obtained from three replicates per patient sample, with the resulting overall AUC of 0.994. FIG. 5D presents threshold-dependent sensitivity and specificity curves for the prospective cohort across the same seven repeats. Individual repeats are shown as faded lines, with the mean sensitivity and specificity plotted in solid black and gray, respectively.

[0154] The sensitivity and specificity of the prospective cohort were evaluated using the same decision threshold determined from the retrospective test. FIG. 6A shows the sensitivity of each BC stage based on sensitivity-optimized predictions as follows - Stage 0: 100.0% ± 0%, Stage I: 99.0% ± 2.5%, Stage II: 93% ± 5.8%, Stage III: 100.0% ± 0%, Stage NA: 100.0% ± 0.0%, and the overall sensitivity of 99.0% ± 2.9%. Correspondingly, the horizontal dotted line in FIG. 6A represents the test specificity of 94%. FIG. 6B shows the sensitivity of each BC stage based on specificity-optimized predictions as follows - Stage 0: 96.0% ± 4%, Stage I: 96.0% ± 2.5%, Stage II: 93% ± 5.8%, Stage III: 90.0% ± 20%, Stage NA: 100.0% ± 0.0%, and the overall sensitivity of 95.0% ± 2.9%. Correspondingly, the horizontal dotted line in FIG. 6B represents the test specificity of 96%.Example 3

[0155] Further preliminary feasibility studies based on 5-fold cross-validation analysis for lung cancer detection demonstrated a sensitivity of 94% (212 / 226 samples) and a specificity of 99.4% (2146 / 2158 healthy samples), with only 1% inconclusive results. A plot with this data was generated and is illustrated in FIG. 11B. For colorectal cancer, the platform achieved a sensitivity of 96% (238 / 247 samples) and specificity of 99.5% (2151 / 2160 healthy samples), resulting in just 0.6% inconclusive samples. A plot with this data was generated and is illustrated in FIG. 11 A. In prostate cancer preliminary analyses, sensitivity reached 93% (353 / 380 samples) and specificity 99.4% (2140 / 2153 healthy samples), with 1.6% inconclusive results. A plot with this data was generated and is illustrated in FIG. 11C.Example 3

[0156] Unsupervised t-SNE clustering analysis of the preprocessed FTIR spectral data with and without gold nanoparticle rSEIRA from normal and breast cancer samples, with orAttorney Docket No. 19178.0007WOU1 without feature selection based on ANOVA score ranking was performed. The clustering analysis was performed without machine learning.

[0157] First, sample preparation was performed. A small volume of serum or plasma was collected from a patient as a sample, 40 pL was collected for repeated analyses (approximately 5 pL per individual analysis well). The sample was mixed with an OncodeAi Optical Fingerprinting Solution (OOFS), a reagent containing nanoparticles. The nanoparticles were functionalized to enhance the detection of cancer-associated biomolecules such as glycosylated proteins, microRNAs, and metabolites.

[0158] Second, the nanoparticles were integrated. The nanoparticles were either be precoated onto the OncodeAi IR Sampling Device (OIR) or mixed directly with the biological sample in the OFS solution. These nanoparticles induced resonant surface-enhanced infrared absorption (rSEIRA), amplifying the vibrational signals of biomolecules.

[0159] Third, spectral acquisition was performed. The prepared sample was deposited onto the OIR sampling plate, which was a disposable, multi-well infrared reflective array. The sample was dried to form a uniform thin layer, ensuring consistent signal acquisition. The sample was illuminated with infrared (IR) radiation, and the reflected IR spectra were captured. The rSEIRA effect enhanced the spectral signals, allowing for the detection of subtle molecular changes associated with cancer.

[0160] Fourth, spectral preprocessing was performed. The raw IR spectra underwent preprocessing steps, including: (1) Baseline correction: To remove background noise; (2) Normalization: To standardize the data for comparison; (3) Spectral smoothing and second derivative-transformed: To reduce noise and enhance signal clarity; and (4) Feature selection: Key spectral features (e.g., second derivative of absorbance peaks at specific wavenumbers) were identified by statistical analysis, such as scored by ANOVA, and extracted based on the top ranking for further analysis.

