A method and device for designing a short peptide of a protease substrate based on an eve model

Abstract

This application relates to a method and apparatus for designing short peptides of protease substrates based on an EVE model. The method includes: collecting sequence data of the natural substrates of the target protease; training the EVE model, calculating the mutation score for each possible mutation using the trained model, summing the mutation scores to obtain the evolutionary score of the complete sequence; mapping the training set to a latent variable space, fitting the latent variable distribution, sampling and decoding the latent variables, performing temperature scaling and position-by-position sampling to generate candidate sequences; calculating the evolutionary score of the candidate sequences, performing preliminary clustering using cd-hit, selecting the sequence with the lowest evolutionary score in the sequence cluster and calculating the similarity between pairs of sequences, performing dimensionality reduction using PCoA and then performing clustering using K-Means; selecting an initial representative sequence based on proximity to the centroid; if the initial representative sequence is LQ at positions P2 and P1, then calculate; if there is a non-LQ sequence with a distance less than the centroid, then select it as the representative sequence of the cluster.

Description

Technical Field

[0001] This application relates to the field of biotechnology, and in particular to a method and apparatus for designing short peptides as protease substrates based on the EVE model. Background Technology

[0002] The design of protease substrate peptides is of great significance in biomedical research, drug development, and biotechnology applications. Protease substrate peptides are widely used in enzyme activity detection, drug screening, biosensor development, and protein function research. These short peptides typically contain specific amino acid sequences that can be specifically recognized and cleaved by target proteases; the cleavage efficiency directly affects the sensitivity and accuracy of related applications.

[0003] Traditional protease substrate peptide design methods mainly rely on strategies such as empirical design based on known natural substrate sequences, rational design guided by structural biology, and random mutation screening. However, these methods have significant limitations: limited design space, making it difficult to systematically explore novel substrate sequence combinations; lack of accurate sequence-enzyme digestion efficiency prediction capabilities, resulting in low design success rates; traditional methods often rely on extensive experimental screening, leading to high time and economic costs; and difficulty in ensuring substrate sequence diversity while maintaining high digestion efficiency.

[0004] In recent years, deep learning-based protein sequence design methods have developed rapidly, providing new ideas for solving the aforementioned problems. EVE (Evolutionary Model of Variant Effect) is a protein evolution model based on variational autoencoders (VAEs) that can predict mutation effects by learning evolutionary constraints on protein sequences. This model analyzes a large number of natural sequence variations to identify beneficial and harmful mutations and quantify the impact of sequence changes on protein function. Summary of the Invention

[0005] Based on this technology, this invention proposes a method and apparatus for designing short peptides of protease substrates based on the EVE model. It aims to overcome the limitations of traditional design methods, achieve efficient and diverse protease substrate sequence generation and optimization, and provide high-quality substrate sequence candidates for protease-related research.

[0006] Specifically, this application is implemented through the following technical solution: This application provides a method for designing short peptides as protease substrates based on the EVE model, including: Collect sequence data of the natural substrates of the target protease; The EVE model is trained using the sequence data. The mutation score for each possible mutation is calculated using the trained model. The mutation scores are summed to obtain the evolution score of the complete sequence. The training set is mapped to the latent variable space, the latent variable distribution is fitted, the latent variables are sampled and decoded, and candidate sequences are generated by temperature scaling and bit-by-bit sampling. The evolutionary score of the candidate sequences is calculated, and preliminary clustering is performed using cd-hit. The sequence with the lowest evolutionary score in the sequence cluster is selected. The similarity between pairs of sequences in the sequence with the lowest evolutionary score is calculated. Based on the similarity, dimensionality reduction is performed using PCoA, and then clustering is performed using K-Means. The initial representative sequence is selected based on its closest distance to the centroid. If the initial representative sequence is L at position P2 and Q at position P1, then the 15% lower quantile of the distances from all sequences within the cluster to the centroid is calculated. If there exists a distance to the centroid less than 1 / 3 ... For non-LQ sequences, the non-LQ sequence closest to the centroid is selected as the representative sequence of the cluster; where P1: the substrate amino acid residue located on the N-terminus of the cleavage bond and immediately adjacent to the cleavage bond. P2: the second residue of the substrate amino acid located on the N-terminus of the cleavage bond and immediately adjacent to the cleavage bond. LQ: a subsequence in which the amino acids are arranged in the order of leucine-glutamine.

