Single molecule force spectrum signal automatic analysis method based on optical tweezers

Abstract

The invention discloses a single molecule force spectrum signal automatic analysis method based on optical tweezers, and aims to solve the problems of low analysis efficiency, high subjectivity and signal loss of a noise reduction algorithm in the prior art. The core of the method is that a simulation signal slice data set is generated through a single molecule stretching model, and a fusion perception adaptive noise reduction model is trained; after overlapping sliding slicing and normalization are carried out on force-distance signals collected by the optical tweezers, the model is input for denoising, and the signals are recombined; finally, folding sites are automatically positioned through secondary difference, worm chain model automatic fitting and parameter automatic calculation are carried out, parameters such as the molecular folding structure size and free energy are obtained, and analysis automation of single-molecule force spectrum signals collected through optical tweezers is achieved. The method breaks through the balance problem of noise reduction and signal fidelity, significantly improves the efficiency, accuracy and stability of single-molecule force spectrum analysis, and is suitable for single-molecule mechanical research in the fields of biophysics and molecular biology.

Description

Technical Field

[0001] This invention relates to the field of single-molecule detection, and more specifically to an automatic analysis method for single-molecule force spectrum signals based on optical tweezers. Background Technology

[0002] Optical tweezers single-molecule force spectroscopy uses a laser beam to trap microspheres, indirectly manipulating single molecules. By detecting the microsphere's displacement within the trap, the relationship between molecular forces and deformation can be determined. It can be used to study DNA folding, protein conformational changes, antigen-antibody interactions, and other phenomena, and has wide applications in biophysics and molecular biology.

[0003] The analysis of the core signal of single-molecule force spectroscopy, the force-distance curve, relies on manual annotation and parameter extraction, which has limitations such as low efficiency, high subjectivity, and low data utilization. Automated analysis technology, on the other hand, uses algorithms to achieve batch processing, feature recognition, and parameter quantification of force-distance signals, greatly enhancing the efficiency and accuracy of single-molecule force spectroscopy processing.

[0004] Brownian motion causes the light-trapped microspheres to sway randomly, resulting in irregular fluctuations in the force-distance signal. This, combined with system noise, masks the true mechanical response of the molecules, reduces signal accuracy, and interferes with the automatic algorithm's feature recognition, thus limiting its analytical performance. Existing denoising algorithms such as mean filtering, Kalman filtering, low-pass Butterworth filtering, and local weighted regression cause signal loss during denoising, and their denoising accuracy cannot guarantee the stability and accuracy of force-distance signal analysis algorithms. Summary of the Invention

[0005] This invention addresses the shortcomings of existing automatic force-distance signal analysis algorithms, such as insufficient accuracy and low stability under noise interference, by proposing an automatic single-molecule force spectrum signal analysis method based on optical tweezers. By performing sliding slice segmentation on the force-distance signal, followed by denoising using a fusion-sensor adaptive denoising model and slice recombination, a high signal-to-noise ratio force-distance signal is obtained. Finally, methods such as quadratic difference, worm chain model fitting, and free energy integration are used to automatically calculate the molecular structure parameters in the force-distance signal, achieving automatic analysis of single-molecule force spectrum signals from optical tweezers, significantly improving analysis efficiency and accuracy.

[0006] This invention is achieved through the following technical solution:

[0007] 1) Construct a single-molecule stretching model, use the model to generate single-molecule force spectrum simulation signals, including noisy and noiseless force-distance simulation signals, and segment the simulation signals to obtain noisy and noiseless simulation signal slice datasets.

[0008] The single-molecule stretching model is constructed by the following steps: First, a value is randomly selected in the range of 30-80 nanometers as the holding length P, and three different profile lengths L are randomly selected in the range of 500-1000 nanometers. The three sets of holding lengths P and profile lengths L are substituted into the worm chain model to generate three corresponding curves, where the horizontal axis of the curve is distance and the vertical axis is force. Then, a segment of each of these three curves is randomly selected as a signal curve, and adjacent signal curves partially overlap in force value but do not overlap in distance value, thus constructing a simulated force-distance signal. Next, the simulated signal is subjected to analog-to-digital conversion at a frequency of 100 Hz to obtain a discrete force-distance digital signal, i.e., a noise-free force-distance simulation signal. Finally, Gaussian noise is added to the noise-free simulation signal for both the force signal and the distance signal to obtain a noisy force-distance digital signal, i.e., a noisy force-distance simulation signal.