[0161] FIG.s 8A-11D represent the unsupervised t-SNE clustering analysis of the preprocessed FTIR spectral data with and without gold nanoparticle rSEIRA from normal and breast cancer samples, with or without feature selection based on ANOVA score ranking. The clustering analysis was performed without machine learning.Example 4

[0162] An optical fingerprinting solution (OFS) that allows for the normalization of an FTIR spectrum graph output for the early detection of cancer from blook serum samples was made.Attorney Docket No. 19178.0007WOU1

[0163] First, materials were obtained. Molecular Biology Grade (MBG) water (tested for nuclease and bacteria contaminants) was obtained from Sigma Aldrich. Methylene blue was obtained from Sigma Aldrich. Cresyl blue was obtained from Sigma Aldrich. Alcian blue was obtained from Vector Labs. Potassium thiocyanate was obtained from Sigma Aldrich.

[0164] Second, the workspace was set up so that the working area within the Biological Safety Cabinet (BSC) space was clean, materials were gathered, aliquots of Methylene blue, Cresyl blue, Alcian blue, and Potassium thiocyanate were stored in a BSC fridge, and equipment was obtained including a 25mL serological pipette tip, power pipette, 2 50mL conical tubes, lOOOpL pipette, 0.22pm PVDF membrane filter, luerlock 30mL syringe, and a vortex mixer.

[0165] Third, the OFS was made. Using the 25mL serological pipette tip and power pipette, 50mL of MBG water (in two transfers) was exactly transferred into an empty 50mL conical tube. All dyes and the potassium cyanate were vortexed before pipetting. AlOOOpL pipette was used to remove 790 L water and dispose of it. A 200 pL pipette was used to add 165pL of Methylene blue to the water. A 200pL pipette was used to add 250pL of Cresyl blue (in two 125pL transfers) to the water and Methylene blue solution. A 200pL pipette was used to add 250pL of Alcian blue (in two 125pL transfers) to the water, Methylene blue, and Cresyl blue solution. A 200pL pipette was used to add 125pL potassium thiocyanate to the water, Methylene blue, Cresyl blue, and Alcian blue solution. The conical tube with the solution of the water, Methylene blue, Cresyl blue, Alcian blue, and potassium thiocyanate was closed and vortexed until the solution was homogenous.

[0166] Fourth, the OFS was filtered. A 30mL syringe was opened and the membrane filter was attached to the nozzle, and then the syringe plunger was removed. Approximately half of the 50mL solution was poured from the conical tube into the syringe and the plunger was replaced. The plunger was pressed down, and the filtered solution was dispensed into a new 50mL conical tube. The filtering process was repeated with the remaining 25mL unfiltered OFS using a new filter.

[0167] Fifth, the OFS was aliquoted. Once the OFS was filtered, 1.5mL glass vials were obtained and transferred into the BSC. The OFS was vortexed. A lOOOpL pipette was used to add 1400pL (1.4mL) of the OFS to each vial (in two 700pL transfers). Each vial was sealed by twisting on the cap. This process was repeated with new vials until there was no more OFS in the 50mL conical tube. The OFS aliquots were stored in the BSC refrigerator until needed.Attorney Docket No. 19178.0007WOU1

[0168] Sixth, the OFS solution was used. For every 2 plates (192 wells), one vial of the OFS solution was retrieved from the refrigerator. 20pL of nanoparticle solution that had been stored in a fridge was freshly vortexed and pipetted into a vial. OFS and nanoparticle solution were vortexed until well-mixed and label vial with “NP” to indicate the nanoparticle has been added.CLAUSES

[0169] The following numbered clauses define further example aspects and features of the present disclosure:

[0170] 1. A method for analyzing a sample comprising:(a) preparing a solution with nanoparticles;(b) mixing the solution with nanoparticles with a sample to create a reaction mixture;(c) applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array;(d) performing fourier-transform infrared (FTIR) spectroscopy on the reaction mixture to produce at least one IR spectra; and(e) analyzing the at least one IR spectra with one or more machine learning models or algorithms.

