A small extracellular vesicle detection method based on solid-state nanopore and machine learning

Abstract

The application relates to a small extracellular vesicle detection method based on a solid-state nanopore and machine learning. The method comprises the following steps: detecting a to-be-detected sample by using a solid-state nanopore chip, making an electroosmotic flow dominate particle capture and a through hole by applying a voltage; collecting a nanopore ion current signal and performing filtering, baseline correction and detrending pretreatment; extracting equivalent charge loss, blocking amplitude, half-residence time, left edge slope and right edge slope of a single event through hole signal; inputting the features into a trained machine learning model to obtain a specific event correlation score, taking a correlation event ratio in a sample as a discrimination index, and realizing detection of target small extracellular vesicles. Compared with an existing detection method which needs labeling and enrichment, the application can realize rapid, label-free and single-particle-level detection analysis in a complex plasma matrix.

Description

Technical Field

[0001] This invention belongs to the field of biomedical detection and nanosensing technology, specifically involving a method for single-particle electrical characterization of plasma nanoparticles using solid nanopores, and combining machine learning to achieve label-free identification and quantitative detection of small extracellular vesicles. Background Technology

[0002] Extracellular vesicles (EVs), especially small extracellular vesicles (sEVs) with a diameter of approximately 30-100 nm, are secreted into body fluids by various cells via the endosome pathway and are widely present in biological samples such as plasma, serum, and urine. sEVs can carry biomolecular information such as proteins, nucleic acids, and lipids from their source cells, playing a crucial role in intercellular communication, disease development, and the regulation of the tumor microenvironment. In recent years, numerous studies have demonstrated that tumor cell-derived sEVs exhibit high stability and detectability in body fluids, making them highly promising tumor markers for liquid biopsies. However, plasma is a highly complex biological matrix, containing not only sEVs but also lipoprotein particles (such as very low-density lipoprotein VLDL, low-density lipoprotein LDL, and high-density lipoprotein HDL), protein complexes, and amorphous nanoparticles. Some lipoprotein particles significantly overlap with sEVs in terms of particle size range, charge properties, and porosity, posing a significant challenge to physical characteristic-based detection and differentiation. Therefore, achieving rapid, accurate, and specific detection of tumor-associated sEVs in unenriched or unlabeled raw plasma samples remains a technical challenge in the field of liquid biopsy.

[0003] Existing sEV detection technologies mainly include nanoparticle tracking analysis (NTA), dynamic light scattering (DLS), transmission electron microscopy (TEM), flow cytometry, and immunocapture-based detection methods. Among these, NTA and DLS primarily rely on particle size distribution information, making it difficult to distinguish nanoparticles of similar size but from different origins; while TEM can provide high-resolution morphology information, sample preparation is complex and throughput is low, making it difficult to meet clinical testing needs; traditional flow cytometry is limited by optical resolution, typically making it difficult to directly detect particles smaller than 100 nm, and often requires fluorescent labeling or signal amplification steps; although immunocapture or labeling methods can improve specificity, they rely on antibodies or fluorescent reagents, resulting in complex detection procedures, high costs, and the potential introduction of bias or influence on the native state of sEVs.

[0004] Solid-state nanopore technology is a detection method based on single-molecule electrical signals. Its basic principle is to construct nanoscale pores in an insulating film. When charged biological particles pass through the nanopores under an applied electric field, they cause instantaneous blockage of ion currents. By analyzing the amplitude, duration, and morphological characteristics of the blockage event, real-time detection of single particles can be achieved. However, directly applying solid-state nanopore technology to the detection of plasma sEVs still faces several challenges. First, the surface potential of sEVs is usually low; relying solely on electrophoretic forces results in low capture efficiency and a low incidence of through-pore events, leading to insufficient effective signals. Second, a large number of non-target nanoparticles in plasma can also generate current blockage events. Relying solely on traditional low-dimensional parameters such as blockage amplitude or residence time makes it difficult to reliably distinguish tumor-related sEVs in complex contexts. Furthermore, the high viscosity and complex composition of plasma samples easily introduce baseline drift and non-stationary noise, further increasing the difficulty of signal analysis.

[0005] In recent years, some studies have attempted to enhance the capture efficiency of low-charged nanoparticles by introducing electroosmotic flow (EOF) by modifying the buffer system, electrode polarity, or nanopore surface properties. Electroosmotic flow is generated by the overall migration of surface charge on the inner wall of the nanopore and the counterion layer in the solution under the influence of an applied electric field; its direction and intensity can be controlled by adjusting the voltage polarity and solution conditions. However, improving capture efficiency solely through EOF is still insufficient to address the lack of specificity caused by the high degree of overlap in particle types against the complex background of plasma.

[0006] On the other hand, with the development of machine learning technology, pattern recognition methods based on multi-feature fusion have gradually been introduced into the field of biosignal analysis. However, there is still a lack of mature and systematic solutions for how to effectively combine machine learning methods with solid-state nanopore detection processes and construct stable discrimination strategies suitable for complex plasma matrices.

[0007] Therefore, there is an urgent need for a detection method that can achieve both high capture efficiency and high discrimination specificity in unlabeled or enriched plasma samples, in order to realize single-particle-level analysis and sample-level diagnostic discrimination of tumor-related sEVs, thereby providing a rapid, low-cost technical solution with clinical application potential for early cancer screening and auxiliary diagnosis. Summary of the Invention

[0008] Given that existing technologies for detecting small extracellular vesicles in complex biological matrices such as plasma generally suffer from problems such as low capture efficiency, insufficient specificity, reliance on labeling or enrichment steps, and difficulty in achieving reliable single-particle-level discrimination, this invention aims to provide a method for detecting small extracellular vesicles based on solid-state nanopores and machine learning.

[0009] A primary objective of this invention is to enable direct detection of raw plasma samples without relying on exosome enrichment, immune capture, or fluorescent labeling, thereby achieving efficient capture and single-particle electrical characterization of target small extracellular vesicles.

[0010] Another major objective of this invention is to increase the permeability probability of small extracellular vesicles with low surface potential in solid nanopores by introducing a particle-driven mechanism dominated by electroosmotic flow, thereby significantly increasing the number of effective events and overcoming the problem of insufficient capture efficiency when relying solely on electrophoretic drive.

[0011] Another major objective of this invention is to: extract specific complementary electrical and morphological features from the single-particle through-pore signal and combine them with a machine learning classification model to achieve statistical differentiation between the target small extracellular vesicles and other nanoparticles (especially lipoprotein particles) in plasma, thereby improving detection specificity and accuracy.

[0012] To achieve the above-mentioned objectives, the present invention adopts the following technical solution.

[0013] In a first aspect, the present invention provides a method for detecting small extracellular vesicles, the detection method comprising: The sample to be tested is taken and detected using a solid-state nanopore chip. By applying voltage, the particles in the sample are driven to pass through the nanopore, and the ion current signal of the nanopore is collected. The ion current signal is filtered, baseline corrected and detrended, and single particle passing through the pore event is identified to realize the detection and counting of small extracellular vesicles. A constructed machine learning model is used to calculate the score of the output via event as the target small extracellular vesicle event; the input variables of the machine learning model include the equivalent charge defect (ECD) and current blocking amplitude of the single-particle via event. ΔI ), half-dwell time, left edge slope (Left_k1), and right edge slope (Right_k2); The proportion of through-hole events with scores higher than a preset threshold for the target small extracellular vesicle event is statistically analyzed in the total number of through-hole events. Based on the proportion, it is determined whether the sample to be tested is a specific sample, thereby realizing the detection and counting of specific small extracellular vesicles in the sample.