[0009] The formula for the analog-to-digital conversion is: and

[0010] In the formula, x actual This represents the actual displacement between the two microspheres, where x0 is the initial distance between the two microspheres, and v... trap The moving speed of the optical trap is represented by t, which is the discrete time sequence during the movement of the optical tweezers. F is the instantaneous molecular tension applied between the two microbeads, k is the stiffness of the optical trap, and WLC is the worm chain model.

[0011] The segmentation process divides both noisy and noiseless simulation signals into slices of 300 data points each, with no overlap between the slices.

[0012] 2) Construct a fusion-sensing adaptive denoising model suitable for processing force-distance signal slices, and use the simulation signal slice dataset generated in step 1) to train the fusion-sensing adaptive denoising model to obtain the trained fusion-sensing adaptive denoising model.

[0013] The fusion-aware adaptive denoising model comprises six multi-layer fusion-aware modules, two self-attention modules, and one output convolutional layer. Each three multi-layer fusion-aware modules are connected in series with one self-attention module, forming two series-connected combined structures. Skip connections are added between the combined structures. Finally, one output convolutional layer is used to summarize the feature parameters generated by the aforementioned combined structures to output a denoised signal slice.

[0014] The input and output of the fusion sensing adaptive noise reduction model are both two-channel data, which are respectively applied to the force channel and distance channel of the stress-distance signal slice.

[0015] The multi-layer fusion sensing module consists of four one-dimensional convolutional kernels and one max pooling layer. The number of input and output channels is 80. The number of channels of each convolutional kernel is 16 and the size is 7, 15, 29 and 59 respectively. The number of input and output channels of the max pooling layer is 16 and the size of the pooling kernel is 9.

[0016] The output convolutional layer has 80 input channels, 2 output channels, and a one-dimensional convolutional kernel size of 1.

[0017] In step 2), training the fusion-aware adaptive noise reduction model means: the simulated signal slice dataset is divided into a training set and a validation set in an 8:2 ratio; the noisy simulated signal slices are used as the input data of the model, and the noiseless simulated signal slices are used as the output supervision labels of the model; the model parameters are optimized by the Adam optimizer with mean squared error (MSE) as the loss function, and finally the trained fusion-aware adaptive noise reduction model is obtained.

[0018] 3) Overlap and slide the force-distance signal collected by optical tweezers, and use Max-Min normalization to normalize the signal values ​​to the [-1,1] interval to obtain the normalized signal slice; and input it into the fusion perception adaptive denoising model trained in step 2), and output the denoised signal slice after model processing;

[0019] The sliding slice method involves slicing each slice to a length of 300 data points, with an overlap rate of 97% between adjacent slices. For slices shorter than 300 data points, the last value of the slice is used to supplement the length to 300 data points. The formula for the sliding slice method is as follows:

[0020] In the formula, X is the full-length signal sequence, X k For the k-th slice, index last α is the end index of the previous slice, and α is the overlap rate of adjacent slices.

[0021] 4) Perform inverse Max-Min normalization on the denoised signal slices from step 3), and then perform synthesis processing to obtain the reconstructed force-distance signal;

[0022] The synthesis process refers to sequentially connecting all inverse Max-Min normalized slices according to the sliding slice order, and averaging the values ​​of the overlapping parts of adjacent slices; the formula for the synthesis process is: and

[0023] In the formula, S reconstructed The signal is reconstructed, where x is the index of the data point within the slice, and y is the index of the data point within the slice. average(x) represents the average value of the overlapping region, U is the set of full-length signal indices, k is the number of indices x, and y x,k S represents the value at index x in the k-th slice. i It is the i-th slice out of a total of m slices.

[0024] 5) By automatically locating the folding signal sites, automatically fitting the worm chain model, and automatically calculating the parameters of the force-distance signal recombined in step 4), automatic analysis of single-molecule force spectrum signals is achieved.