[0171] 2 The method of clause 1, further comprising after step (b) incubating the reaction mixture at room temperature.

[0172] 3. The method of clause 1 or 2, further comprising after step (c) allowing the reaction mixture to air dry.

[0173] 4. The method of any of clauses 1-3, wherein the nanoparticles are AuNPs.

[0174] 5. The method of any of clauses 1-4, wherein the nanoparticles are nanorods or nanospheres.

[0175] 6. The method of any of clauses 1-4, wherein nanoparticles further comprise nanoshells or magnetic cores.

[0176] 7 The method of any of clauses 1-6, wherein the nanoparticles have a diameter between 30 nm to 250 nm.

[0177] 8. The method of any of clauses 1-6, wherein the nanoparticles have a diameter between 80 and 200nm.

[0178] 9. The method of any of clauses 1-8, wherein the nanoparticles are bare citrate nanospheres.Attorney Docket No. 19178.0007WOU1

[0179] 10. The method of any of clauses 1-8, wherein the nanoparticles have surface chemistries in solution.

[0180] 11. The method of any of clauses 1-8, wherein the nanoparticles are magnetic and functionalized with specific capture molecules such as antibodies, aptamers, or receptor ligands.

[0181] 12. The method of any of clauses 1-8, wherein the nanoparticles are gold nanoparticles with a magnetic material.

[0182] 13. The method of any of clauses 1-12, wherein the solution with nanoparticles has a concentration between IxlO3to IxlO6particles per mm2.

[0183] 14. The method of any of clauses 1-13, wherein the reaction mixture further includes a molecular dye.

[0184] 15. The method of any of clauses 1-14, wherein the reaction mixture further includes a molecular dye selected from H&E, cresyl violet, brilliant cresyl blue, coomassie brilliant blue, bromophenol blue, trypan blue, toluidine blue, or aniline blue.

[0185] 16. The method of any of clauses 1-15, wherein the one or more machine learning models or algorithms are trained with an entire infrared transmittance or absorbance spectrum obtained from SEIRA- or rSEIRA-FTIR analyses as the input data.

[0186] 17. The method of any of clauses 1-15, wherein the one or more machine learning models or algorithms are provided the entire spectral range from 4000 to 400 cm-1 or 8000 to 400 cm-1.

[0187] 18. The method of any of clauses 1-17, wherein the one or more machine learning models or algorithms include a multilayer perceptron (MLP), convolutional neural networks (CNNs), gradient boosting methods such as LightGBM and XGBoost, support vector machines (SVMs), or a weighted ensemble of MLP, LightGBM and / or XGBoost model.

[0188] 19. The method of any of clauses 1-18, further comprising a series of data preprocessing steps such as spectral quality control, normalization, feature engineering, and feature selection or importance, applied to the entire input spectrum for each FTIR measurement prior to the step of analyzing the at least one IR spectra with one or more machine learning models or algorithms.

[0189] 20. A method for analyzing a biological sample comprising:(a) preparing an optical molecular fingerprinting solution (OFS);(b) mixing nanoparticles with a biological sample;Attorney Docket No. 19178.0007WOU1(c) combining the OFS with the biological sample with the nanoparticles to create a reaction mixture;(d) applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array;(e) performing fourier-transform infrared (FTIR) spectroscopy on the reaction mixture to produce at least one IR spectra; and(f) analyzing the at least one IR spectra.

[0190] 21. The method of clause 20, wherein the OFS is an aqueous solution comprising about 0.33% methylene blue (MB), about 0.50% brilliant cresyl blue (BCV), and about 0.50% alcian blue (AB) molecular dyes.

[0191] 22. The method of clause 20, wherein the OFS is an aqueous solution comprising 0.33% methylene blue (MB), 0.50% brilliant cresyl blue (BCV), and 0.50% alcian blue (AB) molecular dyes.

