Novel single-stranded DNA deaminase mining method, system, equipment and medium

Abstract

The invention relates to the technical field of bioinformatics, and discloses a novel single-stranded DNA (deoxyribonucleic acid) deaminase mining method, system, equipment and medium, which break through the limitation that the traditional homologous search has low sensitivity to distant proteins and depends on random experimental verification by utilizing a deaminase family characteristic HMM (Hidden Markov Model) spectrum and combining a deaminase editing activity prediction model. The novel single-stranded DNA deaminase with editing activity can be efficiently mined from massive natural or artificial protein data. According to the method, the accuracy and coverage range of candidate protein screening are remarkably improved, a high-quality core element is provided for developing a high-specificity and low-off-target base editor, meanwhile, design optimization of a multifunctional editor is supported, and the method has important application value in the fields of gene therapy, molecular breeding and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of bioinformatics, and in particular to a novel single-stranded DNA deaminase mining method, system, device and medium. BACKGROUND

[0002] Single base editing is a new gene engineering technology that fuses the CRISPR-Cas system and deaminase acting on single-stranded DNA, enabling direct base substitution without introducing double-strand breaks, with high editing efficiency and strong specificity, and has great application potential in human genetic disease gene therapy, genetic disease model construction, livestock improvement and molecular breeding, etc. Deaminase as the core catalytic element directly determines the editing efficiency, targeting accuracy and safety of the base editor. Currently, researchers have developed adenine base editors (ABE) and cytosine base editors (CBE) by fusing adenine deaminases (such as TadA, etc.) and cytosine deaminases (such as APOBEC / AID, etc.), respectively, to catalyze A-to-G and C-to-T base substitution at the target DNA. Currently, the application of single base editors is mainly limited by side editing, by-products and off-target effects, and the mining of new deaminases is an important way to develop and optimize base editors.

[0003] Recent research has found that TadA family proteins have dual substrate catalytic potential, enabling adenine (A-to-G) and cytosine (C-to-T) deamination editing, providing a new path for the development of multifunctional editors. Thanks to the development and application of high-throughput metagenomics technology, the genomes of environmental microorganisms in a large number of diverse habitats have been sampled and sequenced, providing valuable and rich protein data resources for in-depth mining. At the same time, protein generation technology based on artificial intelligence technology has rapidly developed and applied, providing a new tool for generating specific types of deaminases de novo Design and generation of new tools are expected to generate unique new deaminases in a larger sequence space. How to quickly and accurately screen and mine new single-stranded DNA (cytosine / adenine) deaminases (TadA) with editing activity from a large number of natural or artificially generated protein databases still presents great challenges. Existing methods are generally based on homology search (such as BLASTP) and multiple sequence alignment analysis on small data sets, and then randomly select from the candidate set that meets the known knowledge for experimental verification, which mainly has the following problems: 1. Traditional methods rely on sequence homology alignment, which is less sensitive to distant homologs; 2. Unable to predict functional activity, random and blind when selecting experimental candidate proteins, leading to the fact that natural protein databases have not been deeply searched and mined, and the blindness of candidate protein selection hinders the mining and development of active TadA.

[0004] Therefore, there is an urgent need for a new single-stranded DNA deaminase mining method to quickly and efficiently screen new single-stranded DNA deaminases with editing activity from massive protein data. SUMMARY

[0005] The present application provides a new single-stranded DNA deaminase mining method, system, device and medium to solve the problem of lacking a method that can quickly and efficiently screen new single-stranded DNA deaminases with editing activity from massive protein data (≥5 billion).

[0006] The present application provides a new single-stranded DNA deaminase mining method, comprising: Obtaining protein sequence data to be mined; Obtaining a deaminase family characteristic HMM spectrum (protein hidden Markov model spectrum); According to the deaminase family characteristic HMM spectrum and the protein sequence data to be mined, performing a deep search for deaminase homologous protein sequence data in the protein sequence data to be mined; Preliminary screening of the deaminase homologous protein sequence data; Through a deaminase editing activity prediction model, the deaminase homologous protein sequence data after preliminary screening is subjected to editing activity prediction, and combined with a preset screening condition, the protein sequence data capable of being used as a candidate new single-stranded DNA deaminase is further screened from the deaminase homologous protein sequence data after preliminary screening.

