Method and apparatus for designing protein multifunctional fragment scaffolds based on diffusion model
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-03
- Publication Date
- 2026-08-11
AI Technical Summary
关于条件生成方法，如会议论文集International Conference on LearningRepresentations上记载的文献《Diffusion probabilistic modeling of proteinbackbones in 3d for the motif-scaffolding problem》中提出的SMCdiff模型，只能实现根据给定主题为一个蛋白质功能片段设计支架，并且不能保证图案的存在
Smart Images
Figure CN118571318B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of protein structure design technology, specifically relating to a method and apparatus for designing multifunctional protein fragment scaffolds based on a diffusion model. Background Technology
[0002] Designing proteins with specific functions is crucial for vaccines and enzymes. An important approach is to design stable scaffolds to support the desired protein functional fragments. In wet experiments, protein functional fragment scaffolds have proven significant, as drugs are designed using specific examples of addressing the challenges of creating protein functional fragment scaffolds.
[0003] In existing technologies, diffusion models are often used for protein structure design or reconstruction tasks. A diffusion model first introduces Gaussian noise to make the protein structure appear random, and then uses a reverse diffusion process—that is, gradually removing the noise—to cause the protein structure to gradually converge to a reasonable three-dimensional structure. This process is similar to how proteins in nature gradually fold from a disordered state to a stable structure.
[0004] Specifically, the diffusion model includes a forward process (adding noise) and a backward process (removing noise). In the forward process, the model gradually adds Gaussian noise to the original protein structure in a certain number of steps, transforming it into noisy data. In the backward process, the model uses variational inference and other methods to gradually remove noise, allowing the protein structure to be gradually recovered from the noise, ultimately yielding a reasonable three-dimensional structure.
[0005] However, previous studies have often focused on reconstructing stable protein structures from a single functional protein fragment. The common approach is to immobilize the functional fragment and then perform noise reduction on the free amino acids to obtain a protein structure containing only that fragment. Applying this method to the construction of structures with multiple functional protein fragments requires prior knowledge of their relative positions, which is not easily obtained.
[0006] In addition, some studies have addressed the design problem of protein functional structural fragment scaffolds through conditional generation or repair methods. Regarding conditional generation methods, such as the SMCdiff model proposed in the paper "Diffusion probabilistic modeling of protein backbones in 3D for the motif-scaffolding problem" published in the International Conference on Learning Representations, this model can only design scaffolds for a single protein functional fragment based on a given theme, and cannot guarantee the existence of the pattern.
[0007] Regarding repair methods, the paper "Illuminating protein space with a programmable generative model" published in Nature, Volume 623, pp. 1070–1078, proposes the Chroma protein generation model. This model addresses the problem of protein structure reconstruction of individual protein functional fragments by introducing a diffusion process that follows conformational statistics, an efficient neural architecture, a synthesis layer, and a low-temperature sampling algorithm.
[0008] The paper "De novo design of protein structure and function withrfdiffusion" published in Nature, Volume 620, pages 1089-1100, proposed the RoseTTAFold model, which constructs a generative model capable of generating protein backbones by finely adjusting the prediction network to cope with protein structure generation tasks.
[0009] However, both of the above approaches fix the structural and sequence positions of the desired protein functional fragments. Therefore, when designing scaffolds for multiple protein functional fragments, their positional relationships must be provided to the model in advance, which requires domain knowledge, and this knowledge is not always readily available. Summary of the Invention
[0010] The purpose of this invention is to provide a method and apparatus for designing multifunctional protein fragment scaffolds based on a diffusion model. By adding rotational and translational noise to multiple protein functional fragments to maintain their rigid body kinematics, the noisy protein structure is optimized during the denoising process, resulting in a complete protein structure containing multiple protein functional fragments. This invention innovatively proposes a floating anchored diffusion model that can maintain the rigid body motion of protein functional fragments, enabling automated and highly flexible design of multifunctional protein fragment scaffolds.
