Antibody sequence generation method of SNN diffusion generation model and related equipment

Through the SNN diffusion generation model, LIF neurons and self-attention mechanisms are used to solve the problems of insufficient accuracy and diversity and low training efficiency in antibody sequence generation in traditional deep learning models, achieving more efficient and stable antibody sequence generation and Better generalization ability.

CN119069032APending Publication Date: 2024-12-03NANJING INST OF INTELLIGENT TECH INST OF MICROELECTRONICS OF THE CHINESE ACAD OF
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411148712.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

Traditional deep learning models have problems such as insufficient accuracy and diversity, low training efficiency and stability, and limited generalization capabilities when generating antibody sequences, making it difficult to quickly respond to disease treatment needs.

Method used

Using the SNN diffusion generation model, antibody sequences are generated through diffusion model, residue embedding module and residue sequence prediction module, LIF neurons and self-attention mechanisms improve sequence diversity and accuracy, and stabilize through layer normalization and residual connection Training process enhances the generalization ability of the model.

Benefits of technology

The generated antibody sequences are more accurate and diverse, the training process is more efficient and stable, the model performs better on new data, and has stronger generalization capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119069032A_ABST
    Figure CN119069032A_ABST
Patent Text Reader

Abstract

The invention discloses an antibody sequence generation method of an SNN diffusion generation model and related equipment, and relates to the technical field of artificial intelligence, the method comprises the following steps: obtaining antibody-antigen complex data, the antibody-antigen complex data comprising an antibody-antigen PDB file; inputting the antibody-antigen complex data into an SNN diffusion generation model to obtain a target antibody sequence, wherein the SNN diffusion generation model is composed of a diffusion model, a residue embedding module and a residue sequence prediction module; and processing the target antibody sequence to obtain a predicted antibody sequence.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an antibody sequence generation method and related equipment of a SNN diffusion generation model. Background Art

[0002] At present, antibody drugs play an increasingly important role in the field of biomedicine and are widely used in the treatment of various serious diseases such as cancer and autoimmune diseases. The design and optimization of antibodies are the core links in the development of antibody drugs, which directly affect their therapeutic effects and market value. Traditional antibody design methods rely on experience and high-throughput screening technology, which is costly and inefficient, and it is difficult to quickly respond to the changing needs of disease treatment. In order to solve this problem, researchers began to explore the use of artificial intelligence and machine learning technologies to improve the efficiency and accuracy of antibody design. The technologies in the field of antibody design mainly include traditional experimental methods and modern computational methods. Traditional antibody design mainly relies on laboratory screening technologies, such as phage display technology (Phage Display) and monoclonal antibody preparation. These methods find antibodies that efficiently bind to the target antigen through a large number of experimental cultivation and screening.

[0003] However, experimental methods are time-consuming, labor-intensive, costly, and have low efficiency and success rates. In recent years, deep learning has made significant progress in protein structure prediction and design. For example, the success of AlphaFold demonstrates the great potential of deep learning in protein structure prediction. This type of method can quickly predict the binding mode of antibodies and antigens and generate optimized antibody sequences by training neural network models. Compared with the past annealing algorithms, genetic algorithms, etc., deep learning methods can often design more effective antibody sequences. Although deep learning-based antibody design methods have made significant progress, there are still some technical bottlenecks, such as: Insufficient accuracy and diversity of antibody generation: When traditional deep learning models generate antibody sequences, they are prone to produce sequences that are too similar, making it difficult to cover the diversity of antibody sequences, and the accuracy of generated sequences needs to be improved. Insufficient model training efficiency and stability: The training of deep learning models usually requires a lot of computing resources and time, and the problem of gradient disappearance or explosion is prone to occur during the training process, resulting in slow model convergence and unstable training. Limited generalization ability: When processing new data, existing models show poor generalization ability and are difficult to show good adaptability on different data sets, resulting in poor performance of the model in practical applications. There is no proper method to solve the above problems. Therefore, it is necessary to propose an antibody sequence generation method based on SNN diffusion generation model to at least solve some of the above problems. Summary of the invention

[0004] A series of simplified concepts are introduced in the Summary of the Invention section, which will be further described in detail in the Detailed Description of the Invention section. The Summary of the Invention section of this application does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the scope of protection of the claimed technical solution.

[0005] In a first aspect, an embodiment of the present application provides an antibody sequence generation method of an SNN diffusion generation model, the method comprising:

[0006] Acquiring antibody-antigen complex data, wherein the antibody-antigen complex data includes an antibody-antigen PDB file;

[0007] Inputting the antibody-antigen complex data into the SNN diffusion generation model to obtain the target antibody sequence, wherein the SNN diffusion generation model is composed of a diffusion model, a residue embedding module and a residue sequence prediction module;

[0008] The target antibody sequence is processed to obtain a predicted antibody sequence.