[0007] According to the new single-stranded DNA deaminase mining method provided by the present application, the protein sequence data to be mined includes any one or any combination thereof: natural microbial protein amino sequence data obtained by microbial culture sequencing or metagenomic sequencing, and deaminase protein sequence data artificially designed and generated by using artificial intelligence methods.

[0008] According to the new single-stranded DNA deaminase mining method provided by the present application, the deaminase family characteristic HMM spectrum is obtained, comprising: Downloading the deaminase protein family HMM spectrum from a protein database, wherein the protein database includes any one or any combination thereof: InterPro database (accession number: IPR028883), Pfam database (accession number: PF14437), HAMAP database (accession number: MF_00972); Obtaining deaminase amino acid sequence data known to have deamination activity; According to the deaminase amino acid sequence data and the deaminase protein family HMM profile known to have deaminase activity, a multi-sequence alignment file is constructed and a deaminase family characteristic HMM profile is established.

[0009] According to the deaminase family characteristic HMM profile and the protein sequence data to be mined, the deaminase homologous protein sequence data is obtained by performing a deep search for deaminase homologous protein sequences, and the method comprises the following steps: According to the deaminase family characteristic HMM profile and the protein sequence data to be mined, the deaminase homologous protein sequence data is obtained by setting the E-value threshold to 1e-10 to control false positives and performing a deep search for deaminase homologous protein sequences.

[0010] According to the deaminase family characteristic HMM profile and the protein sequence data to be mined, the deaminase homologous protein sequence data is obtained by setting the E-value threshold to 1e-10 to control false positives and performing a deep search for deaminase homologous protein sequences. Filter out protein sequence data with abnormal translation expression (such as not starting with methionine (M), containing premature stop codon mutations, containing illegal amino acid residues (X), etc.); Filter out protein sequence data with a length not within a preset length range (such as too short (<120 amino acids) or too long (>300 amino acids)); Filter out protein sequence data that does not contain HAE or C**C motifs, wherein * represents any legal amino acid residue; Remove redundant protein sequence data with a sequence identity threshold of 100% and a shorter sequence coverage ratio of 100% as the threshold; Cluster with a sequence identity threshold of 50% and a shorter sequence coverage ratio of 90% as the threshold; Filter out protein sequence data with a sequence identity of ≥80% and a homologous region covering a query sequence or target sequence length of ≥80% with a known single-stranded DNA deaminase sequence.

[0011] According to the deaminase editing activity prediction model, the deaminase homologous protein sequence data after preliminary screening is subjected to editing activity prediction, and the protein sequence data capable of being used as a candidate new single-stranded DNA deaminase is further screened from the deaminase homologous protein sequence data after preliminary screening by combining a preset screening condition, and the method comprises the following steps: According to the deaminase homologous protein sequence data after preliminary screening, sequence feature embedding vectors of the deaminase homologous protein sequence data after preliminary screening are obtained through a protein sequence language model, and three-dimensional structures of the deaminase homologous proteins after preliminary screening are predicted by using a protein structure prediction tool, and structure feature embedding vectors of the deaminase homologous proteins after preliminary screening are obtained through a protein structure perception large model; according to the sequence feature embedding vectors and the structure feature embedding vectors of the deaminase homologous protein sequence data after preliminary screening, deaminase editing activity prediction results based on sequence features and deaminase editing activity prediction results based on structure features are obtained through a deaminase editing activity prediction model; According to the deaminase editing activity prediction results based on sequence features and the deaminase editing activity prediction results based on structure features, protein sequence data capable of serving as candidate novel single-stranded DNA deaminases is further screened from the deaminase homologous protein sequence data after preliminary screening in combination with a preset screening condition. The deaminase editing activity prediction model is trained by using a machine learning framework according to deaminase protein sequence sample data with known editing activity.

[0012] According to the deaminase editing activity prediction results based on sequence features and the deaminase editing activity prediction results based on structure features, protein sequence data capable of serving as candidate novel single-stranded DNA deaminases is further screened from the deaminase homologous protein sequence data after preliminary screening in combination with a preset screening condition. In the deaminase homologous protein sequence data after preliminary screening, deaminase homologous protein sequence data with both the deaminase editing activity prediction results based on sequence features and the deaminase editing activity prediction results based on structure features being editing activity is screened as protein sequence data capable of serving as candidate novel single-stranded DNA deaminases.