[0014] This invention designs a method for detecting small extracellular vesicles based on solid-state nanopores and machine learning. By introducing a particle-driven mechanism dominated by electroosmotic flow, the capture efficiency of small extracellular vesicles with low surface potential is significantly improved. A single-particle electrical signal analysis strategy based on specific feature fusion is adopted, combined with probabilistic discrimination by a machine learning model, which significantly improves the ability to distinguish small extracellular vesicles in complex plasma backgrounds. There is no need for exosome enrichment, immune capture, or fluorescent labeling of plasma samples. The detection process is simple, short, and low-cost, and avoids the bias introduced by labeling or enrichment steps, which is conducive to maintaining the natural state of small extracellular vesicles. Based on single-particle-level signal statistics, sample-level diagnostic indicators are constructed, which can maintain good stability and reproducibility among different samples. It is suitable for various applications such as small extracellular vesicle identification (non-diagnostic purposes) and cancer efficacy monitoring.

[0015] Optionally, the sample to be tested may include plasma.

[0016] Optionally, the small extracellular vesicles include physiological or pathological (such as cancer) related small extracellular vesicles.

[0017] Optionally, the detection process using a solid-state nanopore chip includes: A solid nanopore chip is mounted in a fluid cell, which is divided into two independent storage chambers: a forward chamber and a reverse chamber. Electrolyte solutions are added to the forward and reverse chambers. The sample to be tested is added to the fluid cell, and under the action of voltage, electroosmosis becomes the dominant driving force for particles in the sample to pass through the nanopores.

[0018] In this invention, the electroosmotic flow direction can be aligned with the pore direction of the target small extracellular vesicles by adjusting the voltage polarity, electrolyte conditions, and nanopore surface characteristics, thereby increasing the capture probability of small extracellular vesicles and reducing invalid background events.

[0019] Optionally, the solid-state nanoporous chip includes silicon nitride (SiN). x A thin film, wherein the silicon nitride thin film has nanopores, the diameter of the nanopores being 150~350 nm and the thickness being 15~40 nm.

[0020] Optionally, the solid nanoporous chip is hydrophilized before detection, such as by using a strong oxidation system to clean and activate the chip surface, so that the inner wall of the nanopore forms a stable hydrophilic surface state. Specifically, the strong oxidation system includes a "piranha" solution.

[0021] Optionally, the voltage is 100~300 mV; the frequency of acquiring the nanopore ion current signal is 90~110 kHz; and the filtering frequency is 8~12 kHz.

[0022] Optionally, the electrolyte solution includes a neutral electrolyte buffer, such as PBS solution or physiological saline.

[0023] Optionally, the identification of single-particle via events includes: analyzing the noise statistical characteristics of the signal within a local time window, setting a current change threshold based on the background noise level, and determining the start of a via event when the current signal momentarily deviates from the baseline and exceeds the threshold; and determining the end of the event when the signal recovers to near the baseline.

[0024] Optionally, the method for constructing the machine learning prediction model includes: A known target small extracellular vesicle sample was taken and detected using a solid-state nanopore chip. The small extracellular vesicle was driven to pass through the nanopore by applying voltage, and the nanopore ion current signal was collected. The ion current signal was filtered, baseline corrected and detrended, and single particle passing through the pore event was identified. The equivalent charge deficit, current blocking amplitude, half-dwell time, left edge slope, and right edge slope are extracted from each via event and used as a training set. This set is then input into a machine learning classification model for training. The model outputs a score for each input event to be classified as a target small cell extracellular vesicle event. After training, a machine learning prediction model is obtained.

[0025] Optionally, the machine learning classification model includes a gradient boosting decision tree model (more preferably an XGBoost model), which outputs a continuous numerical score between 0 and 1.

[0026] Optionally, the criterion for determining an input event as a target small cell extracellular vesicle is: the score is higher than a preset threshold.

[0027] Optionally, the preset threshold is selected from 0.05 to 0.3.

[0028] Optionally, the criterion for determining whether a sample to be tested is a specific sample type is: the proportion of specific small extracellular vesicles is higher than a threshold, wherein the threshold is selected from 1% to 10%.

[0029] In a second aspect, the present invention provides a small extracellular vesicle detection system, the system comprising a nanopore detection unit, an ion current acquisition unit, a data processing unit, and a machine learning discrimination unit; The nanopore detection unit is used to perform the following: detecting a sample to be tested using a solid-state nanopore chip, and driving the sample to be tested through the nanopore by applying a voltage; The ion current acquisition unit is used to perform the following: acquiring ion current signals from nanopores; The data processing unit is used to perform the following: filtering, baseline correction and detrending processing of the ion current signal, and identifying single-particle via events; extracting the equivalent charge defect, current blocking amplitude, half-dwell time, left edge slope and right edge slope from each via event; The machine learning discrimination unit is used to perform the following: inputting the data obtained by the data processing unit into the machine learning prediction model, using the machine learning prediction model to calculate the score of the output via event as a target small extracellular vesicle event; calculating the proportion of via events with scores higher than a preset threshold in the total via events, and determining whether the sample to be tested is a specific sample type based on the proportion.

[0030] Optionally, the system is used to perform the steps of the small extracellular vesicle detection method described in the first aspect.

[0031] Optionally, the small extracellular vesicles include cancer-associated small extracellular vesicles.

[0032] Optionally, the cancer includes, but is not limited to, hepatocellular carcinoma, pancreatic cancer, or colorectal cancer.

[0033] The detection system of this invention does not rely solely on the discrimination result of a single event. Instead, it is based on statistical logic to calculate the proportion of tumor-related small extracellular vesicle perforation events in the test sample to the total number of perforation events. This proportion is used as a sample-level diagnostic indicator. By setting reasonable scoring thresholds and proportion judgment criteria, reliable differentiation between cancer samples and non-cancer samples can be achieved.

[0034] Optionally, the nanopore detection unit may further include a solid-state nanopore chip, a fluidic cell assembly, and an Ag / AgCl electrode.