[0011] To achieve the above-mentioned objectives, the present invention provides the following technical solution:
[0012] In a first aspect, the present invention provides a method for designing a protein multifunctional fragment scaffold based on a diffusion model, comprising the following steps:
[0013] Step 1: Construct a training sample set containing known protein structures, including protein functional fragments and protein scaffolds, and build a floating anchored diffusion model;
[0014] Step 2: In the forward noise addition process of the floating anchor diffusion model, noise is added to all amino acid positions in the known protein structure to obtain the noisy protein structure. The noise includes rotational noise and translational noise. The rotational noise and translational noise in the protein functional fragment are respectively passed through the rigid body noise design module to obtain rigid body rotational noise and rigid body translational noise used to keep the protein functional fragment in rigid body motion.
[0015] Step 3: In the reverse denoising process of the floating anchored diffusion model, the predicted noise of all amino acid positions in the noisy protein structure is obtained. The predicted noise includes predicted rotation noise and predicted translation noise. The predicted rotation noise and predicted translation noise within the protein functional fragment are respectively processed by the rigid body noise design module to obtain rigid body predicted rotation noise and rigid body predicted translation noise used to maintain the rigid body motion of the protein functional fragment. By minimizing the loss between the predicted noise and the added noise, the noisy protein structure is optimized to obtain the complete protein structure.
[0016] Step 4: Iterate through steps 2 and 3 until the preset number of iterations is reached to obtain a trained floating anchored diffusion model for the design of protein multifunctional fragment scaffolds.
[0017] Furthermore, in step 1, the protein scaffold is the location of free amino acids, and the protein functional fragment contains any number of amino acids. Preferably, the protein functional fragment contains 10 to 80 amino acids.
[0018] Furthermore, in step 2, noise is added to all amino acid positions within the known protein structure. The added noise is random Gaussian noise that satisfies rigid body kinematics. By sampling random Gaussian noise that satisfies rigid body kinematics, the resulting rigid body rotation noise and rigid body translation noise will still satisfy rigid body kinematics when the rotation noise and translation noise of amino acid positions within the protein functional fragment are subsequently fused using a rigid body noise design module.
[0019] Furthermore, in step 2, the rotational noise and translational noise within the protein functional fragment are respectively processed through the rigid body noise design module to obtain rigid body rotational noise and rigid body translational noise used to maintain the rigid body motion of the protein functional fragment, specifically as follows:
[0020] In the rigid body noise design module, rotational noise at all amino acid positions within a protein functional fragment is fused to obtain rigid body rotational noise, expressed by the formula:
[0021]
[0022] Among them, R Δ ' represents rigid body rotation noise, and R represents the rotation matrix of the amino acid positions. This represents the average rotational noise at all amino acid positions within a functional segment of a protein.
[0023] In the rigid body noise design module, translational noise at all amino acid positions within a protein functional fragment is fused to obtain rigid body translational noise, expressed by the formula:
[0024]
[0025] Among them, X Δ "" represents rigid body translation noise, and X represents the translation vector of the amino acid position. This represents the average translational noise of all amino acid positions within a protein functional fragment. By fusing the rotational and translational noise of all amino acid positions within the protein functional fragment separately, the resulting rigid body rotational and translational noises act on the entire protein functional fragment, thus treating the protein functional fragment as a whole, ensuring that the positions of all internal amino acids remain unchanged during rotation or translation.
[0026] Furthermore, in step 2, the rigid body rotation noise acts on the protein functional fragment, causing the protein functional fragment to undergo rigid body translation. The first segment translation vector used to describe the rigid body translation of the protein functional fragment is expressed by the formula:
[0027]
[0028] Among them, X Δ ' represents the translation vector of the first segment. By quantifying the first translation vector caused by rigid body rotation noise, it is convenient to adjust the displacement of protein functional segments as needed. In actual inference, it is possible to automatically design the relative positions of multiple protein functional segments while keeping the internal structure of the protein functional segments unchanged, thus improving the flexibility of protein structure design.