[0009] In one embodiment of the present invention, the step of obtaining antibody-antigen complex data comprises:

[0010] Obtain antibody-antigen complex data;

[0011] The antibody-antigen complex data are cleaned and normalized to obtain antibody-antigen complex data.

[0012] In one embodiment of the present invention, the step of inputting the antibody-antigen complex data into the SNN diffusion generation model to obtain the target antibody sequence comprises:

[0013] generating a Markov chain of a diffusion process based on the diffusion model;

[0014] Obtaining noise level data according to the Markov chain;

[0015] Obtaining protein molecular structure features and species features according to the antibody antigen PDB file, wherein the species features include atomic species features and residue species features;

[0016] Inputting the noise level data, the protein molecular structure characteristics and the type characteristics into an antibody sequence generator to obtain a target antibody sequence; the antibody sequence generator is composed of a residue embedding module and a residue sequence prediction module;

[0017] The target antibody sequence and the actual antibody sequence are calculated to obtain the model design sequence loss value, and the antibody sequence generator parameters are back-propagated to optimize.

[0018] In one embodiment of the present invention, the step of inputting the noise level data, the protein molecular structure characteristics and the species characteristics into an antibody sequence generator to obtain a target antibody sequence comprises:

[0019] Inputting the protein molecule structural feature into a structural feature embedding module and performing two residual connections to obtain a structural feature embedding vector, wherein the LIF neuron, the linear layer, the identity mapping layer, the LIF layer and the linear layer in the structural feature embedding module are connected in sequence, and the LIF neuron is used to obtain the input protein molecule structural feature;

[0020] The category features are encoded by superimposing positions, and embedded using LIF neurons and embedding layers to obtain category feature embedding vectors;

[0021] Obtaining a sequence feature embedding vector according to the structural feature embedding vector and the category feature embedding vector;

[0022] The sequence feature embedding vector is processed to obtain a target antibody sequence.

[0023] In one embodiment of the present invention, the step of processing the sequence feature embedding vector to obtain the target antibody sequence includes:

[0024] Inputting the sequence feature embedding vector into a multi-layer Spikeformer Decoder for decoding to obtain a decoding result;

[0025] Inputting the decoding result into a linear layer to obtain a first processing result;

[0026] Inputting the first processing result to the LIF layer to obtain a second processing result;

[0027] Inputting the second processing result into a linear layer to obtain a third processing result;

[0028] Inputting the third processing result to the LIF layer to obtain a fourth processing result;

[0029] The fourth processing result is calculated to obtain the target antibody sequence.

[0030] In one embodiment of the present invention, the step of calculating the target antibody sequence and the actual antibody sequence to obtain the model design sequence loss value comprises:

[0031] Obtaining actual antibody sequences;

[0032] The actual antibody sequence and the target antibody sequence are subjected to sequence similarity analysis by Kullback-Leibler divergence to obtain a model design sequence loss value.

[0033] In one embodiment of the present invention, the step of processing the target antibody sequence to obtain a predicted antibody sequence comprises:

[0034] Alphafold is used to predict the structure of the target antibody sequence, and the antibody and antigen are physically spliced ​​according to the antigen PDB file and the ZDOCK tool to obtain a predictive antigen-antibody complex;

[0035] The predicted antigen-antibody complex is evaluated and screened using the ROSETTA antibody binding evaluation method to obtain a predicted antibody sequence.

[0036] In the second aspect, the present application proposes an antibody sequence generation system of an SNN diffusion generation model, the system comprising: a data acquisition module, a calculation module and a data processing module;

[0037] The data acquisition module is configured to: acquire antibody-antigen complex data, wherein the antibody-antigen complex data includes an antibody-antigen PDB file;

[0038] The computing module is configured to: input the antibody-antigen complex data into the SNN diffusion generation model to obtain a target antibody sequence, wherein the SNN diffusion generation model is composed of a diffusion model, a residue embedding module, and a residue sequence prediction module;

[0039] The data processing module is configured to: process the target antibody sequence to obtain a predicted antibody sequence.

[0040] In a third aspect, an electronic device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor is used to implement the steps of an antibody sequence generation method of an SNN diffusion generation model as described in any one of the first aspects above when executing the computer program stored in the memory.

[0041] In a fourth aspect, the present application further proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for generating an antibody sequence of an SNN diffusion generation model according to any one of the first aspects.