[0035] Compared with the prior art, the present invention has at least the following beneficial effects: (1) By introducing a particle-driven mechanism dominated by electroosmotic flow, this invention significantly improves the capture efficiency of small extracellular vesicles with low surface potential, and solves the problem of insufficient number of effective events under traditional electrophoresis-driven conditions. (2) The present invention adopts a single-particle electrical signal analysis strategy of multi-feature fusion and combines it with a machine learning model for probability discrimination. Compared with the traditional analysis method that only relies on a single blocking amplitude or residence time, it significantly improves the ability to distinguish tumor-related small extracellular vesicles in the complex background of plasma. (3) The present invention does not require exosome enrichment, immune capture or fluorescent labeling of plasma samples. The detection process is simple, short and low cost. It also avoids the bias introduced by labeling or enrichment steps and helps to maintain the natural state of small extracellular vesicles. (4) The detection system of the present invention is based on the statistical construction of sample-level diagnostic indicators at the single-particle level, which can maintain good stability and reproducibility among different samples and is suitable for application scenarios such as auxiliary diagnosis, screening and efficacy monitoring of cancer. (5) The method and system provided by the present invention have good scalability and can be adapted to different cancer types or different biological fluid samples through model retraining or feature expansion, and have broad clinical application prospects. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the overall process of the tumor-associated small extracellular vesicle detection method based on solid nanopores and machine learning of the present invention. It shows the overall process of sample introduction, particle perforation dominated by electroosmotic flow, ion current signal acquisition, single particle event identification, multi-feature extraction and sample-level diagnosis and discrimination. Among them, Figure A is the schematic diagram of nanopore sensing principle, and Figure B is the flowchart of small extracellular vesicle perforation detection and signal analysis under the action of electroosmotic flow. Figure 2 This diagram illustrates the principle and physical model of nanopore detection, showing the structure of solid nanopores and their series resistance model. Figure 3 This diagram illustrates the comparison of electron microscopy characterization and perforation signals of different types of nanoparticles in plasma. Figure A shows a transmission electron microscope image of small extracellular vesicles; Figure B shows the principle of nanopore detection of small extracellular vesicles; Figure C shows the current trajectory of the detection signal of small extracellular vesicles; Figure D shows a transmission electron microscope image of other particles in plasma (including serum albumin, high-density lipoprotein, low-density lipoprotein, and very low-density lipoprotein); and Figure E shows the current trajectory of the detection signal of other particles in plasma (including serum albumin, high-density lipoprotein, low-density lipoprotein, and very low-density lipoprotein). Figure 4 This diagram illustrates the feature extraction and machine learning analysis of single-particle perforation signals, used to distinguish between perforation signals of small extracellular vesicles and very low-density lipoproteins. It shows features extracted from perforation events, including equivalent charge deficit, current blocking amplitude, half-dwell time, left edge slope, and right edge slope, as well as the machine learning model built based on these features and its feature contribution analysis results. Figure A shows the features extracted from perforation events, including equivalent charge deficit, current blocking amplitude, half-dwell time, left edge slope, and right edge slope. Figure B shows the results of each feature in the very low-density lipoprotein and small extracellular vesicle groups, showing significant differences. Figure C shows the contribution of each feature in the XGBoost model, with equivalent charge deficit contributing the most and right edge slope contributing the least. Figure D shows the small extracellular vesicle prediction score output by the XGBoost model, showing a significant difference between the two groups. Figure 5This diagram illustrates the quantitative analysis results of tumor-associated small extracellular vesicles in a simulated plasma system. It shows the detection results, event distribution characteristics, and clustering and dimensionality reduction analysis results after introducing different concentrations of tumor cell line-derived small extracellular vesicles into a healthy plasma background. Figure A shows the nanopore detection current trajectory signal results of nanoparticles and tumor-derived small extracellular vesicles in healthy human plasma. Figure B shows the contribution of each feature to the XGBoost model, with equivalent charge depletion contributing the most and the left-side slope contributing the least. Figure C shows the tumor-derived small extracellular vesicle prediction score output by the XGBoost model. The two groups show significant differences. There are significant differences. Figure D shows the results of nanopore detection current trajectory signals in plasma samples with different concentrations of tumor-derived extracellular vesicles. Figure E shows the tumor-derived extracellular vesicle prediction scores output by the XGBoost model in plasma samples with different concentrations of tumor-derived extracellular vesicles. Figure F shows that the tumor-related signal is linearly correlated with the concentration of tumor-derived extracellular vesicles. Figure G shows the clustering results of tumor-derived extracellular vesicles and healthy plasma samples, which show obvious clustering. Figure H shows the results of each feature in the healthy plasma and tumor-derived extracellular vesicle groups, which show significant differences. Figure 6 This is a schematic diagram of the clinical sample diagnostic analysis results, showing the distribution differences of cancer samples and non-cancer samples in terms of sample-level diagnostic indicators, receiver operating characteristic curves, and classification performance evaluation results. Figure A shows the tumor-derived small extracellular vesicle prediction scores and the proportion of vesicles identified as tumor-related small extracellular vesicles output by the XGBoost model for five healthy individuals and five hepatocellular carcinoma patients. Figure B shows the heatmap of the proportion of tumor-related small extracellular vesicles detected in 10 pancreatic cancer patients, 10 hepatocellular carcinoma patients, 10 colorectal cancer patients, 10 colorectal polyp patients, and 15 healthy individuals. Figure C shows the proportion of vesicles in 10 patients... Violin plots were used to detect the proportion of tumor-associated small extracellular vesicles in pancreatic cancer patients, 10 hepatocellular carcinoma patients, 10 colorectal cancer patients, 10 colorectal polyp patients, and 15 healthy individuals. The results showed significant differences between the cancer and non-cancer groups. Figure D shows the ROC curves of tumor-associated small extracellular vesicle signal results in the cancer and non-cancer groups, with an AUC value of 0.9527. Figure E shows the tumor-associated small extracellular vesicle signal results in the cancer and non-cancer groups, showing significant differences. Figure F shows the confusion matrix of tumor-associated small extracellular vesicle signal results in the cancer and non-cancer groups, with a prediction accuracy of 87.27%. Detailed Implementation

[0037] To further illustrate the technical means and effects of this invention, the technical solution of this invention will be further explained below with reference to the accompanying drawings and specific embodiments. However, the following examples are merely simplified examples of this invention and do not represent or limit the scope of protection of this invention. The scope of protection of this invention is determined by the claims.

[0038] Where specific techniques or conditions are not specified in the examples, they shall be performed in accordance with the techniques or conditions described in the literature in this field, or in accordance with the product instructions. Reagents or instruments whose manufacturers are not specified are all conventional products that can be purchased from legitimate channels.

[0039] This invention provides a method for detecting tumor-associated small extracellular vesicles based on solid-state nanopores and machine learning. A flowchart is shown below. Figure 1 As shown, the plasma sample to be tested is added to the detection cell. Under the action of an applied voltage, electroosmotic flow acts as the dominant driving force to push the nanoparticles in the sample through solid nanopores. Figure 1 (Figure A); Ion current signals are collected during particle perforation, the signals are preprocessed and single-particle perforation events are identified; multidimensional electrical and morphological features are extracted for each perforation event, and event-level discrimination results are obtained based on a machine learning model; finally, sample-level diagnostic discrimination is achieved by statistically analyzing the proportion of cancer-related events. Figure 1 (Figure B in the middle)

[0040] Example 1 This embodiment describes the construction of a small extracellular vesicle detection method based on electroosmotic flow.

[0041] This embodiment provides a method for single-particle electrical detection of small extracellular vesicles in plasma samples that have not undergone exosome enrichment or labeling. The method is based on the principle of solid-state nanopore electrical sensing and improves the permeability efficiency of small extracellular vesicles in the complex plasma matrix by introducing a particle-driven mechanism dominated by electroosmotic flow under specific conditions, thereby obtaining stable and analyzable single-particle permeability event signals.