[0029] Furthermore, in step 3, minimizing the loss between the predicted noise and the added noise includes:
[0030] Minimize the loss between rotational noise and predicted rotational noise, and between translational noise and predicted translational noise on the protein scaffold;
[0031] Minimize the loss between rigid body rotation noise and rigid body predicted rotation noise, rigid body translation noise and rigid body predicted translation noise on protein functional fragments;
[0032] Noise prediction is completed when all losses on the protein scaffold and protein functional fragments are minimized.
[0033] Furthermore, in step 4, when the trained floating anchored diffusion model is used for actual inference, given any number of protein functional fragments, the amino acid positions on the protein scaffold are randomly sampled. The number of protein functional fragments and the randomly sampled amino acid positions are input into the trained floating anchored diffusion model. After the reverse denoising process, the complete protein structure containing the number of protein functional fragments is obtained.
[0034] Secondly, in order to achieve the above-mentioned objectives, the present invention also provides a protein multifunctional fragment scaffold design device based on a diffusion model, including a training set construction module, a forward noise addition module, a reverse noise reduction module, and an iterative training module.
[0035] The training set construction module is used to construct a training sample set containing known protein structures, including protein functional fragments and protein scaffolds, to construct a floating anchored diffusion model.
[0036] The forward noise module is used to add noise to all amino acid positions within a known protein structure during the forward noise process of the floating anchor diffusion model, thereby obtaining a noisy protein structure. The noise includes rotational noise and translational noise. The rotational noise and translational noise within the protein functional fragment are respectively passed through the rigid body noise design module to obtain rigid body rotational noise and rigid body translational noise used to maintain the rigid body motion of the protein functional fragment.
[0037] The reverse denoising module is used in the reverse denoising process of the floating anchored diffusion model to obtain the predicted noise at all amino acid positions in the noisy protein structure. The predicted noise includes predicted rotation noise and predicted translation noise. The predicted rotation noise and predicted translation noise within the protein functional fragment are respectively processed by the rigid body noise design module to obtain rigid body predicted rotation noise and rigid body predicted translation noise to keep the protein functional fragment in rigid body motion. By minimizing the loss between the predicted noise and the added noise, the noisy protein structure is optimized to obtain the complete protein structure.
[0038] The iterative training module is used to iterate the forward noise addition module and the reverse noise reduction module until a preset number of iterations is reached to obtain a trained floating anchored diffusion model, which is used for the design of protein multifunctional fragment scaffolds.
[0039] Thirdly, to achieve the above-mentioned objectives, embodiments of the present invention also provide a protein multifunctional fragment scaffold design device based on a diffusion model, including a memory and a processor. The memory is used to store a computer program, and the processor is used to implement the protein multifunctional fragment scaffold design method based on a diffusion model provided in the first aspect of the present invention when the computer program is executed.
[0040] Fourthly, to achieve the above-mentioned objectives, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, wherein the computer program, when used with a computer, implements the protein multifunctional fragment scaffold design method based on a diffusion model provided in the first aspect of the present invention embodiments.
[0041] The beneficial effects of this invention are as follows:
[0042] (1) This invention fuses the noise at all amino acid positions within a protein functional fragment to obtain rigid body rotation noise and rigid body translation noise to describe the entire protein functional fragment, thereby treating the protein functional fragment as a rigid body and maintaining its internal structure during its movement to obtain a protein structure containing the desired function. In addition, based on the design concept of treating the protein functional fragment as a rigid body, the method of this invention is also applicable to the task of restoring the complete structure of several protein functional fragments.
[0043] (2) This invention quantifies the translation of protein functional fragments caused by rigid body rotation noise and describes it as the translation vector of the first fragment. This allows for on-demand adjustment, thereby automatically designing the relative positions between multiple protein functional fragments, solving the problem of designing scaffolds for multiple protein functional fragments, and improving the flexibility of protein structure design. Attached Figure Description
[0044] Figure 1 This is a flowchart of a protein multifunctional fragment scaffold design method based on a diffusion model provided in an embodiment of the present invention.
[0045] Figure 2 This is a detailed flowchart of the protein multifunctional fragment scaffold design method based on the diffusion model provided in the embodiments of the present invention.