[0007] Optionally, the step of mapping the training set to the latent variable space, fitting the latent variable distribution, sampling and decoding the latent variables, and performing temperature scaling and bit-by-bit sampling to generate candidate sequences includes: Calculate the mean and standard deviation of the latent space of all training sets to verify whether the latent vectors of each dimension conform to the assumption of a standard normal distribution; Establish latent variables obtained from sampling the standard normal distribution; The latent variables are input into the decoder of the EVE model, and the decoder receives the latent variables and outputs the unnormalized logarithm fraction of each position for each symbol of the amino acid alphabet. The unnormalized logarithmic fraction is scaled by temperature and then converted into a probability distribution of legal amino acids; and The probability distribution is transformed into candidate sequences through multivariate sampling and character mapping.

[0008] Optionally, the step of selecting the initial representative sequence based on proximity to the centroid includes: Based on the 2D PCoA coordinates of the clusters after K-Means clustering, the mean of the PCoA coordinates of all sequences within the cluster is taken as the centroid. The Euclidean distance from each sequence within the cluster to the centroid is calculated, resulting in a "sequence-distance" list for each cluster; and For each cluster, sort the "sequence-distance" sequences in ascending order of distance, and select the first sequence after sorting as the initial representative sequence.

[0009] Optionally, the mutation score is obtained by calculating the logarithmic difference between the conditional probability of the mutated amino acid and the conditional probability of the wild-type amino acid.

[0010] Optionally, ,in N is the total number of sequences within the cluster. It is a rounding function. It is a set of distances arranged in ascending order.

[0011] Optionally, clustering using K-Means after dimensionality reduction via PCoA based on the similarity includes: performing eigenvalue decomposition on the distance matrix, selecting the first two principal coordinates to construct a two-dimensional representation space; and using the K-Means algorithm to cluster in the two-dimensional representation space to determine the optimal cluster centers and sequence assignments.

[0012] Optionally, the method further includes the step of filtering human-derived sequences.

[0013] Another aspect of this application provides a device for designing short peptides as protease substrates based on the EVE model, comprising: The data collection module is used to collect sequence data of the natural substrates of the target protease; The evolution score acquisition module is used to train an EVE model using the sequence data, calculate the mutation score for each possible mutation using the trained model, and sum the mutation scores to obtain the evolution score of the complete sequence. The candidate sequence generation module is used to fit a normal distribution through the latent space of the training set and generate candidate sequences by mapping the randomness of standard normal sampling based on the temperature hyperparameter. A multi-level clustering module is used to calculate the evolutionary score of the candidate sequences, perform preliminary clustering using cd-hit, filter out the sequences with the lowest evolutionary scores in the sequence clusters, calculate the similarity between pairs of sequences in the sequence with the lowest evolutionary scores, and perform K-Means clustering based on the similarity after dimensionality reduction using PCoA; and The representative sequence determination module is used to select the initial representative sequence based on the closest distance to the centroid. If the initial representative sequence is L at position P2 and Q at position P1, then the 15% lower quantile of the distances from all sequences in the cluster to the centroid is calculated. If there exists a distance to the centroid less than 1 / 3 ... For non-LQ sequences, the non-LQ sequence closest to the centroid is selected as the representative sequence of the cluster; where P1: the substrate amino acid residue located on the N-terminus of the cleavage bond and adjacent to the cleavage bond; P2: the second residue of the substrate amino acid located on the N-terminus of the cleavage bond and adjacent to the cleavage bond; LQ: the subsequence in which the amino acid sequence is arranged as leucine-glutamine.

[0014] In another aspect, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0015] In another aspect, this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method.