[0042] In summary, an antibody sequence generation method of an SNN diffusion generation model in an embodiment of the present application includes: obtaining antibody-antigen complex data, wherein the antibody-antigen complex data includes an antibody-antigen PDB file; inputting the antibody-antigen complex data into the SNN diffusion generation model to obtain a target antibody sequence, wherein the SNN diffusion generation model is composed of a diffusion model, a residue embedding module, and a residue sequence prediction module; processing the target antibody sequence to obtain a predicted antibody sequence. High-quality antibody sequences are generated by gradually adding and removing noise during training. In particular, the discrete classification data diffusion framework used in the diffusion process and the dynamic response characteristics of the LIF (Leaky Integrate-and-Fire) neurons in the SNN (Spiking Neural Network) diffusion generation model make the generated antibody sequences more accurate and more diverse. By using the sparse features of LIF neurons to reduce the amount of calculation, the self-attention mechanism mines the correlation in the input sequence, and the layer normalization and residual connection stabilize the training process, the gradient disappearance problem is reduced, and the training process is more efficient and stable. Considering the widespread feature similarity in biology, through strict data cleaning, normalization, feature extraction and data set partitioning, as well as the characteristics of LIF neurons to enhance local features, the model performs better on new data and has stronger generalization ability.

[0043] The antibody sequence generation method of the SNN diffusion generation model proposed in this application, and other advantages, objectives and features of this application will be partially reflected through the following description, and will also be partially understood by technical personnel in this field through research and practice of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present specification. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0045] Figure 1 A schematic diagram of the process of generating an antibody sequence using a SNN diffusion generation model provided in an embodiment of the present application;

[0046] Figure 2 A schematic diagram of the operation principle of a diffusion model in an antibody sequence generation method of an SNN diffusion generation model provided in an embodiment of the present application;

[0047] Figure 3 A schematic diagram of a residue embedding module in an antibody sequence generation method of an SNN diffusion generation model provided in an embodiment of the present application;

[0048] Figure 4A schematic diagram of a residue sequence prediction module in an antibody sequence generation method of an SNN diffusion generation model provided in an embodiment of the present application;

[0049] Figure 5 A schematic diagram of the Spikeformer structure in an antibody sequence generation method of an SNN diffusion generation model provided in an embodiment of the present application;

[0050] Figure 6 A schematic diagram of the structure of an antibody sequence generation system of an SNN diffusion generation model provided in an embodiment of the present application;

[0051] Figure 7 A schematic diagram of the control electronic device structure of an antibody sequence generation system of an SNN diffusion generation model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] In order to better understand the technical solutions provided by the embodiments of this specification, the technical solutions of the embodiments of this specification are described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of this specification and the specific features in the embodiments are detailed descriptions of the technical solutions of the embodiments of this specification, rather than limitations on the technical solutions of this specification. In the absence of conflict, the embodiments of this specification and the technical features in the embodiments can be combined with each other.

[0053] In this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or equipment. In the absence of more restrictions, the elements limited by the statement "comprise one..." do not exclude the existence of other identical elements in the process, method, article or equipment including the elements. The term "more than two" includes two or more than two situations.

[0054] In recent years, with the improvement of computing power and the development of bioinformatics, computational methods have been widely used in antibody design. The main computational methods include molecular dynamics simulation and computer-aided antibody engineering methods. Molecular dynamics simulation simulates the interaction between antibodies and antigens, predicts the binding force and stability of antibodies, and thus screens suitable antibodies. This method requires high-performance computing resources, takes a long time to simulate, and cannot optimize and design unknown antigens. It belongs to the accelerated screening process. In order to overcome the problems of low antibody performance and poor resistance of unknown antibodies, computer-aided antibody engineering uses algorithms and databases combined with experimental data to design and optimize antibody sequences. This method can improve the efficiency of antibody design, but it depends on the quality of existing data and the degree of optimization of the algorithm. The development of deep learning methods has brought new progress to computer-aided antibody engineering.

[0055] See also Figure 1 , which is a schematic diagram of the process of generating an antibody sequence using an SNN diffusion generation model provided in an embodiment of the present application, and may specifically include:

[0056] S110, acquiring antibody-antigen complex data, wherein the antibody-antigen complex data includes an antibody-antigen PDB file;

[0057] Exemplarily, data acquisition and collection are first performed to acquire antibody-antigen complex data, which includes antibody-antigen PDB files, which are in the format of PDB (Protein Data Bank) files.

[0058] S120, inputting the antibody-antigen complex data into an SNN diffusion generation model to obtain a target antibody sequence, wherein the SNN diffusion generation model is composed of a diffusion model, a residue embedding module, and a residue sequence prediction module;

[0059] Exemplarily, the SNN diffusion generation model mainly includes a diffusion model, a residue embedding module and a residue sequence prediction module. First, the protein molecular structure features and species features are directly extracted from the antibody antigen complex data. Then, these antibody antigen complex data are input into the SNN diffusion generation model, and the noise level data is generated by the diffusion model. Next, the noise level data, protein molecular structure features and species features are input into the antibody sequence generator to generate the target antibody sequence. The antibody sequence generator consists of a residue embedding module and a residue sequence prediction module. In order to ensure that the diffusion generation model of SNN can perform well on new data, that is, it has good generalization ability, data set division is required. The data set division of the training process is as follows: extract the CDR (complementarity-determining region) region amino acid sequence of the antibody antigen complex data; load the sequence classifier; classify the antibody antigen complex data; and construct a training set and a test set based on the classification results. When training for the first time or when a classifier does not exist, use MMseqs2 to build an unsupervised classifier and perform similarity clustering training on the sequence.