[0013] The application further provides a novel single-stranded DNA deaminase mining system, comprising: A first data acquisition module is configured to acquire protein sequence data to be mined. A second data acquisition module is configured to obtain a deaminase family characteristic HMM spectrum. A deep search module is configured to perform deaminase homologous protein sequence deep search according to the deaminase family characteristic HMM spectrum and the protein sequence data to be mined, and obtain deaminase homologous protein sequence data from the protein sequence data to be mined. A preliminary screening module is configured to preliminarily screen the deaminase homologous protein sequence data. The terminal screening module is configured to: predict editing activity of the deaminase homologous protein sequence data after the preliminary screening by a deaminase editing activity prediction model, and further screen the deaminase homologous protein sequence data after the preliminary screening to obtain protein sequence data capable of serving as a candidate novel single-chain DNA deaminase by combining with a preset screening condition.

[0014] The application further provides an electronic device comprising a processor and a memory storing a computer program, wherein the processor implements the novel single-chain DNA deaminase mining method according to any one of the above when executing the computer program.

[0015] The application further provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the novel single-chain DNA deaminase mining method according to any one of the above.

[0016] The application further provides a computer program product comprising a computer program, wherein the computer program can be stored on a non-transitory computer-readable storage medium, and the computer program is executed by a processor to enable a computer to implement the novel single-chain DNA deaminase mining method according to any one of the above.

[0017] The novel single-chain DNA deaminase mining method, system, device and medium provided by the application can at least bring the following beneficial effects: By constructing a deaminase family characteristic HMM spectrum, the recognition sensitivity to distant homologous proteins is significantly improved, the defects of traditional BLASTP and the like depending on local sequence similarity are overcome, potential functional deaminase sequences, including natural or artificially generated distant homologs, can be deeply mined from massive protein data; By integrating a deaminase editing activity prediction model, the screening standard is expanded from simple sequence similarity to functional activity prediction, the blindness of candidate protein selection is significantly reduced, invalid experimental verification is avoided, and the mining efficiency of active deaminases is improved; By combining the deep search of the HMM spectrum and the multi-level screening strategy of the activity prediction model, potential targets in metagenomic data or artificial intelligent generated protein libraries can be systematically covered, high-throughput and automated analysis of massive data is realized, and more high-quality candidate enzymes are provided for the development of novel single / dual functional base editors; By mining novel deaminases with unique editing characteristics (such as high specificity and low side editing / off-target effect), core elements are provided for optimizing existing ABE / CBE or developing multifunctional editors (such as editors targeting A and C at the same time), and the application progress in the fields of gene therapy and molecular breeding is promoted; The method can flexibly adapt to the mining needs of different deaminase families (such as TadA, APOBEC, etc.), and support function screening combined with the protein database generated by artificial intelligence, thereby providing key technical support for the future artificial design-verification closed loop of deaminase. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0019] Figure 1 A flowchart of a novel single-stranded DNA deaminase mining method provided by the present application.

[0020] Figure 2 The ROC curve and AUC value of the three deaminase editing prediction models established in the embodiment are shown.

[0021] Figure 3 A structural diagram of a novel single-stranded DNA deaminase mining system provided by the present application.

[0022] Figure 4 A structural diagram of an electronic device provided by the present application. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments, and should not be understood as limiting the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application. In the description of the present application, it should be understood that the terms used are only for the purpose of description, and should not be understood as indicating or implying relative importance.

[0024] Figure 1 A flowchart of a novel single-stranded DNA deaminase mining method provided by the present application. The execution subject of the novel single-stranded DNA deaminase mining method provided by the present application can be any applicable terminal side device or network side device, such as a novel single-stranded DNA deaminase mining device, etc.

[0025] Referring to Figure 1 The novel single-stranded DNA deaminase mining method provided by the present application can include: S110, obtaining protein sequence data to be mined, wherein the protein sequence data to be mined includes any one or any combination thereof of the following: natural microbial protein amino sequence data obtained by microbial culture sequencing or metagenomic sequencing, deaminase protein sequence data artificially designed and generated by using an artificial intelligence method, and the protein sequence data to be mined is denoted as proteinsDB.faa.

