Nanoengineered immunosensors with unsupervised clustering for multiple circulating biomarkers

Abstract

An immunosensor device is disclosed for rapid detection of circulating biomarkers associated with thrombosis, including C-reactive protein (CRP), calprotectin, soluble platelet selectin (sP-selectin), and D-dimer. The immunosensors were fabricated using fiber-laser engraving of carbon nanotubes and CO2-laser cutting of microfluidic channels and features, along with the electrochemical deposition of gold nanoparticles to conjugate with biomarker-specific aptamers and antibody. Using unsupervised clustering based on 4-biomarker concentrations, thrombotic events were predicted in 49 of 53 patients. The 4-biomarker combination yielded an area under the receiver operating characteristic curve (AUC) of 0.95, demonstrating high sensitivity and specificity for acute thrombosis prediction compared to the AUC values for individual biomarker: CRP (0.773), calprotectin (0.711), sP-selectin (0.683), and D-dimer (0.739). The immunosensor device with unsupervised clustering provides accurate and efficient methods for predicting thrombosis, guiding personalized medicine.

Description

2025-015-2NANOENGINEERED IMMUNOSENSORS WITH UNSUPERVISED CLUSTERING FOR MULTIPLE CIRCULATING BIOMARKERSRelated Application

[0001] This Application claims priority to U. S. Provisional Patent Application No.63 / 725,516 filed on November 26, 2024, which is hereby incorporated by reference in its entirety. Priority is claimed pursuant to 35 U. S. C. § 119 and any other applicable statute.Technical Field

[0002] The technical field generally relates a nanoengineered immunosensor that uses machine learning to detect multiple biomarkers for the rapid prediction of a diseased state or condition. In one embodiment, the immunosensor rapidly detects circulating biomarkers associated with thrombosis, including C-reactive protein (CRP), calprotectin, soluble platelet selectin (sP-selectin), and D-dimer.Statement Regarding Federally SponsoredResearch and Development

[0003] This invention was made with government support under BX004356 awarded by the U. S. Department of Veterans Affairs, and HL159970, HL118650, HL149808, and HL129727 awarded by the National Institutes of Health. The government has certain rights in the invention.Reference to an Electronic Sequence Listing

[0004] The contents of the electronic sequence listing (2025_015_2_seqlisting.xml: Size: 4 kilobytes; and Date of Creation: 11 / 2 / 2025) is herein incorporated by reference in its entiret.Background

[0005] Acute viral infections are emerging as increasingly complex and prevalent threats to public health worldwide. While the incidence of the illness known as the 2019 coronavirus disease (COVID- 19) is declining, a growing association has been reported between respiratory viral infections and clinical thrombotic events. A prothrombotic state predisposes patients to acute coronary syndromes, cerebrovascular accidents, and pulmonary embolism.2025-015-2The presence of COVID-19-associated thromboses raises concern for worse patient recovery¬ rates. Reports of these thrombotic events underscore the need for an accurate and rapid prediction tool that can be used to determine appropriate anticoagulation prophylaxis and to prepare for the next wave of the SARS-CoV-2 pandemic.

[0006] In response to the COVID-19-associated thrombosis crisis, the International COVID-19 Thrombosis Biomarkers Colloquium formulated a panel of recommended biomarkers that can be used to predict the risk of developing thrombosis associated with COVID- 19. This formulated panel is based on retrospective studies of several thousand patients with COVID-19, which have revealed the presence of elevated blood levels of C-reactive protein (CRP), calprotectin, soluble platelet selectin (sP-selectin), and D-dimer in these patients. However, relying on a single biomarker often falls short in terms of sensitivity and specificity for predicting thrombotic events. Furthermore, a primary obstacle in predicting thrombosis is the complex interplay of multiple variables, such as viral strain, virulence, individual health conditions, genetic predisposition, and social determinants of health. A comprehensive analytical approach that utilizes multiple biomarkers simultaneously may improve the accuracy of predicting thrombotic risk during acute illnesses.Summary

[0007] A nanoengineered immunosensor device is disclosed for the simultaneous detection of four (4) key biomarkers, using unsupervised clustering to analyze variations in biomarker concentrations. This approach enhances thrombosis prediction by capturing the nuanced interplay of these biomarkers. In detection, the signal-to-noise ratio of the electrodes is determined by the specific binding affinity- of aptamers or antibodies to the biomarkers. Aptamers and antibodies are immobilized on electrochemically deposited gold nanoparticles (AuNPs) based on fiber laser-engraved carbon nanotubes (CNTs) electrodes, enabling sensitive and specific detection across multiple biomarkers, in contrast to traditional singlebiomarker detection methods like the enzyme-linked immunosorbent assay (ELISA) method that require longer detection times and process optimizations. Ultimately, the customized and scalable approach is poised to improve the healthcare team responses to thrombotic events in future pandemics.

[0008] Following biomarker detection and quantification, a machine-learning technique called unsupervised clustering is applied to the biomarker data to identify inherent patterns within the unlabeled data. This approach allows the data points to be organized into distinct2025-015-2clusters, and the unsupervised clustering algorithms effectively stratified patients based on their thrombotic risk from various illnesses. This stratification is grounded in objective and quantifiable biomarker data, providing a robust framework for the personalized assessment of risk profiles to prevent thrombotic complications, including stroke, pulmonary embolism, and acute coronary syndrome, in high-risk patients.

[0009] In one embodiment, an immunosensor device for determining or predicting a disease state or condition within a subject is disclosed that uses a sample from the subject (e.g., blood or blood plasma). The immunosensor device includes a chip including a substrate having disposed thereon a plurality of working electrodes each formed with carbon nanotubes and having gold nanoparticles deposited thereon and further conjugated to aptamers or antibodies specific to a different biomarker of a plurality of different biomarkers. The chip further includes a reference electrode and a counter electrode. A detection chamber is formed over the plurality of working electrodes. A washing chamber is disposed over the detection chamber, wherein the washing chamber fluidically communicates with the detection chamber.

[0010] The immunosensor device further includes an impedance measuring device that is configured to measure impedance changes at the plurality of working electrodes in response to exposure to the sample and a solution of [Fe(CN)6]3+ / 4+or other redox couple exposed to the working electrodes. A computing device is also provided having software executed thereby that performs unsupervised clustering of concentration data for the plurality of different biomarkers based on the measured impedance changes and outputs a determination or prediction of the disease state or condition. This may include acute thrombosis in one embodiment but it may also include other disease states or conditions.

[0011] In one embodiment, an immunosensor device for determining or predicting a disease state or condition of a subject using a sample from the subject includes a chip including a substrate having disposed thereon a plurality of working electrodes each formed with carbon nanotubes and having gold nanoparticles deposited thereon and further conjugated to aptamers or antibodies specific a different biomarker of a plurality of different biomarkers, a reference electrode, and a counter electrode. A detection chamber is formed over the plurality of working electrodes and a washing chamber is disposed over the detection chamber, wherein the washing chamber fluidically communicates with the detection chamber.2025-015-2

[0012] The immunosensor device may include an impedance measuring device that is configured to measure impedance changes at the plurality of working electrodes in response to exposure to the sample and a solution of [Fe(CN)6]3+ / 4+ or other redox couple exposed to the working electrodes. In one embodiment, the impedance measuring device includes an electrical workstation or dedicated impedance measuring device configured to perform cyclic voltammetry and / or electrochemical impedance spectroscopy.

[0013] The immunosensor device may further include a computing device having software executed thereby that performs unsupervised clustering of measured impedance changes or concentration data for the plurality of different biomarkers and outputs a determination or prediction of the disease state or condition based on the unsupervised clustering. In one embodiment, the measured impedance changes are converted to biomarker concentrations prior to unsupervised clustering. In another embodiment, the measured impedance changes are converted to multi-bit barcode corresponding to biomarker concentrations prior to unsupervised clustering.

[0014] The diseased state or condition that is determined by the immunosensor device may include acute thrombosis and the aptamers or antibodies are specific to C-reactive protein (CRP), calprotectin, soluble platelet selectin (sP-selectin), and D-dimer. The sample may include blood plasma or other biological samples.

[0015] In one embodiment, the detection chamber and washing chamber are formed in a plurality of layers stacked on the substrate of the chip. A top layer has an inlet that fluidically communicates through the washing chamber and the detection chamber to the plurality of working electrodes.

[0016] In another embodiment, a method of using the immunosensor device includes washing the working electrodes of the immunosensor device with a buffer solution; loading a sample into an inlet in the chip and incubating the sample with the immunosensor device for a period of time; washing residual sample from the chip; measuring impedance changes at the plurality of working electrodes with an impedance measuring device in response to exposure to the sample and a solution of [Fe(CN)6]3+ / 4+or other redox couple exposed to the working electrodes; and performing unsupervised clustering of the measured impedance changes or concentration data for the plurality of different biomarkers using a computing device having software executed thereon and outputting a determination or prediction of the disease state or condition based on the unsupervised clustering.2025-015-2Brief Description of the Drawings

[0017] FIGS. 1A-1E illustrate nanoengineered immunosensors detected multiple biomarkers and utilized unsupervised clustering for rapid and accurate prediction of acute blood clots. FIG. 1A shows respiratory viruses (e.g., SARS-CoV-2) infected human lung epithelial cells and bound to a specific receptor (e.g., ACE2). FIG. IB illustrates invasion by the respiratory virus led to abnormal fluctuations in the concentrations of specific biomarkers in blood circulation due to inflammatory and thrombotic processes, including CRP, calprotectin, sP-selectin, and D-dimer. FIG. 1C illustrates the operations and components for the fabrication of a nanoengineered immunosensor device using fiber and CO2laser technologies. FIG. ID illustrates conjugation of aptamers and antibody to the CNTs electrodes enabled the rapid electrochemical detection of targeted biomarkers (BSA: bovine serum albumin). FIG. IE shows the unsupervised clustering analysis stratified 53 patients into distinct levels of thrombotic risk, using the concentrations of 4 distinct biomarkers as clustering features. The clustering results were validated with the ICD-10 diagnostic code for thrombosis provided by the UCLA Biobank (PBBS) Core.

