Beta-secretase activity detection method based on fluorescent scintillation fingerprint intelligent identification

Abstract

The invention relates to a single-molecule peptide recognition and enzyme activity detection method, in particular to a beta-secretase activity detection method based on fluorescent scintillation fingerprints and deep learning recognition. The method comprises the following steps: marking a spontaneous scintillation type fluorescent dye on a specific peptide substrate, and fixing the specific peptide substrate on the surface of a functionalized slide by utilizing click chemistry to realize stable anchoring of a single peptide molecule; then, a total internal reflection fluorescence microscope is adopted to collect a time sequence fluorescence track, and single molecule scintillation fingerprints before and after enzyme digestion are obtained. And track data is classified and analyzed by combining a deep learning model, so that whether the peptide molecules are subjected to enzyme digestion or not can be accurately judged, and quantitative detection of the enzyme activity is further realized. Compared with a traditional method, the method has the advantage that the detection sensitivity, the specificity and the adaptability to complex samples are remarkably improved. The beta-secretase is used as a model, the verification accuracy rate exceeds 88%, a new technical path is provided for enzyme activity analysis and early diagnosis of related diseases, and the method has a wide application prospect.

Description

Technical Field

[0001] This invention relates to the field of biological detection technology, specifically to enzyme activity detection methods based on single-molecule fluorescence imaging and deep learning analysis, and particularly to a highly sensitive enzyme activity detection method that utilizes fluorescence scintillation trajectory fingerprinting technology to determine peptide molecular sequence and enzyme digestion reaction status. Background Technology

[0002] Enzymes, as key biological catalysts in life processes, directly influence cellular metabolism, signal transduction, and the regulation of pathological mechanisms. In research on neurodegenerative diseases such as Alzheimer's disease, β-secretase has attracted much attention due to its cleavage of amyloid precursor protein (APP). Therefore, developing a β-secretase activity detection technology with high sensitivity, high specificity, and strong versatility has significant clinical and research value.

[0003] Currently used methods for enzyme activity detection include fluorescence, colorimetry, radiolabeling, and mass spectrometry. While these methods can accomplish the detection task to a certain extent, they generally suffer from problems such as insufficient sensitivity, poor specificity, large background interference, poor sample adaptability, and complex experimental procedures, making it difficult to meet the needs for high-precision identification of low-abundance, dynamically active molecules such as β-secretase.

[0004] In recent years, the development of single-molecule detection technology has brought breakthroughs to enzyme activity research. Single-molecule fluorescence imaging can track the interaction between substrates and enzymes at nanoscale spatial and millisecond temporal resolution, significantly improving detection sensitivity and information density. However, the fluorescence signals of peptide substrates at the single-molecule level exhibit complex scintillation behavior, which is significantly affected by the microenvironment. Traditional analytical methods rely on manually setting thresholds and extracting features, making it difficult to accurately identify enzyme cleavage behavior or resolve substrate sequence information, thus limiting the versatility and accuracy of the technology.

[0005] Deep learning has demonstrated outstanding performance in biological image processing and temporal signal recognition in recent years. It can automatically extract latent patterns from high-dimensional and complex data, making it suitable for processing temporal signals with nonlinear and multi-scale characteristics, such as single-molecule fluorescence trajectories. By introducing deep learning models into the analysis of single-molecule fluorescence scintillation trajectories, we can not only improve the accuracy of enzyme digestion event recognition but also achieve high-throughput and intelligent assessment of enzyme activity.

[0006] Therefore, developing an intelligent detection method that integrates single-molecule fluorescence imaging and deep learning analysis to achieve highly sensitive identification and accurate quantification of β-secretase activity based on the "fingerprint" of substrate fluorescence scintillation trajectory has significant scientific value and application prospects. This invention is proposed based on the above background, aiming to overcome the technical bottlenecks of existing methods and construct a novel enzyme activity detection platform with high versatility, strong adaptability, and automatic identification capabilities. Summary of the Invention

[0007] The purpose of this invention is to overcome the problems of insufficient sensitivity, specificity and adaptability of existing technologies in enzyme activity detection, and to provide an enzyme activity detection method based on the combination of single-molecule fluorescent scintillation fingerprint and deep learning algorithm.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] (1) Construction and immobilization of peptide substrates:

[0010] Peptide sequences representing the states before and after enzyme digestion were designed and coupled to a hydroxymethylsiloxane (HMSiR) fluorescent dye with spontaneous scintillation properties. The fluorescent peptides were immobilized on surface-treated glass slides via click chemistry (azido-alkynyl cycloaddition reaction) to construct a stable single-molecule detection platform.