[0026] In addition, through literature mining analysis, the TadA protein sequences reported in the literature and verified by experiments for activity can be collected and sorted, including TadA with deamination / editing activity and TadA without deamination / editing activity, and the TadA proteins with experimentally verified activity in the laboratory or research team are combined and stored in a sequence file, denoted as known_TadAs.faa.

[0027] S120, obtaining a deaminase family characteristic HMM profile.

[0028] The protein hidden Markov model profile (HMM profile) is a sequence feature representation model based on probability statistics. Through dynamic modeling of conserved regions in multiple sequence alignment (MSA), the evolutionary constraint strength of amino acid sites is quantified, and through a differential weight mechanism and probability-driven search, it can overcome the linear sequence similarity limitation and more accurately identify distant homologous proteins that have undergone significant evolutionary changes but still retain functional or structural similarities, which provides a key tool for mining new TadA enzymes from massive protein databases.

[0029] In an embodiment, S120 can include: Downloading the deaminase protein family HMM profile from the protein database, wherein the protein database includes any one or any combination thereof of the following: InterPro database (accession number: IPR028883), Pfam database (accession number: PF14437), and HAMAP database (accession number: MF_00972); Collecting the latest literature reported TadA enzyme amino acid sequences with deamination activity through literature mining, using CD-HIT tools to deduplicate (denoted as known_active_TadA_deduplicated.faa) with a sequence similarity threshold of 90%, using hmmalign and HMM profile IPR028883 to construct a multiple sequence alignment file of known active TadA enzyme amino acid sequences, and then using hmmbuild to construct a TadA enzyme characteristic profile; or, Collect the latest literature reports of TadA enzyme amino acid sequences with deaminase activity known from literature mining, and use CD-HIT tools to deduplicate with a sequence consistency threshold of 90% (denoted as known_active_TadA_deduplicated.faa). Use MAFFT software (E-INS-i iterative optimization algorithm) to establish a multiple sequence alignment file, and use hmmbuild to construct a TadA enzyme HMM spectrum. This embodiment uses the TadA enzyme HMM spectrum file activeTadAv1.hmm established in this step as an example for description.

[0030] S130, according to the deaminase family characteristic HMM spectrum and the protein sequence data to be mined, setting the E-value threshold to 1e-10 to control false positives, performing a deep search for deaminase homologous protein sequences, and obtaining deaminase homologous protein sequence data in the protein sequence data to be mined.

[0031] Specifically, in this embodiment, the TadA enzyme characteristic HMM spectrum file activeTadAv1.hmm and the massive protein amino acid sequence database file proteinsDB.faa are used as inputs, the hmmsearch command in the HMMER tool package is used to perform a deep search for TadA enzyme homologous protein sequences, the E-value threshold is set to 1e-10 to control false positives, improve specificity and remote homologous protein discovery capability, and all hit TadA enzyme homologous proteins are obtained according to the target protein access number in the hit record and the proteinsDB.faa data, denoted as potential_homologous_TadAs_hits1.faa.

[0032] S140, preliminary screening of deaminase homologous protein sequence data.

[0033] In one embodiment, S140 includes: Filtering out protein sequence data with abnormal translation expression (such as not starting with methionine (M), containing premature stop codon mutations, containing illegal amino acid residues (X), etc.); Filtering out protein sequence data with a length not within a preset length range (such as too short (<120 amino acids) or too long (>300 amino acids)); Filtering out protein sequence data that does not contain HAE or C**C motifs, where * represents any legal amino acid residue; Using sequence clustering tools (such as CD-HIT, MMseqs2, etc.), setting a sequence consistency threshold of 100% and a shorter sequence coverage ratio of 100% as the threshold for deduplication, and retaining non-redundant protein sequence data; Using a sequence clustering tool (such as CD-HIT, MMseqs2, etc.), clustering is performed with a sequence consistency threshold of 50% and a shorter sequence coverage ratio of 90% as the threshold, and is recorded as potential_homologous_TadAs_hits2.faa; Using a protein sequence alignment tool (such as BLASTP, Diamond, etc.), the sequence consistency of the mined TadA homologous proteins and the experimentally verified active known TadA proteins is obtained by inputting potential_homologous_TadAs_hits2.faa as the query sequence and known_TadAs.faa as the searched database, and removing TadA homologous proteins with sequence consistency ≥80% and homologous region coverage query sequence or target sequence length ≥80% to obtain effective homologous proteins for subsequent analysis, recorded as potential_homologous_TadAs_hits3.faa.