[0192] 23. A method for analyzing a biological sample comprising:(a) preparing an optical molecular fingerprinting solution (OFS);(b) combining the OFS with a biological sample to create a reaction mixture(c) applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array that has been precoated with nanoparticles;(d) performing fourier-transform infrared (FTIR) spectroscopy on the reaction mixture to produce at least one IR spectra; and(e) analyzing the at least one IR spectra.

[0193] 24. The method of clause 23, wherein the OFS is an aqueous solution comprising about 0.33% methylene blue (MB), about 0.50% brilliant cresyl blue (BCV), and about 0.50% alcian blue (AB) molecular dyes.

[0194] 25. The method of clause 23, wherein the OFS is an aqueous solution comprising 0.33% methylene blue (MB), 0.50% brilliant cresyl blue (BCV), and 0.50% alcian blue (AB) molecular dyes.

[0195] 26. A method for disease detection comprising:(a) depositing a biological sample onto an infrared (IR) reflective sampling array plate configured for nanoparticle application to produce resonant surface-enhanced infrared absorption (rSEIRA), the plate or the biological sample containing nanoparticles;(b) illuminating the biological sample with IR radiation and collecting the resulting reflected IR spectra;Attorney Docket No. 19178.0007WOU1(c) processing and analyzing the collected IR spectra with a machine learning model trained to identify molecular spectral fingerprints indicative of disease states; and(d) outputting a diagnostic result correlating to the presence or absence of disease.

[0196] 27. The method of clause 26, wherein the disease being detected comprises: a cancer, including but not limited to breast cancer, lung cancer, colorectal cancer, prostate cancer, or other cancer types; a neurodegenerative disease; or a pre-malignant condition.

[0197] 28. The method of clause 26 or clause 27, wherein the machine learning model comprises advanced computational models, including convolutional neural networks (CNNs), deep learning frameworks, gradient boosting methods (LightGBM, XGBoost), support vector machines (SVMs), or combinations thereof, trained on extensive and diverse patient-derived spectral datasets.

[0198] 29. The method of any of clauses 26-28, further comprising integrating of spectral data of biomarkers, including circulating RNA, protein signatures, or other metabolites to enhance diagnostic accuracy.

[0199] 30. The method of any of clauses 26-29, wherein the nanoparticles are nanoparticle arrays which comprise gold, silver, or composite nanostructures specifically engineered and optimized to yield maximum electromagnetic field enhancement and spectral amplification at disease-relevant IR frequencies.

[0200] 31. A system for detecting early-stage diseases comprising:(a) an IR reflective sampling plate engineered with one or more nanoparticle arrays that are either pre-coated onto a surface of the plate or provided in a solution mixture with a sample or an analyte, wherein the plate is configured to produce resonant surface-enhanced infrared absorption (rSEIRA);(b) an IR illumination source capable of emitting radiation across a relevant IR spectral range and a detector configured to capture reflected, enhanced IR spectral data;(c) a processing unit operatively linked to the detector, configured to execute machine learning models trained to analyze IR spectra and accurately predict the likelihood of early- stage disease presence based on molecular fingerprint analysis; and(d) a user interface configured to display information.

[0201] 32. The system of clause 31, wherein the IR reflective sampling plate is specifically tailored for the detection of early-stage cancers, particularly breast cancer, andAttorney Docket No. 19178.0007WOU1 adaptable to other disease conditions by retraining the machine learning models with corresponding disease-specific datasets or reoptimizing the nanoparticle conditions.

[0202] 33. The system of clause 31 or clause 32, wherein the processing unit employs advanced pattern recognition, spectral feature extraction, and predictive modeling algorithms to reliably identify and quantify subtle molecular alterations within biological samples, thereby facilitating predictive diagnostics, preventive interventions, and ongoing disease monitoring.

[0203] 34. The system of any of clauses 31-33, further comprising automated or semi-automated sample preparation components designed to streamline sample handling, deposition, and drying processes, ensuring reproducibility and ease of integration into routine clinical workflows.