[0018] FIGS. 2A-2F illustrate the fabrication and characterization of nanoengineered immunosensor devices. FIG. 2A is a schematic representation of a multilaminated (10 layers) immunosensor device. FIG. 2B shows the modification of the CNT electrodes w ith Au NPs, Ag / AgCl NPs, and Pt NPs, followed by the conjugation of aptamers and antibody. FIG. 2C illustrates utilization of surface plasmon resonance (SPR) techniques to study the association, dissociation, and regeneration of targeted biomarkers on the aptamer channel. FIG. 2D show s transmission electron microscopy (TEM) images and energy-dispersive X-ray (EDX) elemental mapping of the Au NPs working electrode, the Ag / AgCl NPs reference electrode, and the Pt NPs counter electrode. Scale bar: 100 nm. FIG. 2E shows X-ray photoelectron spectroscopy (XPS) spectra of the modified electrodes. FIG. 2F illustrates comparative responses of control and aptamer channels to targeted biomarkers using SPR techniques. A linear relationship betw een different dissociation times, response units, and biomarker concentrations was established based on the calibrated response units via the control and aptamer channels.

[0019] FIGS. 3A-3F show the performance of the nanoengineered immunosensor device was evaluated. FIG. 3A illustrates cyclic voltammetry (CV) testing compared current density' with (gray curve) and without (light gray curve) the targeted biomarkers. FIG. 3B shows electrochemical impedance measurements of the targeted biomarker at various2025-015-2concentrations. FIG. 3C illustrates the linear relationship between charge-transfer resistance and four (4) biomarker concentrations. FIG. 3D illustrates the independent measurements demonstrating the reproducibility of the individual immunosensor (n = 6). FIG. 3E illustrates the specificity towards targeted biomarkers in the presence of interferents (n = 3). FIG. 3F shows storage stability was established over 14 days. Error bars represent the standard deviation from repeated measurements (Greenspon, #2) unless otherwise specified.

[0020] FIGS. 4A-4F illustrate unsupervised clustering was used for thrombosis prediction with 53 blood specimens from COVID-19 patients. FIG. 4A is a chord diagram illustrates the network relationship among the four (4) biomarkers and 53 patients. FIG. 4B is hierarchical clustering analysis utilized the concentrations of the 4 biomarkers to assess thrombotic risk in the 53 blood specimens. FIG. 4C shows hierarchical clustering outcomes were validated using the ICD-10 diagnostic code for thrombosis. FIG. 4D illustrates PC A analysis was used to cluster the 53 blood specimens. Each arrow represents the influence of a different biomarker concentration in the PCA space. FIG. 4E illustrates t-SNE analysis converted similarities between data points into joint probabilities, facilitating the visualization of patient clusters. FIG. 4F shows UMAP analysis preserved both local and global structures, enabling the effective visualization of specimen clusters that are positive or negative for thrombosis.

[0021] FIGS. 5A-5U illustrate statistical analysis of the concentrations of four (4) biomarkers was performed on 53 patient blood specimens. FIGS. 5A-5 include violin plots compare the concentrations of four (4) biomarkers (FIG. 5 A: CRP; FIG. 5B: Calprotectin; FIG. 5C: sP-selectin; FIG. 5D: D-dimer) between thrombosis-negative (-) and thrombosispositive (+) patients. (E) Proximity matrix among the four (4) biomarkers. FIG. 5F shows the proximity matrix among the 53 patients. FIGS. 5G-5J show receiver operating characteristic (ROC) curve analysis illustrates the relationship between the concentration of a single biomarker (FIG. 5G: CRP; FIG. 5H: Calprotectin; FIG. 51: sP-selectin; FIG. 5J: D-dimer) and thrombosis prediction, using the ICD-10 code for validation. FIG. 5K shows ROC curve analysis using the combination of 4 biomarkers (CRP, Calprotectin. sP-selectin, and D-dimer) for thrombosis prediction. FIGS. 5L-5Q show ROC curve analysis using the combinations of two (2) biomarkers for thrombosis prediction (FIG. 5L: CRP and Calprotectin; FIG. 5M: CRP and sP-selectin; FIG. 5N: CRP and D-dimer; FIG. 50: Calprotectin and sP-selectin; FIG. 5P: Calprotectin and D-dimer; FIG. 5Q: sP-selectin and D-dimer). FIGS. 5R-5U illustrate ROC curve analysis using the combinations of three (3) biomarkers for thrombosis prediction (FIG. 5R: CRP. Calprotectin. and sP-selectin; FIG. 5S: CRP, Calprotectin, and D-2025-015-2dimer; FIG. 5T: CRP, sP-selectin, and D-dimer; FIG. 5U: Calprotectin, sP-selectin, and D-dimer).

[0022] FIGS. 6A-6C illustrate the detection of CRP in human plasma samples was compared using the fabricated immunosensor and ELISA. FIG. 6A shows CRP detection utilizing the fabricated immunosensor, based on three independent experiments (n = 3). FIG.6B shows CRP detection via ELISA, based on three independent experiments (n = 3). FIG.6C is an illustration of the linear relationship between results obtained from the fabricated immunosensor and those from ELISA. The Pearson correlation coefficient (7?) was determined using linear regression analysis.

[0023] FIGS. 7A-7C show the detection of calprotectin in human plasma samples was compared using the fabricated immunosensor and ELISA. FIG. 7A shows calprotectin detection utilizing the fabricated immunosensor, based on three independent experiments (n = 3). FIG. 7B shows calprotectin detection via ELISA, based on three independent experiments (n = 3). FIG. 7C is an illustration of the linear relationship between results obtained from the fabricated immunosensor and those from ELISA. The Pearson correlation coefficient ( ) was determined using linear regression analysis.

[0024] FIGS. 8A-8C show the detection of sP-selectin in human plasma samples was compared using the fabricated immunosensor and ELISA. FIG. 8A shows sP-selectin detection utilizing the fabricated immunosensor, based on three independent experiments (n = 3). FIG. 8B shows sP-selectin detection via ELISA, based on three independent experiments (n = 3). FIG. 8C is an illustration of the linear relationship betw een results obtained from the fabricated immunosensor and those from ELISA. The Pearson correlation coefficient (7?) was determined using linear regression analy sis.

[0025] FIGS. 9A-9C illustrate the detection of D-dimer in human plasma samples was compared using the fabricated immunosensor and ELISA. FIG. 9A shows D-dimer detection utilizing the fabricated immunosensor, based on three independent experiments (n = 3). FIG.9B shows D-dimer detection via ELISA, based on three independent experiments (n = 3). FIG. 9C is an illustration of the linear relationship between results obtained from the fabricated immunosensor and those from ELISA. The Pearson correlation coefficient (A) was determined using linear regression analysis.

[0026] FIGS. 10A-10F illustrate the unsupervised clustering performed using the “4-bit barcode’' method for acute thrombosis prediction. FIG. 10A is a chord diagram illustrating the network relationship among the 4 biomarkers and 53 patients. FIG. 1 OB is a hierarchical2025-015-2clustering analysis for acute thrombosis prediction. (C) Hierarchical clustering outcomes validated against the ICD-10 diagnostic code for thrombosis. FIG. 10D illustrates PCA analysis for acute thrombosis prediction, where each arrow represents the influence of a different biomarker concentration in the PCA space. FIG. 10E illustrates t-SNE analysis for predicting acute thrombosis. FIG. 10F is UMAP analysis for predicting acute thrombosis.

[0027] FIG. 11 illustrates a process diagram showing the steps for priming the chip, loading it with plasma, obtaining the signal, and regenerating the chip. The immunosensor was prepared by washing with PBS, loading a diluted plasma sample for 45 minutes of incubation to allow biomarker binding, followed by electrochemical measurements using [Fe(CN)6]3- / 4- solution and EIS, after which the chambers were washed, regenerated with glycine-HCl buffer, and stored at 4°C for future use.

[0028] FIG. 12 illustrates a system for using the immunosensor device as disclosed herein. The immunosensor device is illustrated along with the impedance measuring device, computing device with machine learning (ML) software that outputs a determination or prediction of a disease state or condition.Detailed Description of Illustrated Embodiments

[0029] In one embodiment, an immunosensor device 10 (FIGS. 1C, ID, 2A, 2B, 11, 12) is disclosed for determining or predicting a disease state or condition within a subject 100 (e.g., human or mammalian subject) using a sample 102 (FIG. 11) such as blood plasma from the subject 100. The immunosensor device 10 may be formed as a chip 12 that includes a substrate 14 having disposed thereon a plurality of working electrodes 16. In one embodiment, there are four (4) such working electrodes 16 as illustrated in FIG. 1C but other embodiments may have more or less. Each working electrode 16 is located on the substrate 14 of the chip 12 and is formed from carbon nanotube 18 having metallic nanoparticles 20 deposited thereon. The metallic nanoparticles 20 are further conjugated a capture agents 22 specific a different biomarker 104 of a plurality of different biomarkers 104. In one preferred aspect, the metallic nanoparticles 20 are gold nanoparticles. Other metals like silver or platinum may also be used for the nanoparticles 20.

[0030] The capture agents 22 may include to aptamers or antibodies. In one specific embodiment, the diseased state or condition is acute thrombosis and the capture agents 22 are specific to C-reactive protein (CRP), calprotectin, soluble platelet selectin (sP-selectin), and D-dimer. The chip 12 includes additional electrodes including a reference electrode 24 and a2025-015-2counter electrode 26 which are used for taking impedance measurements at the various working electrodes 16. With reference to FIGS. 2A and 12, a detection chamber 28 is formed over the plurality of working electrodes 16 and a washing chamber 30 is disposed over the detection chamber 28, wherein the washing chamber 30 fluidically communicates with the detection chamber 28 via holes or apertures 32 that separate the respective chambers 28, 30 (FIG. 1C). In one embodiment, the chip 12 is formed from various layers that are stacked to form the detection chamber 28, washing chamber 30, as well is inlets / outlets that provide access for the sample and removal of wash or other solutions. A top layer (FIG. 1C) is provided with an inlet 34 that can accommodate the sample 102 or other reagent(s) that fluidically communicates through the washing chamber 30 and the detection chamber 28 to the plurality of working electrodes 16. The various fluids may thus be loaded through this inlet 34 in the top layer.