[0016] The EVE model-based design method for protease substrate short peptides of this invention collects natural substrate sequence data of target proteases, uses the EVE model to learn the evolutionary constraints and functional relationships of sequences, establishes a sequence generation algorithm based on latent space sampling, and combines a multi-level clustering optimization strategy to achieve specialized optimization for protease-substrate interactions, precisely control the balance between functionality and diversity of generated sequences, and provides a systematic sequence evaluation and screening strategy, thereby achieving intelligent design of efficient substrate sequences. Attached Figure Description

[0017] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating the design method for protease substrate short peptides based on the EVE model provided in this specification. Figure 2 This is a schematic diagram of the design device for protease substrate short peptides based on the EVE model provided in this specification. Figure 3 This is a schematic diagram of the entire workflow framework for the design method of protease substrate short peptides based on the EVE model provided in this specification; Figure 4a and Figure 4b Let be the mean distribution and the standard deviation distribution of the latent space; Figure 5 The results of K-Means clustering in the sequence determination module are presented qualitatively based on PCoA. Color represents the clustered categories, and shape indicates whether P2 and P1 are L and Q, respectively. Figure 6a and Figure 6b The 30 sequence logo images used for wet experiments and the sequence logo images of training data were selected from the Nsp5 restriction site (P6-P2') examples of the novel coronavirus. Figure 7 The correlation between logistic regression pseudoprobability and Best RFU / t; Figure 8a and Figure 8bThe bar chart shows the cut efficiency of the sequence and the relative cut efficiency compared to the mean of the positive control. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0019] The technical solutions provided in the various embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0020] Figure 1 This is a schematic flowchart of the protease substrate short peptide design method based on the EVE model provided in this specification. Figure 3 This is a schematic diagram illustrating the overall workflow of the EVE-based protease substrate short peptide design method provided in this specification. The EVE-based protease substrate short peptide design method provided in this specification includes the following steps: S101: Collect sequence data of the natural substrate of the target protease; Natural substrate sequence data of target proteases can be collected using specialized databases. Taking the novel coronavirus as an example, sequences of SARS-CoV-2 were collected from NCBI Virus, and sequences of Nsp5 restriction sites (P6-P2') were extracted, totaling 261 sequences.

[0021] Figure 6a and Figure 6b The image shows the Logo diagrams of 30 sequences selected for wet experiments and the sequence Logo diagrams of training data from the Nsp5 restriction site (P6-P2') of the novel coronavirus.

[0022] S102: Train the EVE model using sequence data, calculate the mutation score for each possible mutation using the trained model, sum the mutation scores to obtain the evolution score of the complete sequence.

[0023] Specifically, prepare input data (multiple sequence alignment files) that conform to the specified format, and then run the training script (train_VAE) provided by EVE to train the model; calculate the mutation score for each possible mutation using the script (compute_evol_indices) provided by EVE based on the trained model; sum the mutation scores to obtain the evolutionary score of the complete sequence by referring to the EVEscape method.

[0024] More specifically, firstly, for each possible single-point mutation in the sequence, its mutation score is calculated. This score is obtained by calculating the negative log-likelihood ratio of the mutated sequence relative to the wild-type sequence, using the following formula: ,in x i Indicates the sequence number i One location, a j This represents the mutated amino acid. a WT This indicates wild-type amino acids. x -i Indicates except the first i The system calculates the fitness score of the complete sequence by weighted summation after obtaining the mutation score at each position outside the sequence. .

[0025] By utilizing the variational autoencoder architecture of the EVE model, functional constraint patterns in protease substrate sequences are learned through analysis of evolutionary information from a large number of natural substrate sequences. These functional constraints (such as cleavage site specificity, sequence length preference, and conservation of key residues) are hidden within the evolutionary patterns of numerous homologous sequences. VAE extracts these constraints through an "encode-decode" process. By learning the variation patterns of a large number of natural substrate sequences, the model ensures that the probability distribution output by the decoder aligns with evolutionarily "acceptable" variations: functionally important positions (such as cleavage sites) have a concentrated probability distribution (high conservation), while less important positions are more dispersed (allowing for more variation). The EVE model quantitatively assesses the impact of sequence variation on fitness by calculating the lower bound of evidence (ELBO), establishing a quantitative relationship between sequence variation and functional effects. Combining ELBO and conditional probability, EVE correlates sequence variation with protease substrate function (such as cleavage efficiency and binding affinity) through the following steps: calculating the ΔELBO of the variation; correlating ΔE with the functional effect; and, to establish a more precise quantitative relationship, incorporating measurements such as mutant cleavage efficiency to further refine the ΔE. E / Δ ELBO Calibration is performed. For example, a VAE is trained on the sequence S of this novel coronavirus protein family to learn its distribution. p ( S ), and approximate the log-likelihood with the lower bound of evidence: log p ( S ) ≥ ELBO ( S ).because ELBO ( S ) is log p ( S The lower bound of ) and mutants S Relative wild type WT The fitness can be approximated by the log-likelihood ratio:

[0026] Probabilistic scoring is obtained through monotonic mapping. .