[0046] Figure 3 This is a flowchart of the reverse denoising process of the noisy protein structure during the actual inference stage provided in the embodiments of the present invention. Detailed Implementation
[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of the invention.
[0048] like Figure 1 As shown in the example, this embodiment provides a method for designing a multifunctional protein fragment scaffold based on a diffusion model, including the following steps:
[0049] S110, Construct a training sample set containing known protein structures, including protein functional fragments and protein scaffolds, and construct a floating anchored diffusion model.
[0050] A large amount of biological data containing known protein structures was collected to construct a training sample set. These known protein structures mainly include protein functional fragments and protein scaffolds. Protein functional fragments carry the functions of the complete protein; that is, multifunctional proteins contain multiple protein functional fragments. Protein scaffolds are framework structures composed of a large number of free amino acids that support the protein functional fragments arranged in a specific order. Protein functional fragments also contain a large number of amino acids.
[0051] Based on the diffusion model, a floating anchor diffusion model (FADiff) is constructed for the design of multifunctional protein fragment scaffolds.
[0052] S120, in the forward noise addition process of the floating anchored diffusion model, noise is added to all amino acid positions in the known protein structure to obtain a noisy protein structure. The noise includes rotational noise and translational noise. The rotational noise and translational noise in the protein functional fragment are respectively passed through the rigid body noise design module to obtain rigid body rotational noise and rigid body translational noise used to keep the protein functional fragment in rigid body motion.
[0053] like Figure 2 As shown, a multifunctional known protein structure is randomly selected from the training sample set. It contains multiple protein functional fragments and a large number of free amino acids. The large number of free amino acids constitute the protein scaffold. All amino acid positions are abstracted as being composed of rotation matrix R and translation vector X.
[0054] In the forward noise addition process of FADiff, Gaussian noise satisfying SO(3) is randomly sampled, including rotational noise R. Δ Translational noise X Δ Adding amino acids to all positions of a known, multifunctional protein structure yields a noisy protein structure.
[0055] Among them, there is rotational noise R at the amino acid positions in the protein scaffold. Δ Translational noise X Δ This allows all free amino acids in the protein scaffold to rotate or translate. For protein functional fragments, to ensure their structure remains unchanged during rotation or translation, this invention proposes a rigid body noise design module that treats the protein functional fragment as a rigid body and maintains rigid body kinematics during its motion.
[0056] Specifically, the rotational noise at all amino acid positions in the protein functional fragment is fused using a rigid body noise design module to obtain fragment rotational noise that describes the protein functional fragment. in, R represents the average rotational noise at all amino acid positions in a functional protein fragment. -1 This is the inverse of the rotation matrix. Based on the fragment rotation noise, we can further obtain the first fragment translation vector used to describe the translation amount of the entire protein functional fragment. Fragment rotation noise R of protein functional fragments Δ 'and segment translation vector X Δ This allows protein functional fragments to move while maintaining their structure, thus satisfying rigid body kinematics.
[0057] Similarly, the translation noise at all amino acid positions in the protein functional fragment is fused using a rigid body noise design module to obtain fragment translation noise used to describe the protein functional fragment. in This represents the average translation noise at all amino acid positions in a functional protein fragment.
[0058] S130, in the reverse denoising process of the floating anchored diffusion model, the predicted noise of all amino acid positions in the noisy protein structure is obtained. The predicted noise includes predicted rotation noise and predicted translation noise. The predicted rotation noise and predicted translation noise in the protein functional fragment are respectively passed through the rigid body noise design module to obtain rigid body predicted rotation noise and rigid body predicted translation noise used to keep the protein functional fragment in rigid body motion. By minimizing the loss between the predicted noise and the added noise, the noisy protein structure is optimized to obtain the complete protein structure.