[0031] The immunosensor device 10 includes an impedance measuring device 40 (FIG.12) that is configured to measure impedance changes at the plurality of working electrodes 16 of the chip 12 in response to exposure to the sample 102 and a solution of [Fe(CN)6]3+ / 4+(or other redox couple) exposed to the working electrodes 16. The [Fe(CN)6]3+ / 4+solution is used to measure impedance changes at the working electrodes 16. In particular, the recognition components of capture agents 22 on the conductive surfaces of the gold nanoparticles 20 changed the charge-transfer resistance ( / Gt) at the working electrode 16 interface, enabling electrical impedance techniques to be highly effective in assessing the degree of binding affinity of biomarkers 104 to the capture agents 22. Distinct impedance signals / measurements are made at each working electrode 16 which, as explained herein, are then used to determine the concentration or quantity of the biomarkers 104 in the sample 102 which are then used to predict or diagnose a disease state or condition of the subject 100. The impedance measuring device 40 may be a specialized or dedicated device for measuring impedance or it may be an electrochemical workstation. The concentration or quantity of biomarkers in the sample 102 may be determined through the charge-transfer resistance ( / Gt ) at the interface of the working electrode 16. FIG. 3C illustrates how there is a linear relationship between Ra and the concentration of biomarkers 104.

[0032] With reference to FIG. 12, the immunosensor device 10 further includes a computing device 50 having software 52 executed thereby that performs machine learning, namely unsupervised clustering of concentration data for the plurality of different biomarkers 104 based on the measured impedance changes detected with the impedance measuring2025-015-2device 10 and outputs a determination or prediction of the disease state or condition (FIG. 12). Specifically, knowing what cluster the concentration data (or the measured impedance changes) falls into may be used to output a determination or prediction of the disease state or condition. The measured impedance changes obtained with the impedance measuring device 40 may be converted to biomarker concentrations prior to unsupervised clustering.Alternatively, the measured impedance changes are converted to multi-bit barcode corresponding to biomarker concentrations prior to unsupervised clustering. The output that is generated by the software 52 may include a qualitative output such as ‘'positive” or “negative.” In other embodiments, the output that is generated may, alternatively or in addition to. include a quantitative output such as a percentage, range of percentage, a concentration or concentration range of the biomarker 104.

[0033] To use the immunosensor device 10, it is first washed with a buffer solution. The buffer solution is input to inlet 34. Next, as seen in FIG. 11, a sample 102 such as blood plasma (or other biological sample) is located into the inlet 34 and the sample 102 incubates within the immunosensor device 10 for a period of time (e.g., 45 minutes to about 1 hour). This is followed by washing residual sample 102 from the chip 12. Impedance changes at the plurality of working electrodes 16 are then measured with the impedance measuring device 40 in response to exposure to the sample and a solution of [Fe(CN)6]3+ / 4+exposed to the working electrodes 16 also added at the inlet 34 to the chip 12. After measuring the impedance changes or signals, unsupervised clustering of the concentration data (impedance changes / signals) is performed by software 52 executed on the computing device 50 for the plurality of different biomarkers 104 based on the measured impedance changes (dimensionality reduction may also be performed as seen in FIG. IE). As a result of the unsupervised clustering a determination or prediction of the disease state or condition of the subject is output using the software 52 executed by the computing device 50.

[0034] As noted herein, in one specific embodiment, the diseased state or condition includes acute thrombosis and the capture agents 22 specific to C-reactive protein (CRP), calprotectin, soluble platelet selectin (sP-selectin), and D-dimer. The immunosensor device 10 may be reusable in some embodiments by washing or exposing the immunosensor device 10 to glycine-HCl buffer as seen in FIG. 11.

[0035] While the plurality of w orking electrodes 16 in this embodiment includes four (4) working electrodes 16 it should be understood that the immunosensor device 10 may include two (2) working electrodes 16, three (3) working electrodes 16. or more than four (4)2025-015-2working electrodes 16. An advantage of the immunosensor device 10 is that the determination or prediction of the disease state or condition is output more quickly than other assays, for example, within about 1.5 hours or less.

[0036] Results

[0037] Nanoengineered immunosensor devices with unsupervised clustering to enhance thrombosis prediction

[0038] Herein, an immunosensor device 10 is used to rapidly detect circulating biomarkers 104 in a sample 102 such as human plasma (FIG. 1C). A desire was to combine the abnormal fluctuations in the concentrations of four (4) biomarkers 104 associated with thrombosis in response to COVID-19 infection: CRP, calprotectin, sP-selectin, and D-dimer (FIGS. 1A and IB). These biomarkers 104 were measured in de-identified blood samples collected from COVID- 19 patients admitted to the medical center. Nanoengineered immunosensor devices 10 were fabricated using fiber and CO2laser technologies (FIG. 1C). The electrochemically deposited CNTs 18 on the working electrodes 16 were conjugated with capture agents 22 (i.e., aptamers and antibody) specific to the biomarkers 104 (FIG. ID). The concentration of these biomarkers 104 was detected and used as features of unsupervised clustering to predict the risk for acute blood clots. Subsequently, different unsupervised clustering methods were compared for classifying thrombotic risk from the blood specimens (FIG. IE). The predicted results were validated against the ICD-10 code for thrombosis provided by the UCLA Pathology Biobanks and Biospecimen Research Core.

[0039] To fabricate the nanoengineered immunosensor device 10, a polyvinyl chloride (PVC) substrate 14 was converted into an electrically conductive substrate (FIG. 1C) by spraying with acid-treated CNTs. Then, a fiber laser (1064 nm, 50W) was used to pattern a pre-designed mask onto the CNTs substrate to form the multiplexed electrodes 16. The detection and wash chambers 28, 30 were engraved with a CO2laser (10600 nm, 50W). To achieve electrochemical detection of the thrombosis biomarkers (FIG. ID), the CNTs electrodes 16 were electrochemically deposited with AuNPs 20 and conjugated with capture agents 22 in the form of aptamers against CRP, sP-selectin, and D-dimer via an Au-S bond (26-28). The capture agent 22 against calprotectin was an antibody conjugated to the acid-treated CNTs working electrode 16 using N-(3-dimethylaminopropyl)-N'-ethylcarbodiimide (EDC)ZN-hydroxysuccinimide (NHS)-based chemistry. The recognition components of aptamers and antibody on the conductive surfaces of the working electrodes 16 changed the charge-transfer resistance (Rct) at the electrode interface, enabling electrical impedance2025-015-2measurements via the impedance measuring device 40 to be highly effective in assessing the degree of binding affinity.

[0040] To develop the thrombosis prediction model using the acquired data, patient blood sample 102 (m) and biomarkers 104 (n) were used to construct an m x n computing matrix (FIG. IE). Dimensionality reduction and unsupervised clustering algorithms were used to facilitate the rapid and accurate prediction of thrombotic risk. In this representation, each dot corresponds to an individual patient. The concentration of the biomarkers 104 determines the positioning of each dot on the two-dimensional plot. The spatial distance between these plotted points forms the basis of the clustering analysis, which, in turn, dictates the final risk classifications assigned to each patient sample 102. This classification enables a visually intuitive and quantitatively robust method to assess the thrombotic risk profile of patients based on the concentration of specific biomarkers 104. It was validated by comparing it with the ICD-10 diagnostic codes for thrombosis, as provided by the UCLA PBBR Core.

[0041] Rapid detection of biomarkers using nanoengineered immunosensor devices

[0042] Given the nature of thrombotic onset in acute infections, rapid detection of a combination of four (4) biomarkers 104 is clinically important. The four (4) biomarkers 104 were detected within 1 hour by employing an electrochemical immunosensor device 10, utilizing a 5 mM [Fe(CN)6]3+ / 4+solution to capture the distinct impedance signals among various concentration levels of biomarkers 104. The CO2laser-engraved washing chamber 30 was designed to eliminate residual elements from plasma samples 102. The regeneration potential of the immunosensor device 10 was verified using the surface plasmon resonance (SPR) technique (FIGS. 2C and 2F), whereby a 0.1 M glycine-HCl buffer (pH = 2) as seen in FIG. 11 removed adherent proteins and effectively regenerated the immunosensor device 10. To augment the detection range, CNTs 18 were employed as the foundational working electrodes 16 for their large and specific surface area, enabling conjugation with the capture agents 22 on the electrochemically deposited Au NPs 20 (FIGS. 2B and 2D). High-resolution transmission electron microscopy (TEM) was applied to image the working electrodes 16 (Au NPs), reference electrodes 24 (Ag / AgCl NPs), and counter electrodes 26 (Pt NPs) (FIG. 2D). Energy-dispersive X-ray (EDX) elemental mapping confirmed the effective integration of Au, Pt, Ag, and Cl elements (FIG. 2D).

[0043] X-ray photoelectron spectroscopy (XPS) was further employed to elucidate the valence states of the fabricated electrodes 16 (FIG. 2E). The Au 4f spectrum revealed a singular peak pair (Au 4f5 / 2 / 4f7 / 2) at 83.6 and 87.2 eV. indicating the Au° state. In the Pt 4f2025-015-2spectrum, two peak pairs were observed (Pt 4f5 / 2 / 4f7 / 2) at 74.5 and 71.0 eV, and 77.4 and 71.8 eV. corresponding to the Pt° and Pt2+states, indicating strong binding of Pt NPs to the supporting CNTs electrode 16. The Ag 3d spectrum exhibited two peak pairs (Ag 3d3 / 2 / 3d5 / 2) at 373.2 and 367.2 eV, and 367.7 and 373.6 eV, indicating the presence of Ag° and Ag+states and the chlorination of the Ag NPs. Complementing these findings, EDX elemental mapping verified the uniform distribution of Cl elements in the Ag NPs, supporting the efficient synthesis of the Ag / AgCl NPs reference electrodes. SPR techniques were employed to investigate the association and dissociation kinetics between the modified aptamer immobilized on an Au surface and its targeted biomarker 104 (FIGS. 2C and 2F). The results demonstrate robust interactions between the aptamer capture agents 22 and the biomarker 104, as depicted in FIG. 2F. As the biomarker concentration increased, there was a linear rise in the response signals. This linearity persisted with minimal variation during the dissociation phase, affirming the efficacy of this aptamer-based immunosensor device 10 in targeted biomarker detection. The robust interactions were attenuated in a pH 2 buffer solution when rejuvenating the immunosensor surface, suggesting the presence of strong non-covalent forces, including hydrogen bonding and Van der Waals interactions.