[0011] (2) Single-molecule imaging:

[0012] Using a high numerical aperture oil immersion microscope and TIRFM, time-series intensity data of a single fluorescent molecule under the action of an excitation laser were collected, and the scintillation trajectory of each peptide molecule was recorded to form a high-resolution scintillation fingerprint database.

[0013] (3) Data preprocessing:

[0014] The collected fluorescence flicker trajectory data is preprocessed to remove background noise and invalid trajectories, and valid flicker trajectories are selected by using a fluorescence intensity threshold. The preprocessing process includes trajectory smoothing, drift removal, and normalization operations to ensure data quality and consistency, and to ensure the quality of the input data for the deep learning model.

[0015] (4) Deep learning modeling:

[0016] An end-to-end trajectory classification network was constructed, including: a) a one-dimensional convolutional layer (1D-CNN) to extract local trajectory variation features; b) a two-layer long short-term memory network (LSTM) to model the temporal dynamic change trend of the trajectory; c) a normalization layer and dropout mechanism to improve the robustness of the model; and d) a fully connected layer and a softmax output layer to classify substrates and products. The model can automatically learn the temporal features in fluorescence scintillation trajectories without the need for pre-extraction of manual features. The model accurately distinguishes the scintillation fingerprints of peptide sequences before and after enzyme digestion, achieving precise identification of the enzyme digestion state of peptide molecules.

[0017] (5) Construction of standard curve and activity calculation:

[0018] β-secretase reaction systems with different concentration gradients were set up, and the product proportions at different time points were recorded. The product quantity percentages were statistically analyzed using a deep learning model to establish a nonlinear standard curve between product proportion and enzyme concentration. After model prediction, test samples were matched to the standard curve based on product proportions to inversely estimate the activity concentration.

[0019] The method of the present invention has the following advantages:

[0020] This invention presents a method with high sensitivity and specificity, enabling precise identification of enzyme digestion states through single-molecule scintillation fingerprinting combined with deep learning, without the need for additional fluorescent substrates. Compared to traditional methods that rely on changes in fluorescence intensity or endpoint readings, this method is more suitable for detecting low-concentration enzymes, offers a high degree of automation, and has a low false positive rate. It is applicable to the rapid, dynamic, and quantitative analysis of various enzyme activities, and has significant scientific research and clinical translational value. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0022] Figure 2 (a) Substrate sequence design diagram before and after enzyme digestion; (b) Schematic diagram of substrate immobilization process;

[0023] Figure 3 The image shows the single-molecule fluorescence scintillation trajectories of (a) the substrate and (b) the product molecule.

[0024] Figure 4 This is a diagram of the deep learning model structure;

[0025] Figure 5 The graph shows the changes in (a) accuracy and (b) loss function during the model training process.

[0026] Figure 6 This is the classification confusion matrix diagram of the model;

[0027] Figure 7 This is a standard curve of enzyme activity. Detailed Implementation

[0028] This embodiment uses β-secretase as the detection target to verify the enzyme activity detection technology of the present invention based on single-molecule fluorescent scintillation fingerprinting and deep learning methods. Figure 1 The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0029] Example 1

[0030] I. Design and immobilization of peptide substrates:

[0031] (1) As Figure 2As shown, this invention designs and synthesizes two peptide substrate sequences corresponding to the substrate and product before and after β-secretase digestion. Figure 2 a) Both peptides introduce an alkyne group at N, and are then coupled to a hydroxymethylrhodamine fluorescent dye with spontaneous scintillation properties. Figure 2 b. Click chemistry (azido-alkynyl cycloaddition reaction) immobilizes the modified peptide sequence on the surface of a pretreated glass slide to form a stable single-molecule fluorescence detection platform.

[0032] (2) Using TIRFM to image fluorescent peptide molecules fixed on the surface of a glass slide. For example... Figure 3 As shown, single-molecule images acquired by microscopy were selected based on fluorescence intensity to obtain images suitable for recording each β-secretase digestion process before and after digestion. Figure 3 a substrate and Figure 3 b. The scintillation trajectory of the product molecule over time. The acquired trajectory data records the fluorescence intensity in frames, forming a time-series "scintillation fingerprint".

[0033] (3) Figure 4 As shown, a deep learning model for classifying flickering trajectories is constructed. During model training, the input is the raw flickering trajectory without manual feature extraction, and the labels are trajectory annotations with known pre / post-β-secretion enzyme digestion states. A deep neural network model is constructed, including a one-dimensional convolutional layer to extract local features, two LSTM layers to model time dependencies, and a fully connected and softmax output layer to complete the classification. Training uses the cross-entropy loss function and the Adam optimizer, and is conducted for 500 epochs (4500 iterations). Accuracy and loss convergence are recorded during training. Figure 5 a,5b), such as Figure 5 As shown, the results indicate that the model can converge effectively.