[0027] 103: Map the training set to the latent variable space, fit the latent variable distribution, sample and decode the latent variables, and perform temperature scaling and bit-by-bit sampling to generate candidate sequences.

[0028] Specifically, the training set encoding mean is estimated to follow a multivariate normal distribution in the latent variable space, and the latent variables are sampled from the standard normal distribution. The latent variables are then input into the decoder to obtain the unnormalized logarithmic fractions of each position relative to the amino acid alphabet. These fractions are then temperature-scaled and normalized to a probability distribution, and candidate sequences are generated by sampling position by position.

[0029] More specifically, the latent space state is the result of the Encoder's computation ( Figure 3 The sampling process is a simple sampling from a high-dimensional (20-dimensional) standard normal distribution. First, the latent space distribution is verified and key parameters are calculated to provide a basis for subsequent generation. The mean and standard deviation of the latent space of all training sets are calculated to verify whether the latent vectors of each dimension conform to the assumption of a standard normal distribution. Figure 4a and Figure 4b The normal distribution can be written as Z~N(0, 1). We establish latent variables obtained from sampling the standard normal distribution. z Hidden variables z The input model decoder generates candidate sequences. The diversity of generated sequences is controlled by the temperature hyperparameter. Higher temperatures generate more diverse sequences but also have higher evolutionary scores. Lower temperatures generate less diverse sequences that are more conserved relative to naturally occurring sequences but also have lower evolutionary scores.

[0030] S104: Calculate the evolutionary score of the candidate sequences, perform preliminary clustering using cd-hit, select the sequence with the lowest evolutionary score in the sequence cluster, calculate the similarity between pairs of sequences in the sequence with the lowest evolutionary score, and perform K-Means clustering based on the similarity after dimensionality reduction using PCoA.

[0031] Multilevel clustering analysis is a key technique for achieving a balance between sequence diversity and functionality. The first level employs the CD-HIT algorithm for preliminary clustering based on sequence similarity. The CD-HIT algorithm constructs clusters by calculating the similarity between sequences; the similarity calculation uses the ratio of the number of matching residues to the length of the shorter sequence. In practical applications, the sequence similarity threshold is set to 0.65, representing 65% sequence similarity. After initial clustering, each sequence cluster is analyzed, and the sequence with the lowest evolutionary score (i.e., optimal) in each cluster is selected as the representative of that cluster, while sequences with evolutionary scores higher than the threshold are filtered out. The threshold is a hyperparameter that can be adjusted manually based on experimental results. This step effectively reduces sequence redundancy while retaining functionally optimal sequences. The second-level clustering uses a more refined method, first constructing a sequence similarity matrix. This matrix is ​​calculated through global sequence alignment, using the BLOSUM80 amino acid substitution matrix to evaluate the similarity between sequence pairs, and introducing a gap penalty mechanism (open gap penalty -10.0, extended gap penalty -0.5). The specific calculation formula is as follows:

[0032] The alignment_score is the best alignment score obtained using the Needleman-Wunsch algorithm. The final distance matrix is ​​obtained through the following transformation: .

[0033] To perform clustering analysis effectively in a low-dimensional space, principal coordinate analysis (PCoA) is used to map the distance matrix. By performing eigenvalue decomposition on the distance matrix, the first two principal coordinates are selected to construct a two-dimensional representation space. Where U is the eigenvector matrix and Λ is the eigenvalue diagonal matrix. In the two-dimensional representation space, the K-Means algorithm is used for final clustering, and the objective function is iteratively optimized. To determine the optimal cluster centers and sequence assignments.

[0034] Specifically, this method generates 100,000 sequences (temperature set to 1), removes 105 duplicate sequences that appeared in both the training and test sets, and calculates the evolutionary score of the remaining sequences. Preliminary clustering is performed using cd-hit, and the sequences with the lowest evolutionary scores (3305 in total) from the pure generated sequence clusters are selected for subsequent wet experimental validation. The similarity between each of the 3305 sequences is calculated, and based on the similarity, dimensionality reduction is performed using PCoA, followed by K-Means clustering to 30 clusters (the desired number of sequences). Figure 5 As shown.