[0044] Performances of nanoengineered immunosensor devices in biomarkers detection

[0045] Cyclic voltammetry (CV) was performed to investigate the electrochemical behavior of the immunosensor device 10 before and after the specific adsorption of the targeted biomarker 104 (FIG. 3 A). Notably, after the adsorption of the biomarker 104 onto the immunosensor device 10, a pronounced reduction in peak current density was observed. This suggests that the formation of the specific immune complex served as a blocking layer, inhibiting charge transfer on the surface of the working electrode 16. To validate the capability of the immunosensor device 10 to detect various concentrations of four (4) biomarkers 104, electrochemical impedance spectroscopy (EIS) analysis was conducted to examine the linear relationship between multiple concentrations of the targeted biomarkers 104 and charge-transfer resistance (Rct) using the Randles equivalent circuit model (FIGS. 3B and 3C). In the Nyquist plots (FIG. 3B), the semicircular region at high frequencies correlates with the charge-transfer limited process, while the linear segment at low frequencies is associated with the diffusion-limited process. A consistent increase in Ret value was observed with increasing concentrations of targeted biomarkers 104. This specific adsorption reduced the charge transfer between the working electrodes 16 and the [Fe(CN)₆]3- / 4-redox probe,2025-015-2consistent with the CV findings. As presented in FIG. 3C, the calibration curve illustrates the proportional changes in Rctvalues with the concentrations of targeted biomarkers 104, indicating robust linear relations within specified ranges. These linear relations were observed over the ranges of 0.1 pg mL to 104pg mL1for CRP (R2= 0.983), 0.1 ng niL1to 104ng mL1for calprotectin (R2= 0.987), 0.1 ng mL1to 104ng mL1for sP-selectin (R2= 0.982), and 1 ng mL1to 105ng mL1for D-dimer (R2= 0.994). The established limits of detection (LOD, S / N = 3) were determined to be 0.023 pg mL1for CRP, 0.035 ng mL1for calprotectin, 0.019 ng mL-1for sP-selectin, and 0.035 ng mL-1for D-dimer, as detailed in the Tables 1-4. Initially, the four (4) biomarkers 104 combination was proposed for acute thrombosis prediction and manufactured 4-channel immunosensor device 10 to detect the four (4) biomarkers simultaneously. This robust detection capability can be attributed to the large surface area of the CNTs-based immunosensor device 10 and the high specificity of aptamers and antibody.

[0046] The reproducibility, specificity, and storage stability of the manufactured immunosensor device 10 was evaluated for detecting standard and known biomarkers 104 in PBS (pH 7.4) solution. The reproducibility of the nanoengineered immunosensor device 10 was assessed using six independently constructed sensors (FIG. 3D), with relative standard deviation (RSD) values ranging from 3.2% to 5.8%. To assess specificity7, biomarkers 104 were mixed, resulting in distinct sensing signals for each biomarker 104: 95.3% for CRP, 103.2% for calprotectin, 93.2% for sP-selectin, and 103.7% for D-dimer (FIG. 3E).Specificity measurements were repeated three times. In the evaluation of the immunosensor’s reproducibility7(n = 6) and specificity7(n = 3), the repeated measurements showed slight variations, confirming the excellent reproducibility7and specificity7of the nanoengineered immunosensor device 10. Next, the long-term storage stability of the nanoengineered immunosensor device 10 was assessed, revealing sustained sensing signals above 91.1% for CRP, 93.1% for calprotectin, 92.1% for sP-selectin, and 90.5% for D-dimer over two weeks (FIG. 3F), demonstrating the robust stability of the fabricated immunosensor devices 10.

[0047] In summary, the nanoengineered immunosensor device 10 boasted wide-ranging and rapid detection of multiple circulating biomarkers within 1 hour, compared with enzyme-linked immunosorbent assay7(ELISA), which takes about 5 hours.

[0048] Unsupervised clustering to enhance thrombosis prediction

[0049] A chord diagram was used to illustrate the relational network among four (4) biomarkers and 53 blood specimens (FIG. 4A). Notably, the diminished arc span of patient2025-015-249 indicates prediction challenges despite the combination of multiple biomarkers 104. Utilizing the various concentrations detected from the biomarkers 104, unsupervised hierarchical clustering was performed to categorize blood samples 102 into three groups, as depicted in FIG. 4B. To ensure patient confidentiality, a numeric pseudonymization approach was adopted for each patient, aligning the clustering outcomes with the ICD-10 diagnostic codes for thrombosis. Notably, the clustering was solely influenced by the concentrations of the four (4) biomarkers. Patient IDs were organized based on diagnostic outcomes to ensure clarity and facilitate comparative analysis.

[0050] As illustrated in FIGS. 4B and 4C, patients 47, 48, 50, 51, 52, and 53 were identified with the highest thrombotic risk (scores = 2), while patients 3, 17, and 36 exhibited a moderate risk (scores = 1). The other patients were identified with the lowest thrombotic risk (scores = 0). To validate these findings, the results were cross-referenced them with the ICD-10 codes for thrombosis from hospitalized COVID-19 patients. The patient identifiers were reorganized based on the diagnostic outcomes of positive and negative thrombosis. According to the diagnostic results, 46 patient samples 102 were negative for thrombosis, and 7 were positive. For comparative analysis, the negative thrombosis samples 102 were arranged from 1 to 46 and the positive thrombosis samples 102 from 47 to 53. The results of the unsupervised hierarchical clustering were validated against the medical ICD code for thrombosis (FIG. 4C) and accurately predicted positive or negative thrombosis in 49 out of 53 samples 102. A moderate risk rating (score = 1) was assigned to patients 3, 17, and 36, which led to false positive predictions. In contrast, the lowest risk rating (score = 0) was assigned to patient 49, resulting in a false negative prediction.

[0051] Principal Component Analysis (PCA), t-distributed Stochastic Neighbor Embedding (t-SNE), and Uniform Manifold Approximation and Projection (UMAP) are dimensionality reduction techniques, each with distinct characteristics suitable for unsupervised clustering. Although the PCA algorithm correctly clustered patient 49 into the thrombosis-positive group, many false positive samples 102 were identified, including patients 18, 25. 30. 31, 33, 40, and 44. In contrast. t-SNE, a non-linear technique, effectively visualizes high-dimensional data clusters by preserving local structures (FIG. 4E) and shows results similar to those of hierarchical clustering (FIG. 4B). However, t-SNE incorrectly classified patient 49 as part of the thrombosis-negative group, resulting in a false negative prediction. It was identified that the thrombosis-positive group harbors a higher risk (score = 2), including patients 47. 48. 50. 51. 52, and 53. Nevertheless, the distance between the2025-015-2thrombosis-positive group with a higher risk (score = 2) and the thrombosis-positive group with a lower risk (score = 1), including patients 3. 17. and 36. is small, leading to three false positive predictions. UMAP maintains both local and global structures and offers faster computational speeds (FIG. 5F). The single thrombosis-positive group, which includes patients 47, 48, 50, 51, 52, and 53, demonstrates the effectiveness of combining holistic data structures with nonlinear strategies for predicting thrombosis, especially when using biomarker concentrations as indicative features.

[0052] This research focused on predicting acute thrombosis. The four selected biomarkers 104 were associated with acute responses. The pattern of these antigens being predominantly at low or high levels, rather than intermediate levels, was characteristic of acute response markers. These biomarkers 104 typically exhibited a binary expression pattern, reflecting a rapid clinical response to a triggering event, such as acute thrombosis formation. In the context of acute-phase responses, the immune system quickly upregulated the production of these markers during an acute event, leading to a sharp increase in their levels. Conversely, in the absence of such an event, these markers remained at baseline, low levels. The absence of intermediate levels suggested that the regulatory mechanisms controlling these markers function in a switch-like manner, turning on or off rather than increasing or decreasing gradually.

[0053] In contrast, a “4-bit barcode” method (FIGS. 10A-10F) was used to evaluate thrombotic risk, compared to the "‘accurate 4-biomarker concentrations” method (FIGS. 4A-4F). Low biomarker concentrations were coded as 0, while high concentrations were coded as 1. The “4-bit barcode” method was applied as an input feature in various unsupervised machine learning models, including hierarchical clustering (FIGS. 10B and 10C), PCA (FIGS. 10D). t-SNE (FIG. 10E), and UMAP (FIG. 10F). The performance of the “4-bit barcode” method for acute thrombosis prediction was slightly worse than the “accurate 4-biomarker concentrations” method. Specifically, both the “4-bit barcode” method (FIGS. 10B and 10C) and the “accurate 4-biomarker concentrations” method (FIGS. 4B and 4C) produced a false negative for patient 49 in the hierarchical clustering analysis. In the thrombosis-negative group, the “accurate 4-biomarker concentrations” method performed better than the “4-bit barcode” method. Only patients 3, 17, and 36 received false positive predictions using the “accurate 4-biomarker concentrations” method in the hierarchical clustering analysis (FIGS. 4B and 4C). However, patients 3, 6, 17, 36, and 46 received false2025-015-2positives using the “4-bit barcode” method in the hierarchical clustering analysis (FIGS. 10B and IOC).

[0054] Additionally, it was difficult to distinguish between the thrombosis-positive and thrombosis-negative groups in the PCA analysis using the “4-bit barcode” method (FIG. 10D). The t-SNE analysis using the “4-bit barcode” method also incorrectly classified patient 49 as part of the thrombosis-negative group, resulting in a false negative prediction, while identifying patients 3, 6, 17, 36, and 46 as part of the thrombosis-positive group, leading to false positives (FIG. 10E). Similarly, the UMAP analysis using the “4-bit barcode” method incorrectly classified patient 49 as part of the thrombosis-negative group, resulting in a false negative, and classified patients 3, 6, 12, 17. 36, and 46 as part of the thrombosis-positive group, leading to false positives (FIG. 10F). Despite these discrepancies, the “4-bit barcode” method (FIGS. 10A-10F) remains an effective strategy for acute thrombosis prediction and has a comparable, though slightly worse, predictive capability for acute thrombotic events in the COVID-19 cohort compared to the “accurate 4-biomarker concentrations” method (FIGS.4A-4F).