[0204] Having described the preferred aspects and implementations of the present disclosure, modifications and equivalents of the disclosed concepts may readily occur to one skilled in the art. However, it is intended that such modifications and equivalents be included within the scope of the claims which are appended hereto.

Claims

Attorney Docket No. 19178.0007WOU1What is claimed is:

1. A method for analyzing a sample comprising:(a) preparing a solution with nanoparticles;(b) mixing the solution with nanoparticles with a sample to create a reaction mixture;(c) applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array;(d) performing fourier-transform infrared (FTIR) spectroscopy on the reaction mixture to produce at least one IR spectra; and(e) analyzing the at least one IR spectra with one or more machine learning models or algorithms.

2. The method of claim 1, further comprising after step (b) incubating the reaction mixture at room temperature.

3. The method of claim 1, further comprising after step (c) allowing the reaction mixture to air dry.

4. The method of claim 1, wherein the nanoparticles are AuNPs.

5. The method of claim 1, wherein the nanoparticles are nanorods or nanospheres.

6. The method of claim 1, wherein nanoparticles further comprise nanoshells or magnetic cores.

7. The method of claim 1, wherein the nanoparticles have a diameter between 30 nm to 250 nm.

8. The method of claim 1, wherein the nanoparticles have a diameter between 80 and 200nm.

9. The method of claim 1, wherein the nanoparticles are bare citrate nanospheres.Attorney Docket No. 19178.0007WOU110. The method of claim 1, wherein the nanoparticles have surface chemistries in solution.

11. The method of claim 1, wherein the nanoparticles are magnetic and functionalized with specific capture molecules such as antibodies, aptamers, or receptor ligands.

12. The method of claim 1, wherein the nanoparticles are gold nanoparticles with a magnetic material.

13. The method of claim 1, wherein the solution with nanoparticles has a concentration between IxlO3to IxlO6particles per mm2.

14. The method of claim 1, wherein the reaction mixture further includes a molecular dye.

15. The method of claim 1, wherein the reaction mixture further includes a molecular dye selected from H&E, cresyl violet, brilliant cresyl blue, coomassie brilliant blue, bromophenol blue, trypan blue, toluidine blue, or aniline blue.

16. The method of claim 1, wherein the one or more machine learning models or algorithms are trained with an entire infrared transmittance or absorbance spectrum obtained from SEIRA- or rSEIRA-FTIR analyses as the input data.

17. The method of claim 1, wherein the one or more machine learning models or algorithms are provided the entire spectral range from 4000 to 400 cm1or 8000 to 400 cm18. The method of claim 1, wherein the one or more machine learning models or algorithms include a multilayer perceptron (MLP), convolutional neural networks (CNNs), gradient boosting methods such as LightGBM and XGBoost, support vector machines (SVMs), or a weighted ensemble of MLP, LightGBM and / or XGBoost model.

19. The method of claim 1, further comprising a series of data preprocessing steps such as spectral quality control, normalization, feature engineering, and feature selection or importance, applied to the entire input spectrum for each FTIR measurement prior to the stepAttorney Docket No. 19178.0007WOU1 of analyzing the at least one IR spectra with one or more machine learning models or algorithms.

20. A method for analyzing a biological sample comprising:(a) preparing an optical molecular fingerprinting solution (OFS);(b) mixing nanoparticles with a biological sample;(c) combining the OFS with the biological sample with the nanoparticles to create a reaction mixture(d) applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array;(e) performing fourier-transform infrared (FTIR) spectroscopy on the reaction mixture to produce at least one IR spectra; and(f) analyzing the at least one IR spectra.

21. The method of claim 20, wherein the OFS is an aqueous solution comprising about 0.33% methylene blue (MB), about 0.50% brilliant cresyl blue (BCV), and about 0.50% alcian blue (AB) molecular dyes.

22. The method of claim 20, wherein the OFS is an aqueous solution comprising 0.33% methylene blue (MB), 0.50% brilliant cresyl blue (BCV), and 0.50% alcian blue (AB) molecular dyes.