[0042] In this embodiment, solid silicon nitride (SiN) is selected. x A nanopore chip (purchased from Norcada, NXPR5002Y-C0.20-AO-HR) is used as the core component for electrical detection. The nanopore chip consists of a thin film substrate with a single nanoscale via formed in the central region of the film. Both the film thickness and the pore size of the nanopore are in the nanoscale range, allowing small extracellular vesicles to pass through smoothly while generating clearly distinguishable blockage events in the ion current signal. The electrical equivalent model of the nanopore and its relationship with the ion current signal are described below. Figure 2As shown, by treating the nanopore as an equivalent series resistance model consisting of the internal resistance and the access resistance, the generation mechanism of the current blocking signal during the nanoparticle passage through the pore can be quantitatively understood.

[0043] Before detection, the nanoporous chip undergoes surface treatment to improve its hydrophilicity and electrochemical stability. This treatment employs a strong oxidation system to clean and activate the chip surface, creating a stable hydrophilic surface state on the inner walls of the nanopores. This reduces signal instability caused by surface contamination or bubble adsorption during detection. The nanoporous chip treated in this way exhibits stable open-pore current and low background noise levels in subsequent detection, providing a fundamental basis for reliable acquisition of single-particle signals.

[0044] During the detection process, a neutral electrolyte buffer solution is added to both sides of the nanopore to form an ion conduction channel. By applying an external voltage across the nanopore, ions in the solution migrate directionally under the influence of the electric field, thereby establishing a stable ion current within the nanopore. Since the inner wall of the silicon nitride nanopore carries a surface charge under neutral conditions, an electrical bilayer structure forms near the pore wall. When an external electric field is applied, the counterion layer in the electrical bilayer migrates as a whole, thus forming an electroosmotic flow (EOF) within the nanopore.

[0045] In this embodiment, by selecting an appropriate voltage polarity, the direction of the resulting electroosmotic flow is aligned with the transport direction of the target small extracellular vesicles, thus making electroosmotic flow the dominant driving force for nanoparticle permeation through the nanopores. Compared to traditional nanopore detection conditions that rely solely on electrophoretic forces, this driving method significantly reduces the dependence on particle surface potential, making it more conducive to propelling small extracellular vesicles with lower surface charges through the nanopores. Experimental results show that under electroosmotic flow-dominated conditions, the frequency of permeation events of small extracellular vesicles in the nanopores is significantly increased, thereby significantly increasing the number of effective events available for analysis.

[0046] In this embodiment, the samples tested included healthy donor plasma, plasma samples after ultracentrifugation to remove extracellular vesicles, and small extracellular vesicle samples derived from tumor cell lines. All of these different sample types underwent nanopore measurements under the same detection conditions to allow for a systematic comparison of their permeability. Typical ion current signals and event distribution characteristics of different samples in nanopore detection are shown below. Figure 3 As shown. Specifically, the experimental procedure includes: (1) Cell culture: HepG2 cells are adherent cells. The cell culture medium was prepared with 10% fetal bovine serum, 1% antibiotics and DMEM high glucose medium. All cells were cultured in an incubator at 37°C, 5% CO2 and saturated humidity.

[0047] (2) Extraction of extracellular vesicles: After culturing cells in serum-free medium for 24 h, when the cells proliferate to the point of covering the bottom of the dish to 60%, the medium is collected into centrifuge tubes and subjected to multiple centrifugation steps (4℃, 300 rcf, centrifugation for 10 min, removing dead cells, transferring the supernatant to a new centrifuge tube; 4℃, 2000 rcf, centrifugation for 10 min, removing cell debris, transferring the supernatant to a new centrifuge tube; 4℃, 10000 rcf, centrifugation for 30 min, removing large vesicles, transferring the supernatant to a centrifuge tube with an angle rotor for an ultracentrifuge, ensuring no air bubbles are present in the tube and the tube opening is strictly sealed; 4℃, 120000 rcf, centrifugation for 120 min, discarding the supernatant, using PBS to agitate the tube wall, resuspending the sEVs, and transferring the sEVs suspension to a centrifuge tube with a flat rotor for an ultracentrifuge; 4℃, 120000 rcf, centrifugation for 120 min). After extraction, discard the supernatant, and resuspend the sEVs by blowing the bottom of the tube with PBS. Once the sEVs are extracted, they can be directly used for subsequent experiments, or they can be aliquoted and frozen for storage. Freeze the sEVs suspension in the PBS system directly to -80°C, using small aliquots to avoid repeated freeze-thaw cycles. (3) TEM sample preparation of extracellular vesicles: Add an equal volume of 4% paraformaldehyde to the freshly extracted sEVs suspension, shake on a rotary mixer for 20 min, and then proceed with TEM sample preparation. Take 10 μL of the above suspension and drop it onto a 400-mesh copper mesh (carbon support membrane). After standing for 20 min, blot the liquid with filter paper, add another 10 μL of the above suspension, and let it stand for 20 min. Rinse the copper mesh 3 times with PBS, then rinse 5 times with deionized water, and blot the liquid with filter paper. Add 10 μL of 3% phosphotungstic acid negative staining solution, let it stand for 10 min, and blot the liquid with filter paper. After air drying, store in a desiccant oven away from light.

[0048] (4) Detection of nanopores in extracellular vesicles To improve the hydrophilicity of the nanopores and prevent clogging, the silicon nitride nanopores were placed in a "piranha" solution (concentrated sulfuric acid and 30% hydrogen peroxide, volume ratio 3:1), heated in a 95°C water bath for 120 min, rinsed 3-5 times with deionized water, and rinsed once with PBS buffer. The cleaned nanopores were then placed in a fluidization chamber, sealed with gaskets to prevent leakage. After securing the nanopores, PBS buffer was added to one side of the fluidization chamber, and a PBS-diluted sEVs suspension was added to the other side and agitated thoroughly. Ag / AgCl electrodes were then inserted on both sides, and the chamber was covered with a shielding box. The detection signals were recorded on a computer display system. The entire experiment was conducted in a Faraday cage. The sampling frequency was 90–110 kHz, and the filtering frequency was 8–12 kHz. All experimental data were processed and extracted using Clampfit software.

[0049] likeFigure 3 As shown in the figures, Figure A is a transmission electron microscope (TEM) image of small extracellular vesicles; Figure B is a schematic diagram of the nanopore detection method for small extracellular vesicles; Figure C is the current trajectory of the detection signal for small extracellular vesicles; Figure D is a TEM image of other particles in plasma (including serum albumin, high-density lipoprotein, low-density lipoprotein, and very low-density lipoprotein); and Figure E is the current trajectory of the detection signal for other particles in plasma (including serum albumin, high-density lipoprotein, low-density lipoprotein, and very low-density lipoprotein). Compared with healthy plasma samples, the number and distribution characteristics of pore events in plasma samples with extracellular vesicles removed changed significantly, indicating the presence of other nanoscale particles in plasma capable of generating current-blocking signals. Furthermore, pore events generated by small extracellular vesicle samples derived from tumor cell lines under the same detection conditions showed systematic differences from background plasma particles in terms of blocking amplitude, duration, and signal morphology. This result demonstrates that in the complex background of plasma, reliable differentiation of small extracellular vesicles cannot be achieved solely through the number of events or a single blocking parameter.