[0055] Statistical analysis to evaluate different combinations of biomarkers

[0056] Based on the ICD-10 diagnostic code, patients 1-46 were negative for thrombosis, and 47-53 were positive. Specifically, the two cohorts were juxtaposed: those tested negative (-) and those positive (+) for thrombosis (FIGS. 5A, 5B, 5C, and 5D). The various concentrations of CRP (FIG. 5A), calprotectin (FIG. 5B). sP-selectin (FIG. 5C), and D-dimer (FIG. 5D) were determined using the fabricated immunosensor devices 10. The accuracy of analyzing these four (4) biomarkers was validated against ELISA as the reference standard (FIGS. 6A-6C through 9A-9C). Of 53 specimens, CRP (p = 0.0191) and D-dimer (p = 0.0427) were statistically significant for predicting thrombosis (FIGS. 5A and 5D). whereas calprotectin (p = 0.0751) and sP-selectin (p = 0.1251) were not statistically significant (FIGS.5B and 5C). In addition, the proximity matrix of four (4) biomarkers and 53 patients is displayed in FIGS. 5E and 5F, respectively.

[0057] A binary logistic regression analysis was performed, correlating the concentration of the four (4) biomarkers 104 of CRP, calprotectin, sP-selectin, and D-dimer with the ICD-10 code for thrombosis prediction. As delineated in FIGS. 5G-5U, an AUC value of 0.95 from the receiver operating characteristic (ROC) curve indicated a high true positive rate (high sensitivity) and a low false positive rate (1 -specificity) when combining these four (4) biomarkers 104 for thrombosis prediction (FIG. 5K). Single-biomarker ROC values were2025-015-2lower than those of the combined biomarkers 104: CRP (AUC: 0.773) (FIG. 5G), calprotectin (AUC: 0.711) (FIG. 5H), sP-selectin (AUC: 0.683) (FIG. 51), and D-dimer (AUC: 0.739) (FIG. 5J). In addition, the AUC was calculated for combinations of two (2) and three (3) biomarkers 104. These AUC values were below the combination of four (4) biomarkers 104 (AUC: 0.95; FIG. 5K): CRP and calprotectin (AUC: 0.891; FIG. 5L); CRP and sP-selectin (AUC: 0.885; FIG. 5M); CRP and D-dimer (AUC: 0.929; FIG. 5N); calprotectin and sP-selectin (AUC: 0.879; FIG. 50); calprotectin and D-dimer (AUC: 0.919; FIG. 5P); sP-selectin and D-dimer (AUC: 0.916; FIG. 5Q); CRP, calprotectin, and sP-selectin (AUC: 0.885; FIG. 5R); CRP, calprotectin, and D-dimer (AUC: 0.916; FIG. 5S); CRP, sP-selectin, and D-dimer (AUC: 0.916; FIG. 5T); calprotectin, sP-selectin, and D-dimer (AUC: 0.929; FIG. 5U). Thus, combining the four (4) biomarkers 104 of CRP, calprotectin, sP-selectin, and D-dimer demonstrates high sensitivity and specificity for thrombosis prediction.

[0058] Specifically, using only CRP and D-dimer yielded an AUC of 0.929, which is close to the 4-biomarker AUC of 0.95. Among the 53 specimens, CRP (p = 0.0191) and D-dimer (p = 0.0427) were statistically significant for predicting thrombosis (FIGS. 5A and 5D), while calprotectin (p = 0.0751) and sP-selectin (p = 0.1251) were not statistically significant (FIGS. 5B and 5C). Among the four biomarkers 104 evaluated, CRP exhibited the most significant statistical association with thrombosis, as evidenced by the lowest p-value (p = 0.0191) and the highest area under the curve (AUC = 0.773) in receiver operating characteristic analysis, outperforming calprotectin, sP-selectin, and D-dimer. CRP is an acute-phase protein produced by the liver in response to inflammatory cytokines, with levels markedly elevated in patients with COVID- 19, correlating positively with disease severity and mortality. Retrospective studies have consistently demonstrated a positive correlation between CRP levels and COVID- 19 severity, further establishing its role in predicting thrombosis risk. When incorporated into predictive models, CRP’s low detection limits enhance the sensitivity of the model, particularly in detecting early or subclinical inflammatory changes that may precede thrombotic events. This attribute positions CRP as an effective early warning signal within a layered prediction model, warranting closer monitoring or additional testing with more specific biomarkers 104. While CRP is highly sensitive, its lack of specificity’ for thrombosis necessitates its use within a multi-biomarker framework to achieve a balance between sensitivity and specificity. The proposed 4-biomarker 104 combination ensures that the predictive model remains responsive to early inflammatory changes while maintaining precision in identifying thrombotic events.2025-015-2

[0059] DISCUSSION

[0060] The machine learning-assisted prediction, using nanoengineered immunosensor devices 10 that detect multiple circulating biomarkers 104, represents an accurate and rapid strategy for predicting acute blood clots. Fiber-laser engraving, CO2-laser cutting, and other techniques disclosed herein were employed to create microchannels, chambers 28, 30, holes or apertures 32, and an inlet 34 and used electrochemical deposition of AuNPs 20 on CNTs 18 for conjugation with capture agents 22 (e.g., aptamers and antibody). This approach enabled high-throughput fabrication of nanoengineered immunosensor devices 10. This strategy provides rapid electrochemical detection of multiple biomarkers 104, followed by unsupervised clustering performed using software 52 to enhance thrombosis prediction.

[0061] The nanoengineered immunosensor device 10 was fabricated using fiber laser-engraved CNTs 18 for working electrodes 16 and CO2laser cutting for microfluidic features, resulting in a large surface area for the electrochemical deposition of AuNPs 20 and conjugation with specific aptamers and antibody as the capture agent 22. Electrical impedance measurements made using an impedance device 40 were highly effective in assessing the binding affinity at the recognition components of the aptamers and antibody, and the conductive surfaces altered the charge-transfer resistance (ΔRct) at the electrode interface. Electrochemical impedance spectra (EIS) showed a robust linear relationship between various concentrations of the targeted biomarkers and Rct, as depicted in FIG. 2F. As the concentration of biomarkers 104 increased, response signals showed a proportional rise. This linearity persisted with minimal variation during the dissociation phase. The interactions were attenuated in a pH 2 buffer solution when rejuvenating the immunosensor surface, suggesting the presence of strong non-covalent forces, including hydrogen bonding and Van der Waals interactions. The established limits of detection (LOD. S / N = 3) were 0.023 pg mL1for CRP, 0.035 ng mL1for calprotectin, 0.019 ng mL-1for sP-selectin, and 0.035 ng mL1for D-dimer, respectively.

[0062] In response to the urgency to combat acute illness-induced thrombosis nanoengineered immunosensor devices 10 were fabricated that are designed to quantify specific circulating biomarkers 104. Currently, ELIS As are the reference standard for protein detection and quantification. The limit of detection in the ‘‘sandwich” ELISA strategy is determined by the binding affinity and specificity of the selected antibody pairs, which were screened and purchased from Thermo Fisher Scientific for optimal performance. However, ELISAs are limited by several experimental constraints that affect the rapid and accurate2025-015-2prediction of thrombosis. These include multi-step labeling processes, costly labeling reagents, and extended detection durations. In this context, the development of the nanoengineered immunosensor device 10 is unique for its rapid fabrication, specific targeting of biomarkers 104, and broad detection spectrum. This enables a customized solution for early prediction of acute illnesses, thus preventing thrombosis-associated complications in vital organ systems.

[0063] Blood clots from thrombosis are well -recognized to cause complications, including pulmonary embolism in the lungs, stroke in the brain, and acute coronary syndrome in the heart, all of which are associated with high morbidity and mortality. CRP is primarily synthesized in the liver in response to inflammatory cytokines, with interleukin-6 (IL-6) being an essential inducer. As an acute-phase protein. CRP levels increase rapidly during systemic inflammation, making it a crucial biomarker 104 for assessing the inflammatory status of patients. Mounting evidence supports a positive correlation between CRP levels and the severity of COVID-19. In contrast to IL-6, CRP levels have emerged as a potential predictive marker for the risk of thrombosis in these patients. Calprotectin levels (S100A8 / S100A9), a cytosolic component of neutrophils, have been independently- associated with thrombosis. Upon activation, platelets express P-selectin, a molecule adept at binding to and activating leukocytes. The vascular inflammatory response induces endothelial cell dysfunction, accompanied by elevated levels of adhesive molecules like P-selectin, which promote thrombus formation. This process underscores the interconnected roles of inflammation and thrombosis, particularly in disease states like COVID-19, where systemic inflammation is prevalent. Subsequent enzymatic processes can cleave P-selectin into its soluble form. D-dimer, the fibrin degradation product, is a biomarker 104 for coagulation and fibrinolysis. D-dimer levels may serve as a valuable tool for thrombosis screening. However, factors such as disease severity, progression, and medication use can vary between patients, limiting the interpretation of changes in individual biomarkers 104. For the selection of four specific biomarkers 104 associated with acute thrombosis, the consensus of the COVID International Thrombosis Biomarkers Colloquium was followed, which recommended a panel of individual biomarkers 104 to predict the risk of developing thrombosis. A strategy was then proposed of combining four biomarkers 104 for acute thrombosis prediction. CRP and calprotectin are typically elevated in various inflammatory and acute infection states. Additionally, sP-selectin and D-dimer were included, which are associated with thrombosis risk. Thus, a strategy was developed that combines these four biomarkers 104 for acute2025-015-2thrombosis prediction. Using this approach, thrombosis was accurately predicted in 49 out of 53 patient specimens through unsupervised clustering based on the concentrations of the four biomarkers 104 (FIGS. 4A-4F). In addition, it was demonstrated that the combined four (4) biomarkers 104 could enhance the sensitivity and specificity for thrombosis prediction (FIGS.5A-5U).