23. A method for analyzing a biological sample comprising:(a) preparing an optical molecular fingerprinting solution (OFS);(b) combining the OFS with a biological sample to create a reaction mixture(c) applying the reaction mixture onto a surface of a curvature-enhanced infrared reflective (OIR) array that has been precoated with nanoparticles;(d) performing fourier-transform infrared (FTIR) spectroscopy on the reaction mixture to produce at least one IR spectra; and(e) analyzing the at least one IR spectra.Attorney Docket No. 19178.0007WOU124. The method of claim 23, wherein the OFS is an aqueous solution comprising about 0.33% methylene blue (MB), about 0.50% brilliant cresyl blue (BCV), and about 0.50% alcian blue (AB) molecular dyes.

25. The method of claim 23, wherein the OFS is an aqueous solution comprising 0.33% methylene blue (MB), 0.50% brilliant cresyl blue (BCV), and 0.50% alcian blue (AB) molecular dyes.

26. A method for disease detection comprising:(a) depositing a biological sample onto an infrared (IR) reflective sampling array plate configured for nanoparticle application to produce resonant surface-enhanced infrared absorption (rSEIRA), the plate or the biological sample containing nanoparticles;(b) illuminating the biological sample with IR radiation and collecting the resulting reflected IR spectra;(c) processing and analyzing the collected IR spectra with a machine learning model trained to identify molecular spectral fingerprints indicative of disease states; and(d) outputting a diagnostic result correlating to the presence or absence of disease.

27. The method of claim 26, wherein the disease being detected comprises: a cancer, including but not limited to breast cancer, lung cancer, colorectal cancer, prostate cancer, or other cancer types; a neurodegenerative disease; or a pre-malignant condition.

28. The method of claim 26 or claim 27, wherein the machine learning model comprises advanced computational models, including convolutional neural networks (CNNs), deep learning frameworks, gradient boosting methods (LightGBM, XGBoost), support vector machines (SVMs), or combinations thereof, trained on extensive and diverse patient-derived spectral datasets.

29. The method of claim 26, further comprising integrating of spectral data of biomarkers, including circulating RNA, protein signatures, or other metabolites to enhance diagnostic accuracy.Attorney Docket No. 19178.0007WOU130. The method of claim 26, wherein the nanoparticles are nanoparticle arrays which comprise gold, silver, or composite nanostructures specifically engineered and optimized to yield maximum electromagnetic field enhancement and spectral amplification at diseaserelevant IR frequencies.

31. A system for detecting early-stage diseases comprising:(a) an IR reflective sampling plate engineered with one or more nanoparticle arrays that are either pre-coated onto a surface of the plate or provided in a solution mixture with a sample or an analyte, wherein the plate is configured to produce resonant surface-enhanced infrared absorption (rSEIRA);(b) an IR illumination source capable of emitting radiation across a relevant IR spectral range and a detector configured to capture reflected, enhanced IR spectral data;(c) a processing unit operatively linked to the detector, configured to execute machine learning models trained to analyze IR spectra and accurately predict the likelihood of early- stage disease presence based on molecular fingerprint analysis; and(d) a user interface configured to display information.

32. The system of claim 31, wherein the IR reflective sampling plate is specifically tailored for the detection of early-stage cancers, particularly breast cancer, and adaptable to other disease conditions by retraining the machine learning models with corresponding diseasespecific datasets or reoptimizing the nanoparticle conditions.

33. The system of claim 31 or claim 32, wherein the processing unit employs advanced pattern recognition, spectral feature extraction, and predictive modeling algorithms to reliably identify and quantify subtle molecular alterations within biological samples, thereby facilitating predictive diagnostics, preventive interventions, and ongoing disease monitoring.

34. The system of claim 31, further comprising automated or semi-automated sample preparation components designed to streamline sample handling, deposition, and drying processes, ensuring reproducibility and ease of integration into routine clinical workflows.

Images