[0034] S150, by the deaminase editing activity prediction model, the deaminase homologous protein sequence data after preliminary screening is edited to predict the editing activity, and combined with the preset screening condition, the protein sequence data capable of being used as a candidate new single-stranded DNA deaminase is further screened from the deaminase homologous protein sequence data after preliminary screening.

[0035] In an embodiment, S150 can include: According to the deaminase homologous protein sequence data after preliminary screening, the sequence feature embedding vector of the deaminase homologous protein sequence data after preliminary screening is obtained by a protein sequence language model, and the three-dimensional structure of the deaminase homologous protein after preliminary screening is predicted by a protein structure prediction tool, and the structure feature embedding vector of the deaminase homologous protein after preliminary screening is obtained by a protein structure perception large model; According to the sequence feature embedding vector and the structure feature embedding vector of the deaminase homologous protein sequence data after preliminary screening, the deaminase editing activity prediction model is obtained based on the sequence feature and the deaminase editing activity prediction result based on the structure feature, wherein the deaminase editing activity prediction model is trained by a machine learning framework according to known editing activity deaminase protein sequence sample data; In the deaminase homologous protein sequence data after preliminary screening, the deaminase homologous protein sequence data with editing activity is screened based on the sequence feature and the deaminase editing activity prediction result based on the structure feature, and is used as the protein sequence data of the candidate new single-stranded DNA deaminase.

[0036] Specifically, the present embodiment takes the TadA protein with known experimental activity (known_TadAs.faa) as the positive label with editing activity (active) and the negative label without editing activity (inactive). For each TadA protein sequence, a fixed-dimensional sequence feature embedding vector is extracted using the protein sequence language model ESM1b and ESM2, respectively; the three-dimensional spatial structure of the TadA homologous protein sequence is predicted using an artificial intelligence protein structure prediction tool (such as AlphaFold2 / 3, ESMFold, OpenFold, Protenix, Boltz-1 / 2, etc.), and a fixed-dimensional structure feature embedding vector of the TadA homologous protein is extracted using a protein structure perception large model (such as SaProt, ProMEP, etc.) with the predicted three-dimensional structure file as input. An automated machine learning framework AutoGluon-TabularPredictor is used for model training and optimization, a 5x5 bagged cross-validation is set to ensure the robustness of the model, and a Bayesian hyperparameter optimization is used to maximize the F1 score, and the best model is obtained as the TadA editing activity prediction model, denoted as TadA_activity.model.

[0037] Referring to the TadA editing activity prediction model construction process, the sequence feature embedding vector and the structure feature embedding vector of each protein in the set of all effective TadA homologous proteins (potential_homologous_TadAs_hits3.faa) are obtained using the protein language model.

[0038] For each effective TadA homologous protein, the sequence feature embedding vector and the structure feature embedding vector thereof are used as input to predict its editing activity using the trained editing activity prediction model. When both the sequence feature-based model and the structure feature-based model predict that the current protein has editing activity (active), the current protein can be used as a new candidate TadA protein for subsequent synthesis and wet experiment to verify its deamination activity and editing characteristics.

[0039] Two application cases are provided below to further verify the new single-stranded DNA deaminase mining method provided by the present application.

[0040] Application Case 1: Mining TadA enzymes in the E. coli K-12 sub-strain MG1655 protein set The core deaminase ecTadA* used in the ABE editor was first obtained from Escherichia coli str. K-12 substr. MG1655 and evolved to catalyze the deamination of adenine (A) to inosine (I), which is recognized by DNA polymerase as guanine (G), thereby realizing base conversion of target site A-to-G or complementary strand T-to-C. In this case, the method was applied to the protein dataset (GCF_000005845_protein.faa, 4242 proteins) encoded by the genome of the strain to perform screening and evaluation of its effectiveness.