[0025] The automatic location of the folded signal point refers to: performing a second differential operation on the force signal in the recombined force-distance signal, with the differential interval set to 10, to obtain a second differential curve, then identifying the peaks and number of peaks greater than 0.5 in the second differential curve, and automatically locating the values ​​of the force signal and distance signal at each peak point according to the time point at which each peak is located, which is the folded signal point;

[0026] The formula for the quadratic difference operation is:

[0027] In the formula, Y is the force signal value, and t is the data point index value;

[0028] The automatic fitting of the worm chain model refers to: segmenting the recombined force and distance signals according to the folding signal sites, and then automatically fitting the curves of each segment using the worm chain model to obtain the holding length and contour length corresponding to each segment of force and distance signal, which are the parameters of the worm chain model;

[0029] The formula for the worm chain model is:

[0030] In the formula, F(x) is the force at a distance of x, and k B Where is Boltzmann constant, T is absolute temperature, P is the persistence length of the DNA strand, and L0 is the outline length of the DNA strand.

[0031] The automatic parameter calculation refers to: based on the parameters fitted by the worm chain model, automatically calculating the parameters of the folded structure of the force-distance curve, i.e., the single-molecule force spectrum signal, to obtain the true size and free energy of the folded structure, as well as the DNA handle information attached to the single molecule;

[0032] The method for calculating the actual size of the folded structure is as follows: First, substitute the force values ​​at the points after the folded structure is unfolded into the worm chain model before it is unfolded, and calculate the distance value corresponding to the model; then, calculate the absolute value of the difference between the distance value and the distance value at the corresponding points after the folded structure is unfolded, and the absolute value is the size of the folded structure.

[0033] The method for calculating the free energy of the folded structure is as follows: based on the relationship between free energy and integral area in thermodynamics, the area enclosed by the line connecting the fitting curve of the adjacent worm chain model and the folding signal site is integrated, and the integral value is the free energy of the folded structure.

[0034] Compared with the prior art, the present invention has the following beneficial technical effects:

[0035] Compared with traditional single-molecule force spectroscopy analysis methods, this invention employs an automatic single-molecule force spectroscopy signal analysis method based on optical tweezers. This method achieves precise processing of force-distance signals affected by Brownian motion and system noise, overcoming the challenge of balancing noise reduction and signal fidelity in traditional denoising algorithms. Compared with analysis methods relying on manual annotation and existing denoising techniques that easily cause signal loss, this method not only achieves batch processing of force-distance signals, automatic feature recognition, and parameter quantification, but also improves denoising accuracy while ensuring the integrity of the true mechanical response signal of molecules. This significantly enhances the efficiency, accuracy, and stability of single-molecule force spectroscopy analysis, effectively solving the problems of low efficiency, strong subjectivity, and insufficient denoising accuracy of traditional algorithms that restrict analytical performance. Attached Figure Description

[0036] Figure 1 This is a flowchart of the entire process of this invention;

[0037] Figure 2 This is a schematic diagram of the construction of the single-molecule stretching model of the present invention;

[0038] Figure 3 This is a schematic diagram of the digital-to-analog conversion of the force-distance simulation signal in this invention;

[0039] Figure 4 This is a structural diagram of the fusion-sensory adaptive noise reduction model of the present invention;

[0040] Figure 5 This is a structural diagram of the multi-layer fusion sensing module in the fusion sensing adaptive noise reduction model of the present invention;

[0041] Figure 6 This is a schematic diagram of the force-distance signal decomposition and sliding slicing method of the present invention;

[0042] Figure 7 This is a schematic diagram of the slice denoising method based on the fusion-sensing adaptive denoising model of the present invention;

[0043] Figure 8 This is a schematic diagram of the noise reduction force-distance signal slice synthesis processing method of the present invention;

[0044] Figure 9 This is a schematic diagram of the automatic location of the folded signal position (folded point) using the quadratic difference method of the present invention;

[0045] Figure 10 This is a schematic diagram of the automatic location of folding points and automatic fitting of worm chain model for multi-folded force-distance signals in this invention;

[0046] Figure 11 This is a schematic diagram of the integral method for calculating the free energy of the folded structure according to the present invention; Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The full process flowchart of the present invention is as follows: Figure 1 As shown.