[0060] S130, processing the target antibody sequence to obtain a predicted antibody sequence.

[0061] Exemplarily, after the output target antibody sequence undergoes structural prediction by Alphafold, the antibody-antigen complex data is physically spliced, and then the obtained splicing results are evaluated and screened to obtain the predicted antibody sequence.

[0062] In summary, the antibody sequence generation method of the SNN diffusion generation model proposed in the embodiment of the present application generates high-quality antibody sequences by gradually adding and removing noise during training. In particular, the discrete classification data diffusion framework used in the diffusion process and the dynamic response characteristics of the LIF (Leaky Integrate-and-Fire) neurons in the SNN model make the generated target antibody sequence more accurate and rich in biodiversity. By using LIF neuron sparse features to reduce the amount of calculation, the self-attention mechanism mines the correlation in the input sequence, and stabilizes the training process through layer normalization and residual connection, reduces the gradient disappearance problem, and makes the training process more efficient and stable. Considering the feature similarity that is widely present in biology, through strict data cleaning, normalization, feature extraction and data set partitioning, and the characteristics of LIF neurons to enhance local features, the model performs better on new data and has stronger generalization ability.

[0063] In some examples, the step of obtaining antibody-antigen complex data includes:

[0064] Obtain antibody-antigen complex data;

[0065] The antibody-antigen complex data are cleaned and normalized to obtain antibody-antigen complex data.

[0066] Exemplarily, data acquisition and collection are first performed, and the original antibody-antigen complex data file is collected from a database or user input, wherein the antibody-antigen complex in the file is specifically a product described by crystal structure observation technology after the antibody and antigen are physically combined.

[0067] Data preprocessing is a process that must be performed before the SNN diffusion generation model is calculated. The antibody-antigen complex data is cleaned and normalized to remove noise and outliers to obtain the antibody-antigen complex data. During training, data preprocessing includes the following steps and is processed in sequence: reading the antibody-antigen complex data provided from the database; checking the type of the antibody-antigen complex data; loading the PDB parser; confirming whether there is an antibody Fab (fragment of antigen binding) region; confirming whether there are missing values ​​in the PDB file; confirming whether there are multi-protein structural conformations; confirming whether there are unconventional amino acids; amino acid sequence rearrangement; antibody sequence floating point vectorization; continuity check; CDR region marking. For PDB files with problems, repair or process them, and for those that cannot be repaired, terminate the calculation and prompt the link problem. Finally, the preprocessed antibody-antigen complex data is formed. During inference generation, the difference in data preprocessing is that there is no step of reading the data provided from the database and checking the antigen type.

[0068] In some examples, the step of inputting the antibody-antigen complex data into the SNN diffusion generation model to obtain the target antibody sequence includes:

[0069] generating a Markov chain of a diffusion process based on the diffusion model;

[0070] Obtaining noise level data according to the Markov chain;

[0071] Obtaining protein molecular structure features and species features according to the antibody antigen PDB file, wherein the species features include atomic species features and residue species features;

[0072] Inputting the noise level data, the protein molecular structure characteristics and the type characteristics into an antibody sequence generator to obtain a target antibody sequence; the antibody sequence generator is composed of a residue embedding module and a residue sequence prediction module;

[0073] The target antibody sequence and the actual antibody sequence are calculated to obtain the model design sequence loss value, and the antibody sequence generator parameters are back-propagated to optimize.

[0074] Exemplarily, the preprocessed antibody-antigen complex data needs to undergo a feature construction process. Feature construction can extract important features related to antibody function from the preprocessed antibody-antigen complex data. The feature construction process is as follows: generate a mask based on user input or pre-set CDR regions; antigen region labeling; antibody random mutation; binding site feature calculation.

[0075] like Figure 2 As shown, this is a schematic diagram of the operation principle of the diffusion model in the antibody sequence generation method of the SNN diffusion generation model proposed in this application. The diffusion model generates high-quality antibodies by gradually adding and removing noise. The diffusion model training process is as follows: starting from the input antibody-antigen complex data, the process of gradually adding random noise to it is the forward diffusion process; starting from pure noise, gradually removing the noise, and restoring to the original antibody sequence is the reverse diffusion process. During the training process, different degrees of noise are added to each input training sample to generate data pairs with different noise levels. By comparing the output of the antibody sequence generator and the actual antibody sequence, the loss is calculated and the parameters of the antibody sequence generator are optimized by back propagation to complete the training. The noise in the above training process is affected by the data type. When the purpose is to predict the type of antibody sequence, the discrete classification data diffusion framework agreed upon by D3PM is adopted, as follows.