[0059] The noisy protein structure is input into the reverse denoising process of FADiff, such as... Figure 2 As shown, the predicted noise for all amino acid positions in the noisy protein structure output by FADiff includes predicted rotation noise. And predict translation noise
[0060] For protein functional fragments, the predicted rotation noise and predicted translation noise at all amino acid positions are input into the rigid body noise design module, and the predicted rotation noise is fused to obtain the rigid body predicted rotation noise that keeps the protein functional fragment in rigid body motion. in This is the average value. This is the rotation matrix after adding noise. Rigid body prediction rotation noise. This brings the translation vector of the second segment. in This is the average value. This is the translation vector after adding noise. This allows the noisy protein structure to maintain the translational amount of rigid body motion as a whole during the reverse denoising process, which helps to restore the complete protein structure.
[0061] Similarly, by fusing the predicted translation noise of all amino acid positions on the protein functional fragment into the rigid body noise design module, a rigid body predicted translation noise is obtained that allows the protein functional fragment to maintain rigid body motion. in, This represents the average predicted translation noise at each amino acid position on a functional protein fragment.
[0062] Through a forward noise addition process and a backward noise reduction process, the predicted noise and the added noise are made as equal as possible. Figure 2 As shown, the loss between rotational noise and predicted rotational noise, translational noise and predicted translational noise on the protein scaffold, rigid body rotational noise and rigid body predicted rotational noise on the protein functional fragment, and rigid body translational noise and rigid body predicted translational noise is minimized. Then, by subtracting all the noise from the reverse process from the noisy protein structure, the complete protein structure can be obtained.
[0063] S140, iterate through S120 and S130 until the preset number of iterations is reached to obtain a trained floating anchored diffusion model, which is used for the design of protein multifunctional fragment scaffolds.
[0064] Known protein structures are selected multiple times from the training sample set. The resulting functional protein fragments and scaffolds are then subjected to a forward noise addition process (S120) and a reverse noise reduction process (S130). When the preset number of iterations is reached, the trained FADiff is obtained.
[0065] When using the trained FADiff for actual inference, such as Figure 3As shown in the figure, this embodiment uses two protein functional fragments, virtual-1 and virtual-2, represented by orange and blue respectively. The green part represents the protein scaffold composed of a large number of free amino acid sites. The two protein functional fragments and the protein scaffold are input into the trained FADiff. After multiple rounds of reverse denoising, the noise is reduced step by step, and finally the complete protein structure containing the input two preset protein functional fragments is output.
[0066] Based on the same inventive concept, this invention also provides a protein multifunctional fragment scaffold design device based on a diffusion model, including a training set construction module, a forward noise addition module, a reverse noise reduction module, and an iterative training module.
[0067] The training set construction module is used to construct a training sample set containing known protein structures, including protein functional fragments and protein scaffolds, to build a floating anchored diffusion model.
[0068] The forward noise module is used to add noise to all amino acid positions in a known protein structure during the forward noise process of the floating anchor diffusion model, resulting in a noisy protein structure. The noise includes rotational noise and translational noise. The rotational noise and translational noise in the protein functional fragment are respectively passed through the rigid body noise design module to obtain rigid body rotational noise and rigid body translational noise used to keep the protein functional fragment in rigid body motion.
[0069] The reverse denoising module is used in the reverse denoising process of the floating anchored diffusion model to obtain the predicted noise at all amino acid positions in the noisy protein structure. The predicted noise includes predicted rotation noise and predicted translation noise. The predicted rotation noise and predicted translation noise within the protein functional fragment are respectively processed by the rigid body noise design module to obtain rigid body predicted rotation noise and rigid body predicted translation noise to keep the protein functional fragment in rigid body motion. By minimizing the loss between the predicted noise and the added noise, the noisy protein structure is optimized to obtain the complete protein structure.
[0070] The iterative training module is used to iterate the forward noise addition module and the backward noise reduction module until the preset number of iterations is reached, resulting in a trained floating anchored diffusion model for the design of protein multifunctional fragment scaffolds.