[0048] Step 1) Construction of single-molecule force spectrum simulation signals and generation of dataset

[0049] In this embodiment, a single-molecule stretching model is first constructed to generate a force-distance simulation signal slice dataset for model training and validation. The single-molecule stretching model is based on a worm chain model, simulating stretching behavior under different molecular conformations through random parameter generation and folding site selection.

[0050] Specifically, such as Figure 2 As shown, a value was randomly selected within the range of 30-80 nm as the persistence length P of the DNA molecule, and three distinct profile lengths L were randomly selected within the range of 500-1000 nm. Substituting each set of P and L parameters into the worm chain model formula, three force-distance stretching curves were generated. During generation, the curves partially overlapped in the force direction but not in the distance direction, thus simulating the continuous occurrence of multiple folding events in a real single-molecule stretching experiment. Subsequently, as... Figure 3 As shown, a signal segment is randomly selected from the continuous stretching curve to obtain a continuous force-distance analog signal. The signal is then processed by analog-to-digital conversion according to the sampling conditions of the optical tweezers experiment. The preferred sampling frequency is 100Hz, resulting in a noise-free discrete force-distance digital signal. Based on this, Gaussian white noise is superimposed on both the force and distance signals to simulate the combined effects of Brownian motion and system measurement noise, ultimately forming a set of noisy force-distance simulation signals.

[0051] To meet the input requirements of the subsequent model, the noisy and noiseless simulation signals were processed into non-overlapping slices of 300 data points each, resulting in a dataset of paired simulation signal slices. Each pair of slices served as the input sample and the corresponding noiseless supervision label for the model.

[0052] Step 2) Construction and training of the fusion-perceptual adaptive noise reduction model

[0053] In this embodiment, taking advantage of the characteristic that force-distance signals simultaneously contain force and distance channels, a fusion-sensing adaptive noise reduction model is constructed. The structure of the model is as follows: Figure 4 As shown, this is used for joint denoising of sliced ​​force-distance signals.

[0054] The model consists of six multi-layer fusion perception modules, two self-attention modules, and one output convolutional layer. Each set of three multi-layer fusion perception modules is concatenated with one self-attention module, forming two sets of concatenated structures. Skip connections are used between these sets to enhance gradient propagation during deep feature extraction. The multi-layer fusion perception modules employ multi-scale one-dimensional convolutional structures, such as... Figure 5 As shown, it consists of four parallel one-dimensional convolutional kernels and one max-pooling layer. The convolutional kernel sizes are 7, 15, 29, and 59, used to extract force-distance variation features at different time scales; the pooling layer is used to enhance local robustness and suppress high-frequency noise interference. The model's input and output are both dual-channel data, corresponding to the stress signal channel and the distance signal channel respectively, enabling the fusion-sensory adaptive denoising model to establish a joint mapping model on the force and distance signals, ensuring the robustness and fidelity of denoising.

[0055] During the training phase, the simulated signal slice dataset was divided into training and validation sets in an 8:2 ratio. Noisy signal slices were used as model input, and corresponding noise-free signal slices were used as supervision labels. The Adam optimizer was employed, with mean squared error as the loss function, to iteratively optimize the model parameters, ultimately yielding a trained fusion-aware adaptive noise reduction model.

[0056] Step 3) Slicing and normalization of the experimental force-distance signal

[0057] In practical applications, the raw force-distance signal acquired through optical tweezers experiments is input into the method of this invention for automatic analysis. First, the complete signal is processed by sliding slicing, with each slice length set to 300 data points, and the overlap rate between adjacent slices preferably being 97%. Figure 6 As shown. When there are fewer than 300 data points at the end of the signal, the last sampled value is repeated to make up the difference. Then, Max-Min normalization is applied to each slice to map the force signal and distance signal values ​​to the [-1,1] interval, so as to eliminate the influence of signal amplitude differences under different experimental conditions on the model inference results.