[0050] During the detection process, the signal of ion current change in the nanopore over time was continuously recorded. When nanoparticles pass through the nanopore under electroosmotic flow-dominated conditions, they cause transient perturbations to the ion channels within the pore, which manifest as brief current blocking events in the current time series. Each blocking event corresponds to a single-particle passage process, and the amplitude, duration, and morphology of the blocking signal are closely related to the physical properties of the particle and the through-pore dynamics. The differences in the statistical characteristics of through-pore events from particles of different origins provide direct experimental evidence for multi-feature extraction and machine learning-based discrimination in subsequent embodiments.

[0051] In summary, this embodiment achieved stable perforation of small extracellular vesicles and acquisition of single-particle electrical signals in plasma samples without exosome enrichment or labeling by constructing solid-state nanopore detection conditions dominated by electroosmotic flow. Furthermore, through systematic comparison with different control samples, it revealed the differences in perforation event behavior in complex plasma backgrounds, laying an experimental foundation for subsequent discriminant analysis based on multiple features and machine learning.

[0052] Example 2 This embodiment performs signal preprocessing, multi-feature extraction, and statistical discrimination characterization of single-particle through-hole events.

[0053] Based on the nanopore ion current time series obtained in Example 1 under the condition of electroosmotic flow dominance, this example further provides a signal processing and multi-feature extraction method for single-particle perforation events. This method is used to stably and repeatedly extract multi-dimensional features that reflect the differences in nanoparticle perforation behavior from the raw electrical signals in a complex plasma background, providing reliable input for subsequent event-level discrimination based on machine learning.

[0054] In actual testing, plasma samples are characterized by complex composition, high viscosity, and the presence of multiple dissolved or suspended components. Therefore, the directly acquired ion current signals typically contain various noise components and slowly changing baseline drift. Consequently, preprocessing of the raw current signals is necessary before identifying and analyzing single-particle perforation events to improve the accuracy of event identification and the stability of feature extraction.

[0055] In this embodiment, the original ion current time series is first digitally filtered to suppress high-frequency random noise components. The filtered signal significantly reduces the background noise level while maintaining the main morphological characteristics of the via event. Subsequently, baseline correction and detrending processing are performed on the filtered signal to eliminate low-frequency baseline drift caused by temperature changes, electrode polarization, or plasma matrix. Through the above preprocessing steps, a current time series with a stable baseline and a high signal-to-noise ratio can be obtained, laying the foundation for subsequent event identification.

[0056] In the preprocessed current signal, a threshold-based event detection method is used to identify single-particle via events. Specifically, by analyzing the noise statistical characteristics of the signal within a local time window, a current change threshold adapted to the background noise level is set. When the current signal momentarily deviates from the baseline and exceeds the threshold, it is determined as the start of a via event; when the signal recovers to near the baseline, the event is determined as the end of the event. This event recognition method can adapt to changes in noise levels under different samples and detection conditions, thus maintaining good event recognition stability in the complex background of plasma.

[0057] In actual testing, the identified via events include both simple single-peak blocking events and complex blocking events with multiple local extrema. For the latter, this embodiment treats it as a continuous blocking behavior generated by a single nanoparticle during its passage through the nanopore, rather than multiple independent particle events, thereby avoiding repeated counting of the same particle.

[0058] After completing via event identification, multiple complementary electrical and morphological features are extracted for each event. Specifically, this embodiment extracts at least the following five types of features from each via event: equivalent charge deficit (ECD), current blocking amplitude (...). ΔI The half-dwell time, the left edge slope of the event (Left_k1), and the right edge slope of the event (Right_k2) are also mentioned.

[0059] The equivalent charge defect (ECD) is defined as the integral area of ​​the current blocking curve relative to the baseline during a via event. This feature comprehensively reflects the overall effect of particle residence time and blocking intensity within the pore, and can characterize the degree of perturbation of ion transport within the pore by nanoparticles. Compared to simple blocking amplitude or residence time, ECD is more robust to changes in signal morphology.

[0060] Current blocking amplitude ΔI Defined as the maximum change in current relative to the baseline during a via event, this characteristic is primarily related to the effective volume of the nanoparticle and its degree of blockage within the nanopore. The half-dwell time characterizes the kinetics of particle residence within the nanopore, reflecting the overall transport timescale of the particle during the via process.

[0061] Furthermore, to further characterize the morphological features of the via event signal, this embodiment introduces the left-side slope Left_k1 and the right-side slope Right_k2. Left_k1 describes the rate of current change as the current signal drops from the baseline to the blocking extreme, while Right_k2 describes the rate of current change as the current signal recovers from the blocking extreme to the baseline. These two features can reflect the kinetic asymmetry and interfacial interaction characteristics when particles enter and leave the nanopore.

[0062] By jointly extracting the above five features, single-particle perforation events can be characterized from multiple dimensions, including blockage intensity, duration, overall perturbation effect, and signal morphology. Experimental results show that nanoparticles from different sources exhibit different statistical distribution characteristics in the above feature space.

[0063] Based on the aforementioned feature extraction method, statistical analysis of perforation events in plasma and control samples revealed differences in the distribution of different particle types within a multidimensional feature space. Specifically, the experimental procedure included: placing silicon nitride nanopores in a "piranha" solution (concentrated sulfuric acid to 30% hydrogen peroxide, volume ratio 3:1), heating in a 95°C water bath for 120 min, washing 3-5 times with deionized water, and rinsing once with PBS buffer. The cleaned nanopores were then placed in a fluidized bed and sealed with gaskets to prevent leakage. After fixing the nanopores, PBS buffer was added to one side of the fluidized bed, and a PBS-diluted low-density lipoprotein suspension was added to the other side and agitated thoroughly. Ag / AgCl electrodes were then inserted on both sides, the box was covered, and the detection signal was recorded on a computer display system. The entire experiment was conducted in a Faraday cage. The sampling frequency was 90–110 kHz, and the filtering frequency was 8–12 kHz. Experimental data were processed and extracted using Clampfit software. The Extreme Gradient Boosting (XGBoost) algorithm was used to implement machine learning-based classification modeling. Five quantitative nanopore signal parameters were extracted from a single translocation event using MATLAB scripts: equivalent charge deficit (ECD), blocking amplitude, half-retention time, left edge slope (Left_k1), and right edge slope (Right_k2). The dataset was randomly divided into a training set (80%) and a test set (20%). The study optimized model hyperparameters through cross-validation to minimize mean squared error and maximize classification accuracy. Shapley additive interpretation (SHAP) analysis was used to evaluate feature importance and model interpretability, clarifying the contribution of each feature to the model output. Very low density lipoprotein events and extracellular vesicle events were identified by calculating classification probabilities.

[0064] Relevant results are as follows Figure 4 As shown in the figure, the system demonstrates the results of the analysis of via events based on the five features, as well as the role of different features in distinguishing tumor-associated small extracellular vesicles from plasma background particles. Figure A shows the features extracted from via events, including equivalent charge depletion, current blocking amplitude, half-dwell time, left edge slope, and right edge slope. Figure B shows the results of each feature in the very low density lipoprotein and small extracellular vesicle groups, showing significant differences. Figure C shows the contribution of each feature to the XGBoost model, with equivalent charge depletion contributing the most and right edge slope contributing the least. Figure D shows the small extracellular vesicle prediction score output by the XGBoost model, showing significant differences between the two groups.