[0064] In the cohort of 53 blood samples 102 collected from hospitalized COVID-19 patients, single biomarkers 104 proved insufficient for predicting thrombotic risk due to inadequate sensitivity and specificity. However, combining all four (4) biomarkers 104 in an integrative approach greatly improved prediction. Four anomalies were identified within the blood specimens: patient 49 rendered a false-negative outcome, and patients 3, 17, and 36 were misclassified during unsupervised hierarchical clustering. Despite this, the thrombotic risk prediction scores for the other 49 patients were consistent with their clinical evaluations. Notably, patient 49 was correctly classified within the thrombosis-positive group using the PCA algorithm, while patients 3, 17, and 36 were classified as thrombosis-negative using the UMAP algorithm. This emphasizes the need for multiple unsupervised learning techniques and cross-comparisons.

[0065] In summary, this disclosure elucidates the combined utility of four (4) biomarkers 104 of CRP, cal protectin, sP-selectin, and D-dimer in augmenting the prediction of acute blood clots. This enhancement is achieved by applying nanoengineered immunosensor devices 10 with unsupervised learning techniques for clustering and analysis. The nanoengineered immunosensor device 10 allows for sensitive and specific detection of a wide range of biomarker concentrations, facilitating unsupervised clustering. Thus, the nanoengineered immunosensor device 10 provides accurate and rapid prediction of acute blood blots, enabling timely responses to acute illnesses and public health crises.

[0066] In the analysis on acute thrombosis prediction, it was demonstrated that combining four biomarkers 104 enhanced the sensitivity and specificity of predicting acute thrombosis using 53 blood specimens from hospitalized COVID- 19 patients (FIG. 5K). 43 out of 46 specimens were successfully predicted in the thrombosis-negative group and 6 out of 7 in the thrombosis-positive group, achieving an overall accuracy of 49 out of 53 patient specimens. Only COVID-19 patient samples were analyzed, with the majority showing no indication of thrombotic risk. This focus raises concerns about the generalizability of the findings to other patient populations with different underlying conditions and risk profiles. Specifically, cancer patients, HIV patients, trauma patients with hemorrhage, septic patients, and those on2025-015-2anticoagulants exhibit distinct pathophysiological mechanisms that may influence biomarker expression differently from COVID-19 patients. For instance, cancer patients often present a hypercoagulable state due to tumor-associated procoagulant factors, while HIV patients may experience chronic immune activation that affects coagulation pathways. Patients on anticoagulants, meanwhile, might exhibit suppressed biomarker levels, potentially leading to false negatives if not accounted for in the predictive model. This represents a limitation of the research.

[0067] The current nanoengineered immunosensor device 10 with unsupervised clustering may be extended to detect cancer patients, HIV patients, trauma patients with hemorrhage, septic patients, or those on anticoagulants, the selected combination of four biomarkers 104 specific to acute thrombosis may not effectively predict these conditions based on the current biomarker concentrations. However, when analyzed through unsupervised clustering, the strategy using the nanoengineered immunosensor device 10 holds potential for predicting these diseases. The layer-by-layer procedures for manufacturing the nanoengineered immunosensor device 10 and the unsupervised clustering analysis strategy are detailed in the Materials and Methods section. If a panel of biomarkers 104 associated with these diseases is identified, the nanoengineered immunosensor device 10 can be tailored to detect these biomarkers 104 by modifying the corresponding capture agents 22 (e.g., aptamers or antibodies) on the working electrodes 16. These biomarker concentrations can then be used as inputs for disease prediction models utilizing unsupervised machine learning, generating a disease risk score for conditions such as cancer, HIV, hemorrhagic trauma, sepsis, and for anticoagulated patients. Based on multiple biomarkers 104, this strategy7is expected to provide an accurate and efficient method for predicting various diseases and guiding personalized medicine.

[0068] The sP-selectin is a marker of platelet activation and endothelial dysfunction, and its elevation can be associated with both arterial and venous thrombosis. However, it is more strongly linked to arterial thrombosis due to its association with platelet activation, as platelet-rich thrombi are commonly involved in myocardial infarction or ischemic stroke. While elevated sP-selectin is more indicative of arterial thrombosis, elevated D-dimer is nonspecific and can indicate the presence of a thrombus in either the arterial or venous system. Elevated levels of CRP and calprotectin, both markers of inflammation, are also linked to thrombosis, but neither definitively indicates arterial versus venous thrombosis. Given the stronger association of sP-selectin with arterial thrombosis, combining elevated sP-2025-015-2selectin with CRP, calprotectin, and D-dimer levels could help predict the location of arterial thrombosis. This information can guide anticoagulation strategies, such as choosing between anti-platelet or anti-thrombotic therapy.

[0069] Currently, the detection strategy requires approximately 1.5 hours to provide a thrombotic risk score. This includes about 25 minutes for blood collection and processing, 1 hour for the simultaneous detection of four (4) biomarkers 104 using multi channel / chamber techniques, and around 5 minutes for unsupervised machine learning calculations performed by software 52 executed by a computing device 50 to determine the thrombotic risk score. In comparison, the commercial enzyme-linked immunosorbent assay (ELISA) takes about 5 hours to detect a single biomarker. While the proposed detection strategy shortens the detection time and demonstrates high accuracy, the acute thrombosis readouts highlight the need for bedside, point-of-care technology with a readout time of less than 15 minutes and minimal user interaction. Additionally, the relatively slow generation of plasma could impact clinical decision-making for acute coronary syndromes, stroke, pulmonary embolism, or traumatic injury. With emerging automation techniques that integrate blood collection and processing, multi-chamber detection, and unsupervised machine learning into a single system, a shortened turnaround time is anticipated by minimizing manual operations and further reducing detection time.

[0070] The strengths and shortcomings of the current clinical assays (e.g.. aPTT, PT, NIR, clotting time) and the disclosed immunosensor device 10 are summarized in Table 6. In summary, while aPTT and PT are effective for monitoring anti coagulation therapy, they lack specificity for predicting acute thrombosis. Near-Infrared (NIR) shows promise for non-invasive monitoring but has not yet been widely adopted in clinical practice and may exhibit variable sensitivity and specificity. Cloting time is a quick test but is unreliable for thrombosis risk prediction due to its limited specificity and sensitivity. In contrast, the proposed strategy offers advantages, including high sensitivity, specificity, and accuracy for acute thrombosis prediction, while being simple, rapid, cost-effective, and customizable for multiple biomarkers. The proposed strategy is also readily accessible and user-friendly for blood testing, providing thrombotic risk assessments that could aid in acute thrombosis management and support personalized patient care. However, current limitations include validation restricted to only 53 specimens and its use being limited to research setings. The immunosensor device 10 provides a fundamental and experimental basis to expand validation to larger cohorts for future clinical translation.2025-015-2

[0071] Furthermore, including diverse clinical data (e.g., sex, age, thrombosis risk factors) and medical history (e.g., antiplatelet, anticoagulation drugs) is crucial for conducting a comprehensive analysis to strengthen this research. The diversity of clinical data enhances the generalizability of acute thrombosis prediction. Specifically, men and women often have different risk profiles for thrombosis due to biological differences such as hormone levels, genetic factors, and immune responses. Incorporating sex into the model helps tailor predictions to these differences and ensures that both sexes are adequately represented, improving accuracy across different patient populations. Thrombosis risk generally increases with age due to factors like reduced mobility, increased incidence of comorbidities, and changes in the coagulation system. Including age in the prediction model allows for consideration of these age-related variations, enabling risk stratification into different categories for personalized predictions. Incorporating a wide range of thrombosis risk factors allows the model to provide a more comprehensive risk assessment, capturing complex interactions between factors. Some risk factors may have a more substantial impact when combined, and accounting for these interactions improves the predictive power of the model. Patients on antiplatelet or anticoagulation therapy have modified thrombosis risk. While these medications are designed to reduce clot formation, their effectiveness can vary based on individual patient factors. Including this information in the model helps account for the protective effects of these drugs and potential variations in their efficacy. In summary, the diversity of clinical data is essential for developing a robust acute thrombosis prediction model. By incorporating factors such as sex, age, thrombosis risk factors, and medication use, the model can better account for individual variability, resulting in more personalized and reliable predictions.

[0072] MATERIALS AND METHODS

[0073] Materials and reagents

[0074] Gold(III) chloride hydrate (-52% Au basis), hexachloroplatinic(IV) acid hydrate (-40% Pt basis), silver(I) nitrate (>99.0%), potassium chloride hydrate, sodium hypochlorite solution (available chlorine 4.00-4.99%), sulfuric acid (95.0-98.0%), nitric acid (70%). potassium hexacyanoferrate(II) trihydrate (>98.5%), potassium ferricyanide(III) (99%), toluene (99.9%), xylene (99%), HEPES solution, Tween 20, glycine (>98.5%), silver chloride (99%), ethanol (>99.5%), sodium chloride (>99.0%), phosphate buffer solution (PBS, 1.0 M, pH 7.4), N-hydroxysuccinimide (NHS, 98%), N-ethylcarbodiimide (EDC, >97.0%), and2025-015-2human CRP, human calprotectin, human sP-selectin, human D-dimer, and bovine serum albumin (BSA) were all purchased from Sigma Aldrich.

[0075] The CNTs 18 (XFM01, multi-walled, 5-15 nm in diameter, 10-30 pm in length) were sourced from XFNANO Technology Co. Ltd. The polyvinyl chloride (PVC) substrate 14 was procured commercially from Takiron Co. Ltd. Styrene-isoprene styrene (SIS, DI 113) elastomer was acquired from Kraton Corporation. Adhesive VHB tapes were obtained from 3M Company. The aptamers (i.e., capture agent 22) were synthesized and purified by Integrated DNA Technologies (Table 5). The anti-calprotectin (S100A8 / S100A9, MABF291) antibody (i.e., capture agent 22) was purchased from Sigma Aldrich. The Invitrogen™ CRP Human ELISA Kit, Invitrogen™ Calprotectin Human ELISA Kit, Invitrogen™ sP-selectin Human ELISA Kit, and Invitrogen™ D-Dimer Human ELISA Kit were all obtained from Thermo Fisher Scientific. The ultrapure water (18.2 megohm cm) was purified using a Millipore system.