[0035] S105: Select the initial representative sequence based on its closest distance to the centroid; if the initial representative sequence is L at position P2 and Q at position P1, then calculate the 15% lower quantile of the distances from all sequences within the cluster to the centroid. If there exists a distance to the centroid less than 1 / 3 ... For non-LQ sequences, the non-LQ sequence closest to the centroid is selected as the representative sequence of the cluster; where P1: the substrate amino acid residue located on the N-terminus of the cleavage bond and immediately adjacent to the cleavage bond; P2: the second substrate amino acid residue located on the N-terminus of the cleavage bond and immediately adjacent to the cleavage bond. LQ: a subsequence in which the amino acid sequence is arranged as leucine-glutamine.

[0036] Specifically, based on the 2D PCoA coordinates of the 30 clusters after K-Means clustering, the intra-cluster baseline and distance are first determined. For each cluster, the mean of the PCoA coordinates of all sequences within the cluster is taken as the centroid. The Euclidean distance from each sequence within the cluster to the centroid is calculated, resulting in a "sequence-distance" list for each cluster. For example, , It is the first in the cluster i PCoA coordinates (two-dimensional vectors) of the sequence, Let Euclidean distance be the sequence from the centroid. ,in Indicates the first c Clusters, These are the centroid coordinates of the cluster in the PCoA 2D space. For each "sequence-distance" pair, sort them in ascending order of distance, and select the first sequence after sorting (closest distance) as the initial representative sequence. If the initial representative sequence has LQ distances at both P2 and P1, proceed to the next correction step. Calculate the 15% lower quantile of the distances from all sequences within the cluster to the centroid. ;in , Let N be the set of distances arranged in ascending order, and N be the total number of sequences within the cluster. This is the floor function. If a non-LQ sequence is less than 15% lower quantile from its centroid... Then, the non-LQ sequence closest to the centroid is selected as the representative sequence of the cluster, resulting in 18 non-LQ sequences.

[0037] Wet test evaluation The efficiency of enzyme cleavage sites was quantified using biochemical experiments. This experiment involved adding corresponding fluorescent groups to the left and right of the peptide (designed cleavage site). When the two groups met after cleavage, fluorescence was emitted, and the total amount of cleaved peptide was quantified based on the fluorescence intensity. The cleavage rate was quantified by calculating the rate of change of fluorescence intensity over time. The quantitative biochemical experiment completed the synthesis and measurement of most peptides (23 / 30). Data were processed using the method described in Zhang, Jing, et al. "Protocol for high-throughput screening of SARS-CoV-2 main protease inhibitors using a robust fluorescence polarization assay." STAR protocols 3.4 (2022):101794. The fluorescence difference (ΔRFU) between the experimental and control groups was calculated at each time point. Then, starting from the first 10 time points, the time points were gradually increased to find the period that yielded the best linear fit (highest R² value) to determine the initial linear reaction phase (i.e., zero-order reaction kinetics where the substrate concentration is high and the reaction rate is constant). The slope Best RFU / t (e.g., ...) was calculated. Figure 7 Furthermore, if R² < 0.8 or p > 0.05, the fit was considered unsuccessful, and its cutoff efficiency was set to 0. The negative control, Pep12, Pep20, and Pep23 all showed fit failures; reviewing the original data revealed that none of them could be cut off. Each experiment was designed with three replicates.

[0038] Plot a bar chart of the cutting efficiency for each sequence ( Figure 8a and Figure 8b And calculate the relative cutting efficiency relative to the mean of the positive control.

[0039] Furthermore, the Best RFU / t of the filtered results was significantly correlated with the evolutionary score.

[0040] Filtering human source sequences We searched for fragments that perfectly matched our designed sequence in all human protein sequences. Using NCBI's BLAST tool, we searched the generated and screened 30 sequences in all human (9606) protein sequences. The results showed that only one sequence had a perfect match in the human protein database.