[0065] like Figure 4As shown, when using only a single feature, there is a large overlap in the feature distribution of different particle types, making reliable differentiation difficult. However, when multiple features are used in combination, the distribution differences of particles from different sources in the multidimensional feature space become more significant. This result indicates that multi-feature fusion can significantly improve the ability to distinguish different nanoparticle types against a complex plasma background.

[0066] Furthermore, by analyzing the contribution of each feature to the model, it can be found that different features have different importance in distinguishing tumor-associated small extracellular vesicles. This analysis shows that the discrimination of perforation events does not depend on a single physical parameter, but is determined by a combination of multiple electrical and kinetic characteristics, thus providing a reasonable physical and statistical basis for subsequent event-level discrimination based on machine learning.

[0067] In summary, this embodiment achieves high-dimensional characterization of single-particle perforation events by systematically preprocessing the nanopore ion current signal, stably identifying events, and extracting multidimensional features; and demonstrates this through experimental results (such as...). Figure 4 As shown, the effectiveness of multi-feature joint analysis in distinguishing nanoparticles from different sources in complex plasma backgrounds was verified, providing a direct data foundation and theoretical basis for introducing machine learning-based event-level discrimination and sample-level detection methods in subsequent embodiments.

[0068] Example 3 This embodiment constructs and verifies a machine learning-based method for via event-level discrimination and sample-level detection.

[0069] 1. Collection and preparation of clinical data and samples A retrospective analysis was conducted, collecting samples from patients with hepatocellular carcinoma (HCC), pancreatic cancer (PC), colorectal cancer (CRC), colorectal polyps (Polyp), and healthy donors (HD).

[0070] All cancer patients met the following inclusion criteria: a clear pathological diagnosis and Edmondson staging of the tumor; complete basic information for all cases; and exclusion of the following: pregnant patients, germ cell tumors, malignant tumors in other organs, severe infectious diseases, and severe diseases of other vital organs (such as heart, lung, kidney, etc.).

[0071] Patients with colorectal polyps must meet the following inclusion criteria: a confirmed pathological diagnosis of colorectal polyps; no history of any malignant tumor-related diseases within the past 12 months; complete basic information; and exclusion of the following: pregnant patients, patients with severe infectious diseases, and patients with serious diseases of other vital organs (such as the heart, lungs, and kidneys).

[0072] Healthy donors must meet the following selection criteria: no malignant tumor-related diseases within the past 12 months; all basic information must be complete; and the following conditions must be excluded: pregnant patients, patients with serious infectious diseases, and patients with serious diseases of other vital organs (such as heart, lung, kidney, etc.).

[0073] Collect 5 mL of peripheral venous blood from all subjects in EDTA vacuum anticoagulation blood collection tubes. After sampling, invert the sampling tubes 5-6 times, transport them at 4°C, and process and use them within 24 hours. Alternatively, after routine centrifugation, freeze the plasma at -80°C and use it within 3 years.

[0074] 2. In Example 2, a feature set capable of characterizing the pore behavior of nanoparticles from multiple dimensions was obtained by systematically preprocessing and extracting multiple features from single-particle pore events. Building upon this, this example further provides a machine learning-based event-level discrimination and sample-level detection method for statistical identification and sample-level differentiation of tumor-related small extracellular vesicles in complex plasma backgrounds.

[0075] In this embodiment, the multidimensional features corresponding to each via event are used as independent data samples to construct an event-level discrimination problem. Specifically, for each via event, there is a corresponding feature consisting of the equivalent charge deficit (ECD), current blocking amplitude (ECD), and multidimensional features (ECD). ΔI) The eigenvector is composed of the half-dwell time, the left-side slope (Left_k1), and the right-side slope (Right_k2). This eigenvector comprehensively reflects the volume effect, dynamic behavior, and signal morphology characteristics of a single particle during its passage through the nanopore.

[0076] Based on the aforementioned feature vectors, a supervised learning method is used to classify and distinguish via events. In this embodiment, the classification model used is a machine learning model based on gradient boosting decision trees. This model, by constructing multiple decision trees and weighting their results, can effectively handle nonlinear relationships between different features, while also possessing strong generalization ability and robustness to changes in feature scale.

[0077] During model training, via event features from different types of samples are used as input data, and the events are labeled according to their source. The dataset is partitioned to construct separate subsets for model training and performance evaluation. After training, the model outputs a score for each input event, indicating whether it belongs to a tumor-related small extracellular vesicle event. This score is a continuous numerical value reflecting the model's confidence in its classification of a single event.

[0078] Unlike traditional hard-threshold-based classification methods, this embodiment does not directly use the classification result of a single event as the final judgment criterion. Instead, it further introduces a statistical analysis strategy to perform a holistic analysis of all via events in the same sample. Specifically, by setting a scoring threshold, events with scores higher than the threshold are classified as cancer-related events, while the remaining events are classified as non-cancer-related events. Subsequently, the proportion of cancer-related events in all via events in the sample is calculated, and this proportion is used as a sample-level diagnostic indicator.

[0079] By employing the aforementioned event-level discrimination and sample-level statistical methods, the impact of individual event discrimination errors on the final diagnostic results can be eliminated to some extent, thereby improving the stability and reliability of sample-level discrimination. This sample-level index not only reflects the relative abundance of tumor-associated small extracellular vesicles in the sample but also implies their overall electrical behavior characteristics in a complex plasma background.

[0080] In this embodiment, the above method was first validated in a simulated plasma system. Specifically, different concentrations of small extracellular vesicles derived from tumor cell lines were introduced into a healthy donor plasma background to construct a series of spiked samples. Nanopore detection was performed on these samples, and analysis was conducted according to the event-level discrimination and sample-level statistical methods described above. The experimental procedure included: mixing purified small extracellular vesicles derived from human hepatocellular carcinoma HepG2 cells with plasma from healthy subjects to prepare artificial plasma samples, thereby simulating different clinical concentration conditions. The extracellular vesicle suspension was serially diluted and added to 200 μL of plasma from healthy subjects to obtain a final concentration of 1×10⁻⁶. 4 2×10 3 1×10 3 2×10 2 Samples were prepared at a concentration of 100 vesicles / μL. The mixture was gently vortexed and equilibrated at room temperature for 15 min before nanopore detection experiments were performed. Plasma from healthy subjects was ultracentrifuged at 100,000 g for 10 h to prepare extracellular vesicle-free plasma, which served as a blank control.

[0081] The silicon nitride nanopores were placed in a "piranha" solution (concentrated sulfuric acid and 30% hydrogen peroxide, volume ratio 3:1) and heated in a 95°C water bath for 120 min. They were then rinsed 3-5 times with deionized water and once with PBS buffer. The cleaned nanopores were placed in a fluidized bed and sealed with gaskets to prevent leakage. After securing the nanopores, PBS buffer was added to one side of the fluidized bed, and the PBS-diluted sample suspension was added to the other side and agitated thoroughly. Ag / AgCl electrodes were then inserted into both sides, and the shielding box was closed. The detection signal was recorded on a computer display system. The entire experiment was conducted in a Faraday cage. The sampling frequency was 90–110 kHz, and the filtering frequency was 8–12 kHz.