[0076] The noise addition operation in the diffusion process is a predefined Markov chain process with a transition probability of q(x t |x t-1 ), where t is the time step. The denoising back diffusion process is a parameterized Markov chain with a transition probability of p θ (x t-1 |x t ), where θ represents the parameters learned by the model. In order to optimize the generative model p θ (x0) to match the initial input data distribution q(x0), usually optimizing the variational lower bound (VLB):

[0077]

[0078] Among them, L is the variational lower bound, which is used to measure the training objective function of the model, and minimizes the loss of the model by maximizing the variational lower bound. is the expected value of the latent variable x1 given x0. logp(x0|x1) is the log-likelihood of the latent variable x1 given x0. KL (q(x T |x0)||p(x T )) is x T In the approximate posterior distribution q(x T |x0) and the prior distribution pθ (x T ) is the Kullback-Leibler (KL) divergence between . is the latent variable x t The expected value under given x0. KL (q(x t-1 |x t ,x0)||p θ (x t-1 |x t )) is x t-1 Given x0 and x t The approximate posterior distribution q(x T |x0) and the prior distribution p θ (x t-1 |x t ) is the KL divergence between .

[0079] In this method, t represents the diffusion time step, x t represents the sampling result of the probability distribution of the antibody sequence at time t, x t-1 The meaning is similar. In particular, x0 represents the antibody sequence data without noise.

[0080] In this method, q is of the following form:

[0081] q(x t |x t-1 )=Cat(x t ; p = x t-1 Q t ) (2);

[0082] In the formula, Cat represents a category distribution (CategoryDistribution) specified by p that describes the antibody sequence classification at time t, Q t represents a transition probability matrix, [Q t ] ij Represents the value of the i-th row and j-th column in the matrix, and its value is as follows:

[0083]

[0084] In the formula, β t It is a time-dependent parameter in the diffusion process, usually called the noise schedule parameter, which controls the noise level in each diffusion step. K is the number of standard amino acid categories plus the unknown amino acid type identifier, a total of 21 categories.

[0085] Through the Markov chain, the expressions of each item in VLB can be obtained as follows:

[0086]

[0087] in, ⊙ represents element-wise multiplication.

[0088] Based on the forward diffusion process of the Markov chain in the above diffusion model, a series of noise level data with monotonically increasing noise intensity can be obtained, wherein the noise level data is discrete type noise.

[0089] Based on the reverse diffusion process of the Markov chain in the above diffusion model, an antibody sequence generator for predicting antibody sequences can be constructed. The antibody sequence generator consists of a residue embedding module and a residue sequence prediction module. The loss value of the target antibody sequence and the corresponding noise level antibody sequence is calculated by formula (1), and the parameters of the antibody sequence generator are optimized by back propagation to complete the training. After training, the antibody sequence generator that reaches a certain loss value can realize the function of generating effective antibody sequences.

[0090] The residue embedding module in the antibody sequence generator embeds protein molecular structure features and species features. The protein molecular structure features and species features are obtained according to the antibody antigen PDB file, where the species features include atomic species features and residue species features. In the molecular structure features, the angle between the two planes formed by four consecutive atoms in the protein molecule is very important for the dynamic properties and structural stability of the protein, so the angle value is added to the position code as the protein molecular structure feature.

[0091] The noise level data, protein molecular structure characteristics and species characteristics obtained above are input into an antibody sequence generator to obtain a target antibody sequence.

[0092] In some examples, the step of inputting the noise level data, the protein molecular structure characteristics, and the species characteristics into an antibody sequence generator to obtain a target antibody sequence includes:

[0093] Inputting the protein molecule structural feature into a structural feature embedding module and performing two residual connections to obtain a structural feature embedding vector, wherein the LIF neuron, the linear layer, the identity mapping layer, the LIF layer and the linear layer in the structural feature embedding module are connected in sequence, and the LIF neuron is used to obtain the input protein molecule structural feature;

[0094] The category features are encoded by superimposing positions, and embedded using LIF neurons and embedding layers to obtain category feature embedding vectors;

[0095] Obtaining a sequence feature embedding vector according to the structural feature embedding vector and the category feature embedding vector;

[0096] The sequence feature embedding vector is processed to obtain a target antibody sequence.

[0097] For example, Figure 3 As shown, this is a schematic diagram of the residue embedding module in the antibody sequence generation method of the SNN diffusion generation model proposed in this application. The protein molecular structure features will be obtained through the LIF layer (that is, LIF neurons), linear layer, identity mapping layer, LIF layer, linear layer and two residual connections to obtain the structure feature embedding vector.

[0098] In the structure, the LIF (Leaky Integrate-and-Fire) neuron is a simplified neuron model widely used in computational neuroscience and neural network models. This model captures some basic properties of biological neurons while maintaining computational simplicity. The computational characteristics of LIF neurons are as follows:

[0099]

[0100] Where V(t) is the membrane potential at time t; V rest is the resting potential of the neuron; τ is the membrane time constant, which reflects the decay rate of the membrane potential; R is the input resistance; I(t) is the input current.