[0071] Regarding the protein multifunctional fragment scaffold design device based on the diffusion model provided in this embodiment of the invention, since it basically corresponds to the method embodiment, relevant details can be found in the description of the method embodiment. The device embodiments described above are merely illustrative. The units described as separating 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 across multiple network units. Some or all of the modules can be selected to achieve the purpose of this invention according to actual needs. Those skilled in the art can understand and implement this invention without any creative effort.
[0072] Based on the same inventive concept, the embodiment also provides a protein multifunctional fragment scaffold design device based on a diffusion model, including a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to implement the above-described protein multifunctional fragment scaffold design method based on a diffusion model when the computer program is executed.
[0073] The protein multifunctional fragment scaffold design device based on the diffusion model proposed in this invention can be a device such as a computer. The device embodiment can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions from non-volatile memory into memory and executing them. From a hardware perspective, in addition to the processor, memory, network interface, and non-volatile memory, the protein multifunctional fragment scaffold design device based on the diffusion model provided in this invention may also include other hardware depending on the actual functions of the device with data processing capabilities; these will not be elaborated further.
[0074] Based on the same inventive concept, the embodiment also provides a computer-readable storage medium storing a computer program, which, when used by a computer, implements the above-described diffusion-based protein multifunctional fragment scaffold design method.
[0075] The computer-readable storage medium can be an internal storage unit of any data processing device as described in any of the foregoing embodiments, such as a hard disk or memory. Furthermore, the computer-readable storage medium may include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.
[0076] It should be noted that the diffusion-based protein multifunctional fragment scaffold design device, diffusion-based protein multifunctional fragment scaffold design equipment, and computer-readable storage medium provided in the above embodiments all belong to the same concept as the diffusion-based protein multifunctional fragment scaffold design method embodiments. For details of their implementation process, please refer to the diffusion-based protein multifunctional fragment scaffold design method embodiments, which will not be repeated here.
[0077] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the implementation process of the present invention has been described in detail above, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for designing multifunctional protein fragment scaffolds based on a diffusion model, characterized in that, Includes the following steps: Step 1: Construct a training sample set containing known protein structures, including protein functional fragments and protein scaffolds, and build a floating anchored diffusion model; Step 2: In the forward noise addition process of the floating anchored diffusion model, noise is added to all amino acid positions within the known protein structure to obtain a noisy protein structure. The noise includes rotational noise and translational noise. Specifically, the rotational and translational noise within the protein functional fragments are processed through a rigid body noise design module to obtain rigid body rotational noise and rigid body translational noise used to maintain the rigid body motion of the protein functional fragments. In the rigid body noise design module, rotational noise at all amino acid positions within a protein functional fragment is fused to obtain rigid body rotational noise, expressed by the formula: in, Let R represent the rotational noise of the rigid body, and let R represent the rotation matrix at the amino acid positions. This represents the average rotational noise at all amino acid positions within a functional segment of a protein. In the rigid body noise design module, translational noise at all amino acid positions within a protein functional fragment is fused to obtain rigid body translational noise, expressed by the formula: in, Let X represent the translation noise of the rigid body, and let X represent the translation vector of the amino acid position. This represents the average translation noise at all amino acid positions within a functional segment of a protein. Step 3: In the reverse denoising process of the floating anchored diffusion model, the predicted noise of all amino acid positions in the noisy protein structure is obtained. The predicted noise includes predicted rotation noise and predicted translation noise. The predicted rotation noise and predicted translation noise within the protein functional fragment are respectively processed by the rigid body noise design module to obtain rigid body predicted rotation noise and rigid body predicted translation noise used to maintain the rigid body motion of the protein functional fragment. By minimizing the loss between the predicted noise and the added noise, the noisy protein structure is optimized to obtain the complete protein structure. Step 4: Iterate through steps 2 and 3 until the preset number of iterations is reached to obtain a trained floating anchored diffusion model for the design of protein multifunctional fragment scaffolds.
2. The protein multifunctional fragment scaffold design method based on a diffusion model according to claim 1, characterized in that, In step 1, the protein scaffold is the location of free amino acids, and the protein functional fragment contains any number of amino acids.