[0082] Experimental data were processed and extracted using Clampfit software. The Limiting Gradient Boosting (XGBoost) algorithm was employed to achieve machine learning-based classification modeling. Five quantitative nanopore signal parameters were extracted from a single translocation event: equivalent charge deficit (ECD), blocking amplitude, half-retention time, left edge slope (Left_k1), and right edge slope (Right_k2). The dataset was randomly divided into a training set (80%) and a test set (20%). The study optimized model hyperparameters through cross-validation to minimize mean squared error and maximize classification accuracy. Shapley additive interpretation (SHAP) analysis was used to evaluate feature importance and model interpretability, clarifying the contribution of each feature to the model output. Extracellular vesicle events associated with cancer in plasma samples were identified by calculating classification probabilities.

[0083] The analytical results of the above simulated plasma spiking experiment are as follows: Figure 5 As shown, Figure 5 This paper presents the changes in the proportion of cancer-related events at the sample level under different spiking concentrations, as well as the statistical analysis results based on the distribution of event characteristics. Figure A shows the nanopore detection current trajectory signals of plasma nanoparticles from healthy individuals and tumor-derived small extracellular vesicles. Figure B shows the contribution of each feature to the XGBoost model, with equivalent charge depletion contributing the most and the left-side slope contributing the least. Figure C shows the prediction scores of tumor-derived small extracellular vesicles output by the XGBoost model, showing a significant difference between the two groups. Figure D shows the nanopore detection current trajectory signals of plasma samples with different concentrations of tumor-derived small extracellular vesicles. Figure E shows the nanopore detection current trajectory signals of plasma samples with different concentrations of tumor-derived small extracellular vesicles. The XGBoost model output of tumor-derived extracellular vesicle (EGVV) prediction scores from plasma samples containing EGVV were used. Figure F shows a linear correlation between tumor-related signals and the concentration of EGVVV. Figure G shows the clustering results of EGVVV and healthy plasma samples, indicating clear grouping. Figure H shows the results of each feature in the healthy plasma and tumor-derived EGVV groups, showing significant differences. Experimental results indicate that with increasing EGVV concentration, the proportion of samples identified as cancer-related events shows a significant trend, suggesting that the sample-level diagnostic indicators can reflect changes in the relative abundance of EGVV in plasma. Figure 5 It can be seen that the ratio index has good response characteristics and repeatability within a certain concentration range, indicating that the method used in this embodiment is not only suitable for qualitatively distinguishing different samples, but also has a certain semi-quantitative analysis capability.

[0084] After validating the simulated plasma system, the method of this embodiment was further applied to the analysis of real clinical plasma samples. The clinical samples included cancer patient samples and non-cancer control samples, which could include healthy donor samples or benign lesion samples. Nanopore electrical measurements were performed on all samples under the same detection conditions, and the obtained perforation event data were uniformly processed and analyzed.

[0085] Based on the aforementioned event-level discrimination and sample-level statistical methods, the proportion of cancer-related events in different clinical samples was compared and analyzed. Samples with a cancer-related event proportion greater than or equal to 1%~10% (the specific value is related to the sample type) were classified as cancer samples, while samples with a cancer-related event proportion less than 1%~10% (the specific value is related to the sample type) were classified as non-cancer samples. The experimental procedure included: (1) Processing of clinical blood samples: Peripheral blood at 4°C was placed at room temperature for 15 min and then transferred to a 15 mL centrifuge tube; at room temperature, centrifuged at 300 rcf for 15 min and the upper plasma was transferred to a new tube; at 4°C, centrifuged at 2500 rcf for 15 min and then repeated once; after centrifugation, the plasma was aliquoted, labeled and frozen in a -80°C freezer for 3 years.

[0086] (2) Extraction of plasma sEVs: Remove plasma from the -80℃ freezer, heat in a 37℃ water bath until completely thawed, remove the tube, wipe the water stains off the outside of the tube, and transfer on ice. Centrifuge at 10,000 rcf for 10 min at 4℃. Transfer the centrifuged plasma to a new tube, dilute 50 times with PBS, mix thoroughly, and transfer the mixture to a centrifuge tube with an angle rotor for an ultracentrifuge. Ensure there are no air bubbles in the tube and strictly seal the tube opening. Centrifuge at 120,000 rcf for 15 h at 4℃, discard the supernatant, use PBS to agitate the tube wall, resuspend the sEVs, and transfer the sEVs suspension to a centrifuge tube with a flat rotor for an ultracentrifuge. Centrifuge at 120,000 rcf for 2 h at 4℃, discard the supernatant, use PBS to agitate the bottom of the tube, and resuspend the sEVs. After plasma sEVs are extracted, they can be directly used for subsequent experiments, or they can be aliquoted and frozen for storage. The sEVs suspension in the PBS system can be directly frozen to -80°C and aliquoted in small volumes to avoid repeated freeze-thaw cycles.

[0087] (3) Nanoporous single-particle detection of plasma sEVs: To improve the hydrophilicity of the nanopores and avoid pore blockage, the silicon nitride nanopores were placed in a "piranha" solution (concentrated sulfuric acid and 30% hydrogen peroxide in a volume ratio of 3:1), heated in a 95°C water bath for 120 min, washed 3-5 times with deionized water, and rinsed once with PBS buffer. The cleaned nanopores were placed in a fluid cell and sealed with gaskets to prevent leakage. After fixing the nanopores, PBS buffer was added to one side of the fluid cell, and plasma diluted with PBS was added to the other side and agitated. Ag / AgCl electrodes were inserted on both sides, the shielding box was covered, and the detection signal was recorded in the computer display system. The entire experiment was carried out in a Faraday cage. All plasma sEVs assays were performed using an applied voltage of 100 mV (stable baselines could not be produced when voltages of 200 mV, 300 mV, and 400 mV were applied, presumably due to the complex biological components in real plasma; therefore, the 100 mV voltage condition, which provided a more stable baseline, was ultimately selected). The entire experiment was conducted in a Faraday cage. The sampling frequency was 90–110 kHz, and the filtering frequency was 8–12 kHz.

[0088] Experimental data were processed and extracted using Clampfit software. The Limiting Gradient Boosting (XGBoost) algorithm was employed to implement machine learning-based classification modeling. Five quantitative nanopore signal parameters were extracted from a single translocation event using MATLAB scripts: Equivalent Charge Defect (ECD), blocking amplitude, half-retention time, left edge slope (Left_k1), and right edge slope (Right_k2). The dataset was randomly divided into a training set (80%) and a test set (20%). The study optimized model hyperparameters through cross-validation to minimize mean squared error and maximize classification accuracy. Shapley additive interpretation (SHAP) analysis was used to evaluate feature importance and model interpretability, clarifying the contribution of each feature to the model output. Extracellular vesicle events associated with cancer in plasma samples were identified by calculating classification probabilities.