[0034] After the model is trained, predictions are made on the test set, and a confusion matrix is ​​plotted. Figure 6 The model's accuracy in identifying the two types of trajectories is reflected in its performance. The model achieved good classification performance on the test set, distinguishing between the states of the substrate and the product peptide before and after β-secretase digestion.

[0035] (4) Immobilized peptide substrates were subjected to in vitro enzymatic digestion using β-secretase at concentrations of 0.21 nM, 1.00 nM, 2.08 nM, 5.00 nM, and 7.69 nM. Fluorescence trajectory data were acquired at reaction times of 0, 30, 60, 90, and 120 min. After classification using a deep learning model, the ratio of the number of peptide trajectories of the digested product to the total number of peptide trajectories was calculated. Enzyme activity standard curves were constructed based on the substrate peptide ratios under different enzyme concentrations, such as... Figure 7 As shown.

[0036] Experimental results show that the established detection method can accurately quantify β-secretase activity, with a standard curve accuracy exceeding 88%, demonstrating high sensitivity and specificity. The accuracy versus loss function curves, classification confusion matrix, and enzyme activity standard curve during model training all indicate that this method performs excellently in β-secretase activity detection and is suitable for in vitro diagnosis and drug screening of related diseases such as Alzheimer's disease.

[0037] In summary, this embodiment fully verifies the effectiveness and practicality of the enzyme activity detection method based on single-molecule fluorescent scintillation fingerprinting combined with deep learning, and has broad application prospects. Although the invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the invention fall within the scope of protection claimed by this invention.

Claims

1. A method for detecting β-secretase activity based on fluorescent scintillation fingerprint intelligent recognition, characterized in that... The specific steps are as follows: Step 1: Obtain the single-molecule fluorescence scintillation trajectory of the peptide substrate to form scintillation fingerprint data for identification; Step 2: Preprocess the acquired fluorescence scintillation trajectory data to extract standardized scintillation fingerprint features; Step 3: Use a deep learning model to classify and analyze the scintillation fingerprint features in order to determine the enzyme digestion status and quantify the enzyme activity. Step 1 includes: designing polypeptide sequences representing the pre- and post-cleavage states of β-secretase, introducing alkyne groups at their N-termini, and chemically coupling them with hydroxymethylsiloxane (HMSiR) fluorescent dye, which has spontaneous scintillation properties, via azide-alkyne click chemical coupling; then immobilizing the coupling product on a functionalized glass slide surface; and using a total internal reflection fluorescent microscope (TIRFM) to collect time-series fluorescence intensity changes of individual peptide molecules, recording their single-molecule scintillation trajectories, and generating a scintillation fingerprint database. Step 2 includes: denoising, drift removal, and background correction of the trajectory data obtained in step 1; using dynamic thresholding to remove low signal-to-noise ratio or incomplete trajectories; performing intensity normalization and length standardization on the filtered trajectories; and inputting the processed complete trajectory as one-dimensional time-series data into the deep learning model without the need for manual feature extraction. Step 3 includes: constructing an end-to-end deep neural network model, which includes: (1) a one-dimensional convolutional layer (1D-CNN) for extracting local pattern features of the flickering trajectory; (2) a two-layer long short-term memory network (LSTM) for modeling the dynamic information of the trajectory changing over time; (3) a normalization and dropout module for improving the stability and robustness of model training; (4) a fully connected layer and a Softmax output layer for classification prediction; the model uses cross-entropy as the loss function and is trained with the Adam optimizer, and predicts whether the peptide molecule has undergone enzyme digestion through the model output; further, the proportion of trajectories classified as enzyme digestion products by the model is statistically analyzed, and combined with the standard reaction curves at different concentrations, the quantitative detection of β-secretase activity in the sample is realized.

2. The method according to claim 1, characterized in that, Steps 2 and 3 employ an end-to-end temporal deep learning method to standardize and automatically extract the single-molecule fluorescence scintillation trajectories of peptide substrates and products. Combined with a unified input format and dynamic model training, this method enables accurate identification and high-throughput detection of various enzyme substrate reaction states, demonstrating strong versatility and adaptability.

3. The method according to claim 1, characterized in that, The method is applicable to the detection of various enzymes and their corresponding substrate systems.

Images