[0101] LIF neurons have certain behaviors. First, in the absence of action potentials, the input current I(t) charges or discharges the membrane potential V(t) through the membrane resistance R, and the potential accumulates over time. Second, due to the existence of resistance, the membrane potential will naturally decay, and even without input current, the membrane potential will gradually return to the resting potential V rest Finally, when the membrane potential V(t) reaches a certain threshold V th When the neuron generates an action potential (pulse), the membrane potential is instantly reset to the reset potential V rest This gives LIF neurons rich nonlinear capabilities.

[0102] This application utilizes the dynamic response characteristics of LIF neurons to input sequences, which can better model local features of antibody sequence features compared to ordinary linear neurons.

[0103] In some examples, the step of processing the sequence feature embedding vector to obtain a target antibody sequence includes:

[0104] Inputting the sequence feature embedding vector into a multi-layer Spikeformer Decoder for decoding to obtain a decoding result;

[0105] Inputting the decoding result into a linear layer to obtain a first processing result;

[0106] Inputting the first processing result to the LIF layer to obtain a second processing result;

[0107] Inputting the second processing result into a linear layer to obtain a third processing result;

[0108] Inputting the third processing result to the LIF layer to obtain a fourth processing result;

[0109] The fourth processing result is calculated to obtain the target antibody sequence.

[0110] Exemplarily, in the category features, this application adopts the method of atomic category feature embedding and residue category embedding. After superimposed position encoding, the atomic category features and residue category features are both embedded using LIF and embedding layer to obtain atomic category embedding vectors and residue category embedding vectors. After aligning the dimensions of residues and atoms using the broadcast mechanism, a sequence feature embedding vector consisting of atomic structure feature embedding, residue and atom type embedding is constructed.

[0111] The structure of the residue sequence prediction network consists of multiple layers of Spikeformer Decoder and a predictor. Figure 4 As shown, this is a schematic diagram of the residue sequence prediction module in the antibody sequence generation method of the SNN diffusion generation model proposed in this application. The predictor is composed of 2 layers of linear layers and LIF stacking, and the Spikeformer Decoder is composed of multiple modules, each of which is connected in sequence to process the input feature sequence and generate the output feature sequence.

[0112] The key structure in the Spikeformer Decoder is Spike Attention. Figure 5 As shown in FIG. 1 , the schematic diagram of the Spikeformer structure in the antibody sequence generation method of the SNN diffusion generation model proposed in this application is shown. The sequence feature embedding vector is input into the multi-layer Spikeformer Decoder for decoding to obtain the decoding result, thereby realizing the pulse conversion of the input information. The pulse conversion results are respectively recorded as Q sp , K sp , V sp , and then the following calculation is performed:

[0113]

[0114] This process uses the attention mechanism to explore the correlation between different positions in the input sequence, while taking advantage of the sparsity of the spike neural network, which reduces the actual computational complexity and facilitates deployment on a neural computing chip. Here, as with the Transformer, the multi-head attention mechanism is supported. The output after the attention mechanism is processed by layer normalization to standardize the activation values ​​of each layer to stabilize the training process. The normalized output is further processed by a feedforward neural network. A feedforward neural network usually contains two linear transformations and one LIF, which increases the information expression capability of the model by performing non-high-dimensional linear transformations on the input. The output of the feedforward neural network is again processed by layer normalization to further stabilize and standardize the output. Residual connections are introduced between the input and output of the multi-head self-attention mechanism and the feedforward neural network to alleviate the gradient vanishing problem and improve the training effect of the model. By stacking the Spikeformer Decoder with the same structure, a more complex feature representation can be gradually constructed, and efficient sequence feature to sequence feature conversion can be achieved, so that the predictor can better achieve prediction.

[0115] The decoding result is input into the linear layer to obtain a first processing result; the first processing result is input into the LIF layer to obtain a second processing result; the second processing result is input into the linear layer to obtain a third processing result; the third processing result is input into the LIF layer to obtain a fourth processing result; the fourth processing result is calculated to obtain the final target antibody sequence.

[0116] In model research, the parameters that are manually adjusted are generally called hyperparameters. The hyperparameter settings of the diffusion generation model based on SNN include the hyperparameters of the diffusion model and the hyperparameters of the generation model.

[0117] The hyperparameters of the diffusion model are as follows: the number of time steps that determines the degree of refinement of the conversion process from the initial noise to the target distribution; the noise intensity parameter that determines the amount of noise added at each time step; the denoising intensity parameter that determines the amount of noise removed at each time; the noise type parameter that determines what kind of noise to add; and the time scheduling strategy that controls the time step of the denoising generation process. The time step parameter designed by this method is set to 100 after research; the noise intensity parameter and the denoising intensity parameter are constants that are independent of time; the type of noise added is the classified distribution noise shown in step 201; and the time scheduling strategy is fixed time step sampling.