[0058] Step 4) Slicing, denoising, and force-distance signal synthesis

[0059] The normalized force-distance signal slices are input into the trained fusion-sensory adaptive denoising model. The model performs end-to-end inference for each slice and outputs the corresponding denoised signal slice, such as... Figure 7 As shown. Subsequently, the denoised slices undergo inverse Max-Min normalization to restore them to their original data scale. Following the slice generation order, all denoised slices are sequentially stitched together to form a complete signal; for data regions with overlapping areas in adjacent slices, the data points in these regions are averaged, as shown... Figure 8 As shown, in order to avoid the splicing discontinuity problem at the slice boundary, a high signal-to-noise ratio reconstructed force-distance signal is finally obtained.

[0060] Step 5) Automatic identification of fold sites, automatic fitting of worm chain model and automatic calculation of parameters

[0061] After obtaining the reconstructed force-distance signal, a second differential operation is first performed on the force signal in the reconstructed force-distance signal. The preferred differential interval is 10 sampling points. Figure 9 As shown, by detecting the peak positions in the second difference result where the amplitude exceeds a preset threshold (such as 0.5), the automatic identification of single-molecule folding event sites is achieved, and the corresponding force and distance values, i.e., folding sites, are obtained simultaneously.

[0062] Using the identified folding points as boundaries, the complete force-distance signal is divided into multiple segments, such as... Figure 10 As shown, each signal segment was automatically fitted using a worm chain model to obtain the corresponding persistence length and profile length parameters, which were used to characterize the DNA handle and molecular conformation state.

[0063] Based on the fitting results of the worm chain model, the automatic calculation of the true dimensions and free energy of the folded structure is further realized. Specifically, such as... Figure 11 As shown, the site force value after the folded structure unfolds is substituted into the corresponding worm chain model before unfolding to calculate the theoretical distance value. The absolute value of the difference between this distance value and the experimentally measured distance value after unfolding is calculated. This absolute value is the true size of the folded structure, thus effectively eliminating the interference caused by the elastic deformation of the DNA handle. Simultaneously, based on the relationship between free energy and work integral in thermodynamics, the area enclosed by the fitting curve of adjacent worm chain models and the line connecting the folded site is integrated. The integral result is the free energy change of the corresponding folded structure.

[0064] Through the above steps, this invention achieves fully automated analysis of single-molecule force spectrum signals, starting from the original optical tweezers experimental signals, including denoising, folding site identification, model fitting, and calculation of structural parameters and free energies. This method significantly improves the stability, accuracy, and analysis efficiency of single-molecule force spectrum data processing while ensuring the integrity of the true molecular mechanical response information. It is particularly suitable for single-molecule mechanical studies under experimental conditions involving multiple folding events and low signal-to-noise ratios.

[0065] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can make appropriate changes or modifications within the scope of the technology disclosed in the present invention, and such changes or modifications should be covered within the scope of protection of the present invention.

Claims

1. An automatic analysis method for single-molecule force spectrum signals based on optical tweezers, characterized in that, Includes the following steps: 1) Construct a single-molecule stretching model, use the model to generate single-molecule force spectrum simulation signals, including noisy and noiseless force-distance simulation signals, and segment the simulation signals to obtain noisy and noiseless simulation signal slice datasets. 2) Construct a fusion-sensing adaptive denoising model suitable for processing force-distance signal slices, and use the simulation signal slice dataset generated in step 1) to train the fusion-sensing adaptive denoising model to obtain the trained fusion-sensing adaptive denoising model. 3) Overlap and slide the force-distance signal collected by optical tweezers, and use Max-Min normalization to normalize the signal values ​​to the [-1,1] interval to obtain the normalized signal slice; and input it into the fusion perception adaptive denoising model trained in step 2), and output the denoised signal slice after model processing; 4) Perform inverse Max-Min normalization on the denoised signal slices from step 3), and then perform synthesis processing to obtain the reconstructed force-distance signal; 5) By automatically locating the folding signal sites, automatically fitting the worm chain model, and automatically calculating the parameters of the force-distance signal recombined in step 4), the automatic analysis of the single-molecule force spectrum signal is realized.