[0041] Figure 2 This is a schematic diagram of the design device for protease substrate short peptides based on the EVE model provided in this specification, including: The data collection module is used to collect sequence data of the natural substrates of the target protease; The evolution score acquisition module is used to train an EVE model using the sequence data, calculate the mutation score for each possible mutation using the trained model, and sum the mutation scores to obtain the evolution score of the complete sequence. The candidate sequence generation module is used to fit a normal distribution through the latent space of the training set and generate candidate sequences by mapping the randomness of standard normal sampling based on the temperature hyperparameter. A multi-level clustering module is used to calculate the evolution score of the candidate sequences, perform preliminary clustering using cd-hit, filter out the sequences with the lowest evolution score in the sequence cluster, calculate the similarity between pairs of sequences in the sequence with the lowest evolution score, and perform K-Means clustering based on the similarity after dimensionality reduction using PCoA. The representative sequence determination module is used to select the initial representative sequence based on the closest distance to the centroid. If the initial representative sequence is L at position P2 and Q at position P1, then the 15% lower quantile of the distances from all sequences in the cluster to the centroid is calculated. If there exists a distance to the centroid less than 1 / 3 ... For non-LQ sequences, the non-LQ sequence closest to the centroid is selected as the representative sequence of the cluster; where P1: the substrate amino acid residue located on the N-terminus of the cleavage bond and immediately adjacent to the cleavage bond; P2: the second residue of the substrate amino acid located on the N-terminus of the cleavage bond and immediately adjacent to the cleavage bond. LQ: a subsequence in which the amino acid sequence is arranged as leucine-glutamine.

[0042] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The provided method is a short peptide design method for protease substrates based on the EVE model.

[0043] This specification also provides a corresponding... Figure 1 This is an electronic device. At the hardware level, it includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for the business logic. The processor reads the corresponding computer program from the non-volatile memory into memory and then executes it to achieve the above-mentioned functions. Figure 1 The method for designing short peptides of protease substrates based on the EVE model is described above.

[0044] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the methodology). However, with technological advancements, many methodological improvements today can be considered direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved methodology into the hardware circuit. Therefore, it cannot be said that a methodological improvement cannot be implemented using hardware physical modules. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logic function is determined by the user programming the device. Designers can program and "integrate" a digital system onto a PLD themselves, without needing chip manufacturers to design and manufacture dedicated integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing integrated circuit chips, this programming is mostly implemented using "logic compiler" software. Similar to the software compiler used in program development, the original code before compilation must also be written in a specific programming language, called a Hardware Description Language (HDL). There are many HDLs, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, and RHDL (Ruby Hardware Description Language). Currently, the most commonly used are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also understand that by simply performing some logic programming on the method flow using one of these hardware description languages ​​and programming it into an integrated circuit, the hardware circuit implementing the logical method flow can be easily obtained.

[0045] The controller can be implemented in any suitable manner. For example, it can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicon Labs C8051F320. A memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also recognize that, in addition to implementing the controller in purely computer-readable program code form, the same functionality can be achieved by logically programming the method steps to make the controller take the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the means included therein for implementing various functions can also be considered as structures within the hardware component. Alternatively, the means for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0046] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0047] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0048] Those skilled in the art will understand that embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0049] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0050] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0051] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0052] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0053] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0054] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0055] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0056] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this specification may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0058] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0059] The above description is merely an embodiment of this specification and is not intended to limit this specification. Various modifications and variations can be made to this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for designing short peptides as protease substrates based on the EVE model, characterized in that, include: Collect sequence data of the natural substrates of the target protease; The EVE model is trained using the sequence data. The mutation score for each possible mutation is calculated using the trained model. The mutation scores are summed to obtain the evolution score of the complete sequence. The training set is mapped to the latent variable space, the latent variable distribution is fitted, the latent variables are sampled and decoded, and candidate sequences are generated by temperature scaling and bit-by-bit sampling. The evolutionary score of the candidate sequences is calculated, and preliminary clustering is performed using cd-hit. The sequence with the lowest evolutionary score in the sequence cluster is selected. The similarity between pairs of sequences in the sequence with the lowest evolutionary score is calculated. Based on the similarity, dimensionality reduction is performed using PCoA, and then clustering is performed using K-Means. The initial representative sequence is selected based on its closest distance to the centroid. If the initial representative sequence is L at position P2 and Q at position P1, then the 15% lower quantile of the distances from all sequences within the cluster to the centroid is calculated. If there exists a distance to the centroid less than 1 / 3 ... For non-LQ sequences, the non-LQ sequence closest to the centroid is selected as the representative sequence of the cluster; where P1: the substrate amino acid residue located on the N-terminus of the cleavage bond and adjacent to the cleavage bond; P2: the second residue of the substrate amino acid located on the N-terminus of the cleavage bond and adjacent to the cleavage bond; LQ: the subsequence in which the amino acid sequence is arranged as leucine-glutamine.