[0118] The hyperparameters of the generative model are as follows: position encoding method; number of stacking layers parameter of the Spikeformer Decoder module; multi-head parameter in the multi-head self-attention layer of each Spikeformer Decoder; dimension parameter of the hidden layer in the multi-head self-attention layer; dimension parameter of the intermediate layer in the feedforward neural network; scaling factor of the residual connection; predictor dimension parameter. In this application, the sine-cosine position encoding method is adopted; the number of stacking layers parameter of the Spikeformer Decoder module is 4, the multi-head parameter is 8, the dimension of the hidden layer in the multi-head self-attention layer is 336; the scaling factor of the residual connection is 1; the predictor dimension parameter is 128.

[0119] In some examples, the step of calculating the target antibody sequence and the actual antibody sequence to obtain the model design sequence loss value includes:

[0120] Obtaining actual antibody sequences;

[0121] The actual antibody sequence and the target antibody sequence are subjected to sequence similarity analysis by Kullback-Leibler divergence to obtain a model design sequence loss value.

[0122] Exemplarily, the sequence similarity of the actual antibody sequence and the target antibody sequence is calculated by calculating the metric function KL divergence to obtain the model design sequence loss value. In order to simultaneously consider the mean and variance of the fitting data, the loss function of the SNN diffusion generation model (composed of a residue embedding module and a residue sequence prediction module) is obtained:

[0123]

[0124] Among them, λ is an adjustable hyperparameter.

[0125] In some examples, the step of processing the target antibody sequence to obtain a predicted antibody sequence includes:

[0126] Alphafold is used to predict the structure of the target antibody sequence, and the antibody and antigen are physically spliced ​​according to the antigen PDB file and the ZDOCK tool to obtain a predictive antigen-antibody complex;

[0127] The predicted antigen-antibody complex is evaluated and screened using the ROSETTA antibody binding evaluation method to obtain a predicted antibody sequence.

[0128] For example, the protein sequence-based structure prediction function of Alphafold is used to predict the structure of the target antibody sequence generated by the model, and then the existing antigen PDB file is combined with the ZDOCK tool to physically splice the antibody and antigen. For the spliced ​​predicted antigen-antibody complex, the authoritative ROSETTA antibody binding evaluation method is used to evaluate and screen out the predicted antibody sequence.

[0129] like Figure 6 As shown, the present application proposes an antibody sequence generation system of an SNN diffusion generation model, the system comprising: a data acquisition module 21, a calculation module 22 and a data processing module 23;

[0130] The data acquisition module 21 is configured to: acquire antibody-antigen complex data, wherein the antibody-antigen complex data includes an antibody-antigen PDB file;

[0131] The calculation module 22 is configured to: input the antibody-antigen complex data into the SNN diffusion generation model to obtain the target antibody sequence, wherein the SNN diffusion generation model is composed of a diffusion model, a residue embedding module and a residue sequence prediction module;

[0132] The data processing module 23 is configured to process the target antibody sequence to obtain a predicted antibody sequence.

[0133] The effects of the above-mentioned system when applying the above-mentioned method can be found in the description of the above-mentioned method embodiment, which will not be repeated here.

[0134] like Figure 7 As shown, an embodiment of the present application also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor, and when the processor 320 executes the computer program 311, it implements the steps of any method for generating antibody sequences of the above-mentioned SNN diffusion generation model.

[0135] Since the electronic device introduced in this embodiment is a device used to implement an antibody sequence generation device of an SNN diffusion generation model in the embodiment of the present application, based on the method introduced in the embodiment of the present application, technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations. Therefore, how the electronic device implements the method in the embodiment of the present application is not introduced in detail here. As long as the equipment used by technical personnel in this field to implement the method in the embodiment of the present application is within the scope of protection of this application.

[0136] In the specific implementation process, when the computer program 311 is executed by the processor, it can achieve Figure 1Any implementation manner in the corresponding embodiments.

[0137] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0138] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-readable program code.

[0139] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0142] An embodiment of the present application also provides a computer program product, which includes computer software instructions. When the computer software instructions are executed on a processing device, the processing device executes the process of the LDPC decoding method of the solid-state hard disk controller.

[0143] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a server or data center that includes one or more available media integration. The available medium can be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state drive (SSD)), etc.

[0144] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0145] In the several embodiments provided in the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0146] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0147] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0148] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0149] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

[0150] Although the preferred embodiments of this specification have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of this specification.

[0151] Obviously, those skilled in the art can make various changes and modifications to this specification without departing from the spirit and scope of this specification. Thus, if these modifications and variations of this specification fall within the scope of the claims of this specification and their equivalents, this specification is also intended to include these modifications and variations.

Claims

1. A method for generating antibody sequences using a SNN diffusion generation model, characterized in that: The method comprises: Acquiring antibody-antigen complex data, wherein the antibody-antigen complex data includes an antibody-antigen PDB file; Inputting the antibody-antigen complex data into the SNN diffusion generation model to obtain the target antibody sequence, wherein the SNN diffusion generation model is composed of a diffusion model, a residue embedding module and a residue sequence prediction module; The target antibody sequence is processed to obtain a predicted antibody sequence.