[0076] Fabrication of a nanoengineered immunosensor device

[0077] SIS (1 g) was initially dissolved in a 20 mL solvent mixture of toluene and xylene at a 1:1 ratio. This SIS solution was subsequently sprayed onto a PVC substrate 14 (layer 1) to form an adhesive layer (layer 2). Acid-treated CNTs were ultrasonically dispersed in ethanol and sprayed onto the SIS adhesive layer (layer 2). Electrodes and connection wires for the CNTs 18 (layer 3) were patterned using selective laser ablation via a 1064-nm Monport 30W fiber laser engraver and marking system. The CNTs connection wires were subsequently encapsulated using a diluted SIS solution applied through a pre-designed mask, forming layer 4. For the preparation of layer 5, the Au NPs 20, Pt NPs, and Ag NPs were electrochemically deposited onto the four working electrodes 16, the counter electrode 26, and the reference electrode 24 using gold (III) chloride hydrate, hexachloroplatinic (IV) acid hydrate, and silver (I) nitrate solutions, respectively. The Ag NPs were then chloridized using a diluted sodium hypochlorite solution, resulting in AgCl / Ag NPs.

[0078] For the sixth layer of the immunosensor device 10, thiol-modified aptamers (50 pM) specific to CRP, sP-selectin, and D-dimer were incubated on the Au NP working electrodes 16 for one hour, producing the Aptamer / AuNPs / CNTs electrode. To modify the calprotectin antibody onto the carboxylated CNTs, the CNTs were initially incubated in a solution containing EDC (0.15 M) and NHS (0.1 M) for one hour to activate the carboxylic groups. This was followed by adding the calprotectin antibody solution and incubating for an hour.2025-015-2

[0079] The detection chamber 28 (layer 7). washing chamber cover (layer 8) with specific flow-directing holes, washing chamber 30 (layer 9). and top cover (layer 10) were patterned using a 50-W CO2laser cutter from Universal Laser Systems. A 3M VHB double-sided tape was patterned to incorporate four secondary chambers as the detection chamber 28 (layer 7). Using the CO2laser, four distinct holes 32 were cut into a 0.1 mm thick PVC film for the washing chamber cover (layer 8). The washing chamber 30 (layer 9) featured a unified flow channel 36 for waste liquid collection. Lastly, another PVC film was cut to introduce a hole as the liquid inlet 34 for the top cover (layer 10).

[0080] Materials characterization

[0081] X-ray photoelectron spectroscopy (XPS) measurements were conducted using a Kratos AXIS Ultra DLD spectrometer. High-resolution transmission electron microscopy (HRTEM) images and energy-dispersive X-ray (EDX) elemental mapping were acquired with an FEI TITAN transmission electron microscope operating at 120 kV.

[0082] Operational workflow of the nanoengineered immunosensor device

[0083] The priming of the chip 12, loading with plasma, signal acquisition, and regeneration of the nanoengineered immunosensor device 10 are shown in FIG. 11. Briefly, the immunosensor device 10 was washed with PBS (pH 7.4, 1 mL, repeated three times) before use. A 100-fold diluted plasma sample 102 (0.2 mL) w as pipetted into the immunosensor device 10 via the inlet 34. After a 45-minute incubation to allow for the formation of non-covalent solid interactions between the targeted biomarkers 104 in each of the four chambers 28 and the modified aptamers or antibody, the residual plasma was removed with PBS (pH 7.4, 1 mL, repeated three times). Electrochemical measurements were conducted using a Gamry Interface 1010E w orkstation or potentiostat 40 (Gamry Instruments, Inc., USA). Electrochemical impedance spectroscopy (EIS) was performed over a frequency range of 100 kHz to 0.1 Hz. The EIS data were fitted to the Randles model to extrapolate the R parameter. Cyclic voltammetry (CV) was performed with potentials set between 0.4 V and 1.3 V (vs. reversible hydrogen electrode, RHE) at a scan rate of 50 mV / s. For electrochemical measurements, the [Fe(CN)e]3’4‘ solution (0.2 mL) was pipetted into the individual detection chambers (measurement time: 5 minutes). After detecting the four biomarkers 104, the [Fe(CN)6]3' / 4‘ solution was removed from the detection chambers 28, which were then washed with PBS (pH 7.4, 1 mL, repeated three times). Subsequently, the detection chambers 28 were regenerated using 0.1 M glycine-HCl buffer (pH 2, 1 mL, repeated twice) to remove the bound biomarkers 104 from the capture agents 22 (e.g.,2025-015-2aptamers or antibody). Finally, the chambers 28, 30 were washed with PBS and stored at 4°C for future use.

[0084] The performance of the fabricated nanoengineered immunosensor device 10 was compared with previously reported immunosensors and demonstrated its high detection capabilities for the four biomarkers 104 (Tables 1-4).

[0085] Surface plasmon resonance

[0086] Surface plasmon resonance measurements for binding kinetics were performed on a Biacore 2000 system. Assays were conducted at 25°C using a running buffer composed of 10 mM HEPES, 150 mM NaCl, pH 7.4, and 0.005% (v / v) Tween-20. The aptamer was immobilized onto a Series S Sensor Chip SA (GE Healthcare). A 2-fold dilution series of the specific biomarker was injected over the immobilized aptamer at a flow rate of 30 pl / min, with association and dissociation times of 180 s and 900 s, respectively. The aptamer was regenerated using 0.1 M glycine-HCl buffer (pH 2) for two 10-second intervals.

[0087] Ethics and Human Subjects

[0088] All procedures and handling of human blood samples were conducted in accordance with the guidelines set by UCLA and received approval from the UCLA Institutional Review Board (no. IRB-23-1768). Diagnoses of SARS-CoV-2 infection and thrombosis were obtained from the UCLA Pathology7Biobank and Biospecimen Research Core.

[0089] Biosafety

[0090] All blood samples 102 were processed in accordance with the biocontainment procedures established for handling SARS-CoV -2-positive samples. These specimens were approved by the UCLA IRB protocol and provided by the UCLA Pathology7Biobank and Biospecimen Research Core. All participants provided informed consent. The blood samples 102 were de-identified, ensuring that no personally identifiable information was associated wi th them.

[0091] Patient plasma samples

[0092] Patients testing positive for SARS-CoV-2 were recruited into the study subsequent to their diagnosis. The diagnosis of thrombosis was validated with the ICD-10 diagnostic code. Blood samples 102 were handled in strict accordance with the biocontainment procedures designed for SARS-CoV-2-positive specimens.

[0093] Unsupervised hierarchical clustering and dimensionality reduction analysis2025-015-2

[0094] Hierarchical clustering and dimensionality reduction were executed using the Scikit-leam library in Python (scikit-leam.org / stable / index.html, which is incorporated by reference herein) and GraphPad Prism 9 software 52. Scikit-leam is an open-source machine learning library that supports supervised and unsupervised learning. It also provides various tools for model fitting, data preprocessing, model selection, model evaluation, and many- other utilities. Furthermore, data visualization, including Chord diagrams, was accomplished using Origin Lab 2021 and GraphPad Prism 9 software suites. Specifically, Ward’s agglomerative hierarchical clustering technique was employed, performing unsupervised clustering based on Euclidean distances, and emphasized optimizing the cluster count for the datasets. GraphPad Prism software 52 was used for principal component analysis (PCA) and the Scikit-leam library in Python for nonlinear dimensionality- reduction methods, including t-SNE and UMAP, to transform high-dimensional data into two-dimensional spaces.

[0095] Statistical analysis

[0096] Statistical analy ses were conducted utilizing IBM SPSS Statistics 26, GraphPad Prism 9 software, and Python scripts. The p value was determined by the Mann-Whitney U test. A correlation was considered statistically significant at p values below 0.05. All data are presented as the mean ± standard error of the mean (n = 3), unless otherw ise specified.

[0097] ROC curves

[0098] To assess the precision of thrombosis prognosis, the numbers for true positives (TP), true negatives (TN), false positives (FP), and false negatives (FN) were tallied. The ROC curve was illustrated based on sensitivity and 1 -specificity scores. The 95% confidence interval (CI) was computed for each AUC value. Binary- logistic regression is a statistical method used to predict the probability of an event occurring given a set of predictor variables, when the outcome variable is binary: thrombosis (-) and thrombosis (+). A binary logistic regression model was initially calculated using the given dataset. Subsequently, the model's predicted probabilities for the positive class were employed to construct an ROC curve.

[0099] Comparison with previously reported immunosensors and thrombosis prediction strategies.

[0100] The performance of the fabricated nanoengineered immunosensor device 10 was compared with previously reported immunosensors (Tables 1-4). Additionally, the strengths and shortcomings of current clinical assays (e.g., aPTT, PT, NIR, clotting time) and the strategy- proposed in this research are summarized in Table 6.2025-015-2

[0101] Blood collection and processing

[0102] The first step involved collecting blood from the patient using standard phlebotomy techniques into tubes containing an anticoagulant to prevent clotting. The blood was then centrifuged at a low speed of 1500g for 10 minutes, separating it into three layers: red blood cells at the bottom, a buffy coat layer containing white blood cells and platelets in the middle, and plasma at the top. The top plasma layer was carefully aspirated using a pipette, taking care not to disturb the buffy coat layer, as this plasma may still contain some platelets. The aspirated plasma was then centrifuged at a higher speed of 3000g for 10 minutes, causing any remaining platelets to form a pellet at the bottom of the tube. After this second centrifugation, the supernatant (plasma) was carefully pipetted into a new sterile container, ensuring no disturbance of the platelet pellet. This supernatant was the prepared plasma used as the sample 102 for testing thrombosis biomarkers 104 with the nanoengineered immunosensor device 10.

[0103] The screening strategy for antibody pairs in the ELISA kit

[0104] For the '’sandw ich" ELISA strategy, the affinity of the antibody pair (two antibodies) is crucial in determining the limit of detection. During the screening process, potential antibody pairs were tested in various combinations of capture and detection antibodies to identify those with the highest affinity' and specificity7for the antigen. High-throughput screening methods, such as ELISA, were utilized to assess each antibody pair’s binding affinity (the strength of binding to the antigen) and specificity (the ability to bind the antigen without cross-reacting with other molecules). Differences in the limits of detection for the targeted antigens could be attributed to variations in the binding affinity7of the selected antibody pairs. The commercial company (Thermo Fisher Scientific, Cleveland, OH, USA, Invitrogen™) conducted extensive screening of antibody pairs using binding assays to select the high-affinity antibodies incorporated into their ELISA kits. These kits were purchased from Thermo Fisher Scientific (Cleveland, OH, USA, Invitrogen™) to acquire the necessary7antibody pairs.