2. The automatic analysis method for single-molecule force spectrum signals based on optical tweezers according to claim 1, characterized in that, The single-molecule stretching model in step 1) is constructed as follows: First, a value is randomly selected in the range of 30-80 nm as the holding length P, and three different profile lengths L are randomly selected in the range of 500-1000 nm. The three sets of holding lengths P and profile lengths L are substituted into the worm chain model to generate three corresponding curves, where the horizontal axis of the curve is distance and the vertical axis is force. Then, a segment of each of these three curves is randomly selected as a signal curve, and adjacent signal curves partially overlap in force value but have no overlap in distance value. The analog force-distance signal is constructed by stacking the data. Next, the analog signal is converted to digital signal at a frequency of 100 Hz to obtain discrete force-distance digital signals, i.e., noise-free force-distance simulation signals. Finally, Gaussian noise of force and distance signals is added to the noise-free simulation signals to obtain noisy force-distance digital signals, i.e., noisy force-distance simulation signals. The segmentation process involves dividing both the noisy and noise-free simulation signals into slices of 300 data points each, with no overlap between the slices.

3. The automatic analysis method for single-molecule force spectrum signals based on optical tweezers according to claim 1, characterized in that, The fusion-aware adaptive denoising model in step 2) comprises six multi-layer fusion-aware modules, two self-attention modules, and one output convolutional layer. Each set of three multi-layer fusion-aware modules is connected in series with one self-attention module, forming two series-connected structures with skip connections between them. Finally, one output convolutional layer summarizes the feature parameters generated by the aforementioned combined structures to output a denoised signal slice. The input and output of this fusion-aware adaptive denoising model are both two-channel data, corresponding to the force and distance channels of the stress-distance signal slice, respectively. Each multi-layer fusion-aware module consists of four one-dimensional convolutional kernels and one max-pooling layer, with 80 input and 80 output channels. Each convolutional kernel has 16 channels and dimensions of 7, 15, 29, and 59, respectively. The max-pooling layer has 16 input and 16 output channels and a kernel size of 9. The output convolutional layer has 80 input channels, 2 output channels, and a one-dimensional convolutional kernel size of 1.

4. The automatic analysis method for single-molecule force spectrum signals based on optical tweezers according to claim 1, characterized in that, Step 2) of training the fusion-aware adaptive noise reduction model refers to: using noisy simulated signal slices as input data for the model and noiseless simulated signal slices as output supervision labels for the model, optimizing the model parameters using the Adam optimizer with mean squared error as the loss function, and finally obtaining the trained fusion-aware adaptive noise reduction model.

5. The automatic analysis method for single-molecule force spectrum signals based on optical tweezers according to claim 1, characterized in that, In step 3), sliding slicing refers to dividing the force and distance signals acquired by optical tweezers into overlapping slices of 300 data points each, with an overlap rate of 97% between adjacent slices. When the number of data points in a slice is less than 300, the last value of that slice is used to supplement the slice to 300 data points. The synthesis process refers to connecting all the inverse Max-Min normalized slices in sequence according to the sliding slicing order, and averaging the values ​​of the overlapping parts in adjacent slices.

6. The automatic analysis method for single-molecule force spectrum signals based on optical tweezers according to claim 1, characterized in that, In step 5), automatic positioning of the folded signal site refers to: performing a second differential operation on the force signal in the recombined force-distance signal, setting the differential interval to 10, obtaining a second differential curve, then identifying the peaks and number of peaks greater than 0.5 in the second differential curve, and automatically locating the values ​​of the force signal and distance signal at each peak point according to the time point at which each peak is located. This value is the folded signal site.

7. The automatic analysis method for single-molecule force spectrum signals based on optical tweezers according to claim 1, characterized in that, In step 5), automatic fitting of the worm chain model refers to: segmenting the recombined force and distance signals according to the folding signal sites, and then automatically fitting the curves of each segment using the worm chain model to obtain the holding length and contour length corresponding to each segment of force and distance signal, which are the parameters of the worm chain model.

8. The automatic analysis method for single-molecule force spectrum signals based on optical tweezers according to claim 1, characterized in that, The automatic parameter calculation in step 5) refers to: based on the parameters fitted by the worm chain model, automatically calculating the parameters of the folded structure of the force-distance curve, i.e., the single-molecule force spectrum signal, to obtain the true size and free energy of the folded structure, as well as the DNA handle information attached to the single molecule.

Images