2. The antibody sequence generation method of the SNN diffusion generation model according to claim 1, characterized in that: The step of obtaining antibody-antigen complex data comprises: Obtain antibody-antigen complex data; The antibody-antigen complex data are cleaned and normalized to obtain antibody-antigen complex data.

3. The antibody sequence generation method of the SNN diffusion generation model according to claim 1, characterized in that: The step of inputting the antibody-antigen complex data into the SNN diffusion generation model to obtain the target antibody sequence comprises: generating a Markov chain of a diffusion process based on the diffusion model; Obtaining noise level data according to the Markov chain; Obtaining protein molecular structure features and species features according to the antibody antigen PDB file, wherein the species features include atomic species features and residue species features; Inputting the noise level data, the protein molecular structure characteristics and the type characteristics into an antibody sequence generator to obtain a target antibody sequence; the antibody sequence generator is composed of a residue embedding module and a residue sequence prediction module; The target antibody sequence and the actual antibody sequence are calculated to obtain the model design sequence loss value, and the antibody sequence generator parameters are back-propagated to optimize.

4. The antibody sequence generation method of the SNN diffusion generation model according to claim 3, characterized in that: The step of inputting the noise level data, the protein molecular structure characteristics and the type characteristics into an antibody sequence generator to obtain a target antibody sequence comprises: Inputting the protein molecule structural feature into a structural feature embedding module and performing two residual connections to obtain a structural feature embedding vector, wherein the LIF neuron, the linear layer, the identity mapping layer, the LIF layer and the linear layer in the structural feature embedding module are connected in sequence, and the LIF neuron is used to obtain the input protein molecule structural feature; The category features are encoded by superimposing positions, and embedded using LIF neurons and embedding layers to obtain category feature embedding vectors; Obtaining a sequence feature embedding vector according to the structural feature embedding vector and the category feature embedding vector; The sequence feature embedding vector is processed to obtain a target antibody sequence.

5. The antibody sequence generation method of the SNN diffusion generation model according to claim 4, characterized in that: The step of processing the sequence feature embedding vector to obtain the target antibody sequence comprises: Inputting the sequence feature embedding vector into a multi-layer Spikeformer Decoder for decoding to obtain a decoding result; Inputting the decoding result into a linear layer to obtain a first processing result; Inputting the first processing result to the LIF layer to obtain a second processing result; Inputting the second processing result into a linear layer to obtain a third processing result; Inputting the third processing result to the LIF layer to obtain a fourth processing result; The fourth processing result is calculated to obtain the target antibody sequence.

6. The antibody sequence generation method of the SNN diffusion generation model according to claim 3, characterized in that: The step of calculating the target antibody sequence and the actual antibody sequence to obtain the model design sequence loss value comprises: Obtaining actual antibody sequences; The actual antibody sequence and the target antibody sequence are subjected to sequence similarity analysis by Kullback-Leibler divergence to obtain a model design sequence loss value.

7. The antibody sequence generation method of the SNN diffusion generation model according to claim 1, characterized in that: The step of processing the target antibody sequence to obtain a predicted antibody sequence comprises: Alphafold is used to predict the structure of the target antibody sequence, and the antibody and antigen are physically spliced ​​according to the antigen PDB file and the ZDOCK tool to obtain a predictive antigen-antibody complex; The predicted antigen-antibody complex is evaluated and screened using the ROSETTA antibody binding evaluation method to obtain a predicted antibody sequence.

8. An antibody sequence generation system based on a SNN diffusion generation model, characterized in that: The system comprises: a data acquisition module, a calculation module and a data processing module; The data acquisition module is configured to: acquire antibody-antigen complex data, wherein the antibody-antigen complex data includes an antibody-antigen PDB file; The computing module is configured to: input the antibody-antigen complex data into the SNN diffusion generation model to obtain a target antibody sequence, wherein the SNN diffusion generation model is composed of a diffusion model, a residue embedding module, and a residue sequence prediction module; The data processing module is configured to: process the target antibody sequence to obtain a predicted antibody sequence.

9. An electronic device, comprising: A memory and a processor, characterized in that the processor is used to implement the steps of an antibody sequence generation method of an SNN diffusion generation model as described in any one of claims 1 to 7 when executing a computer program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the antibody sequence generation method of the SNN diffusion generation model as described in any one of claims 1 to 7 are implemented.

Citation Information

Patent Citations

  • Deep learning model-based antibody structure optimization method and device

    CN116741260A

  • Antibody de novo design method, device and equipment and storage medium

    CN118380069A

  • Improved Method And System For Predicting Outcomes Based On Spatio / Spectro-Temporal Data

    US20160210552A1

  • Protein structure prediction

    WO2024072980A1