3. The protein multifunctional fragment scaffold design method based on a diffusion model according to claim 1, characterized in that, In step 2, noise is added to all amino acid positions within the known protein structure. The added noise is random Gaussian noise that satisfies rigid body kinematics.
4. The protein multifunctional fragment scaffold design method based on a diffusion model according to claim 1, characterized in that, In step 2, rigid body rotation noise acts on the protein functional fragment, causing the fragment to undergo rigid body translation. The first segment translation vector, used to describe this rigid body translation, is expressed by the formula: in, This represents the translation vector of the first segment.
5. The protein multifunctional fragment scaffold design method based on a diffusion model according to claim 1, characterized in that, Step 3, minimizing the loss between the predicted noise and the added noise, includes: Minimize the loss between rotational noise and predicted rotational noise, and between translational noise and predicted translational noise on the protein scaffold; Minimize the loss between rigid body rotation noise and rigid body predicted rotation noise, rigid body translation noise and rigid body predicted translation noise on protein functional fragments; Noise prediction is completed when all losses on the protein scaffold and protein functional fragments are minimized.
6. The protein multifunctional fragment scaffold design method based on a diffusion model according to claim 1, characterized in that, In step 4, when the trained floating anchored diffusion model is used for actual inference, given any number of protein functional fragments, the amino acid positions on the protein scaffold are randomly sampled. The protein functional fragments and the randomly sampled amino acid positions are input into the trained floating anchored diffusion model. After the reverse denoising process, the complete protein structure containing the protein functional fragments is obtained.
7. A protein multifunctional fragment scaffold design device based on a diffusion model, characterized in that, It includes a training set construction module, a forward noise addition module, a backward noise reduction module, and an iterative training module; The training set construction module is used to construct a training sample set containing known protein structures, including protein functional fragments and protein scaffolds, to construct a floating anchored diffusion model. The forward noise addition module is used to add noise to all amino acid positions within a known protein structure during the forward noise addition process of the floating anchored diffusion model, resulting in a noisy protein structure. The noise includes rotational noise and translational noise. Specifically, the rotational and translational noise within the protein functional fragments are respectively processed by the rigid body noise design module to obtain rigid body rotational noise and rigid body translational noise used to maintain the rigid body motion of the protein functional fragments. In the rigid body noise design module, rotational noise at all amino acid positions within a protein functional fragment is fused to obtain rigid body rotational noise, expressed by the formula: in, Let R represent the rotational noise of the rigid body, and let R represent the rotation matrix at the amino acid positions. This represents the average rotational noise at all amino acid positions within a functional segment of a protein. In the rigid body noise design module, translational noise at all amino acid positions within a protein functional fragment is fused to obtain rigid body translational noise, expressed by the formula: in, Let X represent the translation noise of the rigid body, and let X represent the translation vector of the amino acid position. This represents the average translation noise at all amino acid positions within a functional segment of a protein. The reverse denoising module is used in the reverse denoising process of the floating anchored diffusion model to obtain the predicted noise at all amino acid positions in the noisy protein structure. The predicted noise includes predicted rotation noise and predicted translation noise. The predicted rotation noise and predicted translation noise within the protein functional fragment are respectively processed by the rigid body noise design module to obtain rigid body predicted rotation noise and rigid body predicted translation noise to keep the protein functional fragment in rigid body motion. By minimizing the loss between the predicted noise and the added noise, the noisy protein structure is optimized to obtain the complete protein structure. The iterative training module is used to iterate the forward noise addition module and the reverse noise reduction module until a preset number of iterations is reached to obtain a trained floating anchored diffusion model, which is used for the design of protein multifunctional fragment scaffolds.
8. A device for designing multifunctional protein fragment scaffolds based on a diffusion model, comprising a memory and a processor, wherein the memory is used to store a computer program, characterized in that, The processor is configured to implement the protein multifunctional fragment scaffold design method based on the diffusion model as described in any one of claims 1-6 when executing the computer program.
9. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is used, it implements the protein multifunctional fragment scaffold design method based on the diffusion model as described in any one of claims 1-6.