[0105] In this immunosensor device 10, a label-free strategy (one aptamer or antibody) was employed for specific biomarker detection. The high-affinity aptamers targeting CRP, soluble P-selectin, and D-dimer were synthesized and purified by Integrated DNA Technologies (IDT). Aptamers, which can be synthesized in vitro via PCR methods, provide a cost-effective alternative to antibodies. Additionally, the antibody for calprotectin detection was purchased from Thermo Fisher Scientific (Cleveland, OH, USA). The objective was to2025-015-2achieve rapid prediction of acute thrombosis. Therefore, a label-free detection strategy was utilized to reduce the detection time from ~ 5 hours to ~ 1.5 hour. This approach involved measuring electrochemical impedance signals generated by the formation of specific immune complexes, which act as a blocking layer and inhibit charge transfer on the surface of the working electrode 16. The Randles equivalent circuit was then used to calculate chargetransfer resistance ( / ct). which correlates with the biomarker concentration.

[0106] Table 1. Comparison of various immunosensors for CRP detection.Recognition Detection Range Detection Limit Strategy Detection MethodElement (pgmL'1) (pg mL'1)MSD Antibody Electroluminescent 8xl0‘" -1 8x]0‘"MNPs Antibody Labeled 1.2- 3. M1020.12GNRs Antibody Labeled MlO' lO 0.5 x IO'3Au Aptamer Label-free 1.15xl0 -l.15x10 1.15X10'1CNTs Aptamer Label-free Lld'-lO40.023

[0107] MSD: Multiplexed ELISA with mesoscale discovery reader. MNPs: Magnetic nanoparticles; GNRs: Gold nanorods; CNTs: Carbon nanotubes

[0108] Table 2. Comparison of various immunosensors for calprotectin detection.Recognition Detection Detection Range Detection LimitMaterialElement Strategy (ng ml / 1) (ng mL'1) PtNi@CuMOLs Antibody Labeled 2xl0'4-5 l0 1.377x1 O'4PS NPs Antibody Labeled 3 lO5-2.47 lO’ 3 l05PMMA sphere Imprinted Photonic 0.1-9.8Ag@ZnO Antibody Labeled lxlO2-lxlO4IxlO2CNTs Antibody Label-free MlQ-'-lO40.035

[0109] PtNi@CuMOFs: PtNi nanoparticles functionalized 2D Cu-metal organic framework nanosheets; PS NPs: polystyrene particles; PMMA: poly (methyl methacrylate): CNTs: Carbon nanotubes

[0110] Table 3. Comparison of various immunosensors for sP-selectin detection.Recognition Detection Detection Range Detection LimitMaterialElement Strategy (ng mL'1) (ngmL'1)Antibody Labeled lx 0'2-10 LIO'3CNTs@GNB Antibody Labeled M10^-M1048.5xl0’4Peptidic P-sheet Antibody Labeled 1 1-3.5x10 1.1Antibody Labeled 0.5-1 0.5 CNTs Aptamer Label-free MIO^-IO40.019

[0111] CNTs: Carbon nanotubes; GNB: Gold nanobone2025-015-2

[0112] Table 4. Comparison of various immunosensors for D-dimer detection.Detection Recognition Detection Detection RangeMaterial Limit (ng mL'Element Strategy (ng mL'1)Graphene Antibody Label-free 1— 103. 3x1 O'1ZrO2 Antibody Labeled 5* 1 O'2-6x 1022 1 *102AuNCs Antibody Labeled 5* 102-U1022.92*105Polypyrrole Antibody Label-free IxlO' xlO2IxlO'1CNTs Aptamer Label-free IxlO'-lO40.035

[0113] AuNCs: Gold nanoclusters; CNTs: Carbon nanotubes

[0114] Table 5. The sequences of the aptamers.Aptamers Sequences5 ’ -GCCTGT AAGGTGGTCGGTGTGGCRP CGAGTGTGTTAGGAGAGATTGC-3’ (SEQ IDNO: 1)5 ’ - ACGCUC AACGAGCC AGG AAC AUCG ACGUsP-selectinCAGCAAACGCGAGCGCAACCAGUAAC ACC-3' (SEQ ID NO: 2)5'-GCGCGGTCCCGATTTGGTGTD-dimerAAAATTCCCTCAGCCCTACA-3’ (SEQ ID NO: 3)

[0115] Table 6. Comparison of the strengths and shortcomings of current clinical assays and the immunosensor device for predicting acute thrombosis.Assay Advantages DisadvantagesaPTT - Widely used and standardized. - Not specific for thrombosis.- Useful for heparin therapy. - affected by various factors.- Identifies abnormalities in - Poor sensitivity to coagulation pathways. thrombosis riskPT - Widely available. - Not specific for thrombosis.- Monitors warfarin therapy. - Affected by liver function.- Assesses extrinsic pathway - Limited predictive value for function thrombosis riskNIR - Non-invasive and can be used for - Requires specializedreal-time monitoring. equipment and expertise.- Potential for assessing blood flow - Limited to research settings, and tissue oxygenation. not widely adopted in clinics.- Capable of detecting changes - Sensitivity and specificityblood characteristics. arc variable.Clotting Time - Simple and rapid test. - Influenced by various- Can provide a quick assessment factors.of hemostasis. - Lacks specificity andsensitivity.Current - Simple and rapid test. - Validation in 53 specimens.Immunosensor2025-015-2Device - High sensitivity, specificity, and - Limited to research settings,accuracy for acute thrombosis. not widely adopted in clinics.- Cost effectiveness andcustomizable for multiple_ biomarkers. _

[0116] aPTT: Activated partial thromboplastin time; PT: Prothrombin time; NIR: Nearinfrared spectroscopy; Immunosensor device: Nanoengineered immunosensor device with unsupervised clustering

[0117] While embodiments of the present invention have been shown and described, various modifications may be made without departing from the scope of the present invention. For example, while the immunosensor device 10 is disclosed for the prediction of acute thrombosis, immunosensor design may be designed with other biomarkers 104 for the determination or prediction of other diseases or medical conditions. Additional examples include, by way of illustration and not limitation, pulmonary hypertension, pulmonary embolism, and deep vein thrombosis. The invention, therefore, should not be limited, except to the following claims, and their equivalents.

Claims

1. 2025-015-22.What is claimed is:

1. An immunosensor device for determining or predicting a disease state or condition of a subject using a sample from the subject comprising:4.a chip comprising a substrate having disposed thereon a plurality of working electrodes each formed with carbon nanotubes and having gold nanoparticles deposited thereon and further conjugated to aptamers or antibodies specific a different biomarker of a plurality of different biomarkers, a reference electrode, and a counter electrode;5.a detection chamber formed over the plurality of working electrodes; and6.a washing chamber disposed over the detection chamber, wherein the washing chamber fluidically communicates with the detection chamber.

2. The immunosensor device of claim 1, further comprising an impedance measuring device that is configured to measure impedance changes at the plurality of working electrodes in response to exposure to the sample and a solution of [Fe(CN)6]3+ / 4+or other redox couple exposed to the working electrodes.

3. The immunosensor device of claim 2, further comprising a computing device having software executed thereby that performs unsupervised clustering of measured impedance changes or concentration data for the plurality of different biomarkers and outputs a determination or prediction of the disease state or condition based on the unsupervised clustering.

4. The immunosensor device of claim 3, wherein the measured impedance changes are converted to biomarker concentrations prior to unsupervised clustering.

5. The immunosensor device of claim 3, wherein the measured impedance changes are converted to multi-bit barcode corresponding to biomarker concentrations prior to unsupervised clustering.

6. The immunosensor device of claim 1, wherein the diseased state or condition comprises acute thrombosis and the aptamers or antibodies are specific to C-reactive protein (CRP), calprotectin, soluble platelet selectin (sP-selectin), and D-dimer.2025-015-27. The immunosensor device of claim 1, wherein the sample comprises blood plasma.

8. The immunosensor device of claim 1, wherein the detection chamber and washing chamber are formed in a plurality of layers stacked on the substrate of the chip.

9. The immunosensor device of claim 1, further comprising a top layer having an inlet that fluidically communicates through the washing chamber and the detection chamber to the plurality of working electrodes.

10. The immunosensor device of claim 2, wherein impedance measuring device comprises an electrical workstation or dedicated impedance measuring device configured to perform cyclic voltammetry and / or electrochemical impedance spectroscopy.

11. A method of using the immunosensor device of any of claims 1-10, comprising:17.washing the working electrodes of the immunosensor device with a buffer solution; loading a sample into an inlet in the chip and incubating the sample with the immunosensor device for a period of time;18.washing residual sample from the chip;19.measuring impedance changes at the plurality of working electrodes with an impedance measuring device in response to exposure to the sample and a solution of [Fe(CN)6]3+ / 4+or other redox couple exposed to the working electrodes; and performing unsupervised clustering of the measured impedance changes or concentration data for the plurality of different biomarkers using a computing device having software executed thereon and outputting a determination or prediction of the disease state or condition based on the unsupervised clustering.

12. The method of claim 11, wherein the measured impedance changes are converted to biomarker concentrations prior to unsupervised clustering.2025-015-213. The method of claim 11, wherein the measured impedance changes are converted to multi-bit barcode corresponding to biomarker concentrations prior to unsupervised clustering.

14. The method of claim 11, wherein the diseased state or condition comprises acute thrombosis and the aptamers or antibodies are specific to C-reactive protein (CRP), calprotectin, soluble platelet selectin (sP-selectin), and D-dimer.

15. The method of claim 11, further comprising regenerating the immunosensor device by exposing the working electrodes to glycine-HCl buffer.

16. The method of claim 11, wherein the plurality of working electrodes include at least two electrodes.

17. The method of claim 11, wherein the plurality of working electrodes include four electrodes, wherein each of the four electrodes compnse capture agents specific to different biomarkers.

18. The method of claim 11, wherein the determination or prediction of the disease state or condition is output within about 1.5 hours or less.

19. The method of claim 11, wherein the measured impedance changes or concentration data are subject to dimensionality reduction.

20. The method of claim 11, wherein the unsupervised clustering comprises hierarchical clustering.

Images