[0089] The results of the correlation analysis are as follows Figure 6 As shown, Figure 6The system demonstrates the results of distinguishing between cancer and non-cancer samples based on sample-level diagnostic indicators, including distributional differences between different sample groups and corresponding diagnostic performance assessments. Figure A shows the tumor-derived small extracellular vesicle (SMEV) prediction scores and the proportion of SMEV identified as tumor-associated SMEVs output by the XGBoost model for five healthy individuals and five hepatocellular carcinoma (HCC) patients. Figure B shows a heatmap of the proportion of tumor-associated SMEVs (HD) detected in 10 pancreatic cancer (PC) patients, 10 HCC patients, 10 colorectal cancer (CRC) patients, 10 polyp (colorectal) patients, and 15 healthy individuals. Figure C shows the proportion of SMEVs detected in 10 pancreatic cancer patients... Violin plots were used to detect the proportion of tumor-associated small extracellular vesicles in 10 patients with hepatocellular carcinoma, 10 patients with colorectal cancer, 10 patients with colorectal polyps, and 15 healthy individuals. The results showed significant differences between the cancer group and the non-cancer group. Figure D shows the ROC curves of tumor-associated small extracellular vesicle signal results in the cancer group and the non-cancer group, with an AUC value of 0.9527. Figure E shows the tumor-associated small extracellular vesicle signal results in the cancer group and the non-cancer group, showing significant differences. Figure F shows the confusion matrix of tumor-associated small extracellular vesicle signal results in the cancer group and the non-cancer group, with a prediction accuracy of 87.27%. The experimental results indicate that different population samples show significant differences in the above-mentioned sample-level indicators.

[0090] Furthermore, statistical analysis of sample-level indicators yields high discrimination performance, demonstrating that the method provided in this embodiment can effectively distinguish between cancer and non-cancer samples in real clinical plasma samples. This result further validates the application potential of a method based on single-particle electrical signals combined with machine learning and statistical analysis in the detection of tumor-related small extracellular vesicles in complex plasma backgrounds.

[0091] In summary, this embodiment achieves an effective conversion from single-event signals to sample-level detection results by introducing a machine learning model to perform scoring-based discrimination of single-particle perforation events and combining it with sample-level statistical analysis strategies; and through simulated plasma spiking experiments and real clinical sample analysis (such as... Figure 5 and Figure 6 (As shown) This verifies the feasibility and stability of the method in quantitative analysis and clinical applications.

[0092] The applicant declares that the detailed method of the present invention is illustrated by the above embodiments, but the present invention is not limited to the above detailed method, that is, it does not mean that the present invention must rely on the above detailed method to be implemented. Those skilled in the art should understand that any improvements to the present invention, equivalent substitutions of the raw materials of the product of the present invention, addition of auxiliary components, selection of specific methods, etc., all fall within the protection scope and disclosure scope of the present invention.

Claims

1. A method for detecting small extracellular vesicles, characterized in that, The detection method includes: The sample to be tested is taken and detected using a solid-state nanopore chip. By applying voltage, the particles in the sample are driven to pass through the nanopore, and the ion current signal of the nanopore is collected. The ion current signal is filtered, baseline corrected and detrended, and single particle passing through the pore event is identified to realize the detection and counting of small extracellular vesicles. The constructed machine learning model is used to calculate the score of the output via event as the target small extracellular vesicle event; the input variables of the machine learning model include the equivalent charge defect, current blocking amplitude, half residence time, left edge slope and right edge slope of the single particle via event; The proportion of through-hole events with scores higher than a preset threshold for the target small extracellular vesicle event is statistically analyzed in the total number of through-hole events. Based on the proportion, the sample type of the sample to be tested is determined, thereby realizing the detection and counting of specific small extracellular vesicles in the sample.

2. The method for detecting small extracellular vesicles according to claim 1, characterized in that, The sample to be tested includes plasma; The small extracellular vesicles include physiologically or pathologically relevant small extracellular vesicles.

3. The method for detecting small extracellular vesicles according to claim 1, characterized in that, The process of using a solid-state nanopore chip for detection includes: Solid nanoporous chips are mounted in a fluid pool, which is divided into two independent liquid storage chambers: a common chamber and a reverse chamber. Electrolyte solutions are added to the common chamber and the reverse chamber. The sample to be tested is added into the fluid cell, and under the action of voltage, electroosmotic flow becomes the dominant driving force for particles in the sample to pass through the nanopores. The solid-state nanoporous chip includes a silicon nitride thin film with nanopores on it. The diameter of the nanopores is 150~350 nm and the thickness is 15~40 nm.

4. The method for detecting small extracellular vesicles according to claim 1, characterized in that, The voltage is 100~300mV; the frequency of acquiring the nanopore ion current signal is 90~110 kHz; and the filtering frequency is 8~12 kHz.

5. The method for detecting small extracellular vesicles according to claim 1, characterized in that, The identification of single-particle via events includes: analyzing the noise statistics of the signal within a local time window, setting a current change threshold based on the background noise level, and determining the start of a via event when the current signal momentarily deviates from the baseline and exceeds the threshold; and determining the end of the event when the signal returns to near the baseline.

6. The method for detecting small extracellular vesicles according to claim 1, characterized in that, The method for constructing the machine learning prediction model includes: A known target small extracellular vesicle sample was taken and detected using a solid-state nanopore chip. The small extracellular vesicle was driven to pass through the nanopore by applying voltage, and the nanopore ion current signal was collected. The ion current signal was filtered, baseline corrected and detrended, and single particle passing through the pore event was identified. The equivalent charge deficit, current blocking amplitude, half-dwell time, left edge slope and right edge slope are extracted from each via event and used as a training set. The data are then input into a machine learning classification model for training. The model outputs a score for each input event as belonging to the target small cell extracellular vesicle event. After training, a machine learning prediction model is obtained. The machine learning classification model includes a gradient boosting decision tree model, which outputs a continuous numerical score between 0 and 1. The criterion for determining an input event as a target small extracellular vesicle is: the score is higher than a preset threshold, and the preset threshold is selected from 0.05 to 0.

3.

7. The method for detecting small extracellular vesicles according to claim 1, characterized in that, The criterion for determining that the sample to be tested is a specific sample type is: the proportion of specific small extracellular vesicles is higher than a threshold, which is selected from 1% to 10%.

8. A small extracellular vesicle detection system, characterized in that, The system includes a nanopore detection unit, an ion current acquisition unit, a data processing unit, and a machine learning discrimination unit; The nanopore detection unit is used to perform the following: detecting a sample to be tested using a solid-state nanopore chip, and driving the sample to be tested through the nanopore by applying a voltage; The ion current acquisition unit is used to perform the following: acquiring ion current signals from nanopores; The data processing unit is used to perform the following: filtering, baseline correction and detrending processing of the ion current signal, and identifying single-particle via events; extracting the equivalent charge defect, current blocking amplitude, half-dwell time, left edge slope and right edge slope from each via event; The machine learning discrimination unit is used to perform the following: inputting the data obtained by the data processing unit into the machine learning prediction model, using the machine learning prediction model to calculate the score of the output via event as a target small extracellular vesicle event; calculating the proportion of via events with scores higher than a preset threshold in the total via events, and determining whether the sample to be tested is a specific sample based on the proportion.

9. The small extracellular vesicle detection system according to claim 8, characterized in that, The system is used to perform the steps of the method for detecting small extracellular vesicles according to any one of claims 1-7; The small extracellular vesicles include cancer-associated small extracellular vesicles; The cancers mentioned include hepatocellular carcinoma, pancreatic cancer, or colorectal cancer.

Images