2. The method according to claim 1, characterized in that, The steps of mapping the training set to the latent variable space, fitting the latent variable distribution, sampling and decoding the latent variables, and performing temperature scaling and bit-by-bit sampling to generate candidate sequences include: Calculate the mean and standard deviation of the latent space of all training sets to verify whether the latent vectors of each dimension conform to the assumption of a standard normal distribution; Establish latent variables obtained from sampling the standard normal distribution; The latent variables are input into the decoder of the EVE model, and the decoder receives the latent variables and outputs the unnormalized logarithm fraction of each position for each symbol of the amino acid alphabet. The unnormalized logarithmic fraction is scaled by temperature and then converted into a probability distribution of legal amino acids; and The probability distribution is transformed into candidate sequences through multivariate sampling and character mapping.

3. The method according to claim 1, characterized in that, The selection of the initial representative sequence based on proximity to the centroid includes: Based on the 2D PCoA coordinates of the clusters after K-Means clustering, the mean of the PCoA coordinates of all sequences within the cluster is taken as the centroid. The Euclidean distance from each sequence within the cluster to the centroid is calculated, resulting in a "sequence-distance" list for each cluster; and For each cluster, sort the "sequence-distance" sequences in ascending order of distance, and select the first sequence after sorting as the initial representative sequence.

4. The method according to claim 1, characterized in that, The mutation score is obtained by calculating the logarithmic difference between the conditional probability of the mutated amino acid and the conditional probability of the wild-type amino acid.

5. The method according to claim 1, characterized in that, ,in N is the total number of sequences within the cluster. It is a rounding function. It is a set of distances arranged in ascending order.

6. The method according to claim 1, characterized in that, Clustering based on the similarity after dimensionality reduction using PCoA and then using K-Means includes: performing eigenvalue decomposition on the distance matrix, selecting the first two principal coordinates to construct a two-dimensional representation space; and using the K-Means algorithm to cluster in the two-dimensional representation space to determine the optimal cluster centers and sequence assignments.

7. The method according to claim 1, characterized in that, The method also includes the step of filtering human-derived sequences.

8. A device for designing short peptides as protease substrates based on the EVE model, characterized in that, include: The data collection module is used to collect sequence data of the natural substrates of the target protease; The evolution score acquisition module is used to train an EVE model using the sequence data, calculate the mutation score for each possible mutation using the trained model, and sum the mutation scores to obtain the evolution score of the complete sequence. The candidate sequence generation module is used to map the training set to the latent variable space, fit the latent variable distribution, sample and decode the latent variables, and perform temperature scaling and bit-by-bit sampling to generate candidate sequences. A multi-level clustering module is used to calculate the evolutionary score of the candidate sequences, perform preliminary clustering using cd-hit, filter out the sequences with the lowest evolutionary scores in the sequence clusters, calculate the similarity between pairs of sequences in the sequence with the lowest evolutionary scores, and perform K-Means clustering based on the similarity after dimensionality reduction using PCoA; and The representative sequence determination module is used to select the initial representative sequence based on the closest distance to the centroid. If the initial representative sequence is L at position P2 and Q at position P1, then the 15% lower quantile of the distances from all sequences in the cluster to the centroid is calculated. If there exists a distance to the centroid less than 1 / 3 ... For non-LQ sequences, the non-LQ sequence closest to the centroid is selected as the representative sequence of the cluster; where P1: the substrate amino acid residue located on the N-terminus of the cleavage bond and adjacent to the cleavage bond; P2: the second residue of the substrate amino acid located on the N-terminus of the cleavage bond and adjacent to the cleavage bond; LQ: the subsequence in which the amino acid sequence is arranged as leucine-glutamine.

9. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1 to 7.

Images