[0041] Step 1. Use the TadA enzyme HMM profile file activeTadAv1.hmm and the protein library file GCF_000005845_protein.faa to be searched as input, use hmmsearch and a predetermined Evalue threshold of 1e-10 to search and mine, and get 2 hit sequences (NP_417054.2, NP_414948.1); Step 2. Remove proteins that cannot be normally translated and expressed, such as those starting with methionine (M), containing premature stop codon mutations, containing illegal amino acid residues (X), etc., leaving 2 hit records; Step 3. Remove homologous proteins that are too short (<120 amino acids) or too long (>300 amino acids), and homologous protein NP_414948.1 is excluded due to its length of 367 amino acids exceeding the predetermined threshold, leaving one homologous protein (NP_417054.2); Step 4. Remove proteins that do not contain HAE or C**C motifs, leaving one homologous protein (NP_417054.2); since there is only one effective homologous protein, the 100% sequence identity de-redundancy and 50% sequence identity clustering results are still itself; Step 5. Use the protein sequence alignment tool BLASTP with the NP_417054.2 sequence file as the query sequence and the known active TadA protein (known_TadAs.faa) as the target sequence for homologous search and alignment. The results show that the NP_417054.2 protein has the highest sequence identity with the ecTadA sequence reported in the literature, with a sequence identity of 100%, a homologous alignment Evalue of 5e-128, and a coverage ratio of 100% for the length of the query sequence and the target sequence in the homologous alignment region. At the same time, the annotation information of NP_417054.2 protein in RefSeq database is "RNA adenosine(34) deaminase", which is consistent with the expectation.

[0042] The above results demonstrate that the new single-stranded DNA deaminase mining method provided by the present application can accurately find and screen effective TadA enzyme proteins from the dataset containing target TadA proteins.

[0043] Application Case 2: TadA enzyme mining and identification in a natural microbial protein database of the order of one billion Step 1, natural microbial protein database collection, search for microbial genome and metagenome sequencing and assembly related literatures from academic websites such as PubMed with “microbe genome”, “metagenome”, “metagnome-assembled genomes”, “metagnomics sequencing”, “Shotgun metagenome data” and “genome-resolved metagenomics” as keywords, download the protein amino acid sequences encoded by open reading frames according to the access addresses provided in the literatures, and for the datasets only providing genome sequences, download the datasets and then use the prodigal tool to predict the protein amino acid sequences encoded by the open reading frame regions. In this case, 64 datasets were collected and downloaded from 248 literatures, containing a total of 5904798022 proteins (about 590.05 million), denoted as naturalProteinDB; Step 2, use the TadA enzyme HMM profile file activeTadAv1.hmm and the naturalProteinDB as the inputs of the hmmsearch tool to deeply search and mine TadA homologous proteins in the established natural microbial protein database, and obtain a total of 1323764 hit sequences; Step 3, remove 194422 proteins that cannot be normally translated and expressed due to starting with methionine (M), containing premature stop codon mutations, or containing illegal amino acid residues (X), leaving 1129342 TadA homologous proteins; Step 4, remove 192024 homologous proteins with too short length (<120 amino acids) and 247803 homologous proteins with too long length (>300 amino acids), leaving 689515 TadA homologous proteins; Step 5, remove 159452 proteins not containing HAE or C**C motifs, leaving 530063 TadA homologous proteins; Step 6, use the sequence clustering tool CD-HIT to remove redundancy with sequence identity threshold 100% and shorter sequence coverage ratio threshold 100%, to obtain 279587 non-redundant TadA homologous proteins; Step 7, use the sequence clustering tool CD-HIT) to cluster the 279587 non-redundant TadA homologous proteins with sequence identity threshold 50% and shorter sequence coverage ratio 90% as threshold, and obtain 4409 sequence cluster representative sequences; Step 8, use the protein sequence alignment tool Diamond to input the 4409 sequence cluster representative sequences as query sequences, and input known_TadAs.faa as the searched database, analyze the sequence identity of the 19 mined TadA homologous proteins and the experimentally verified active known TadA proteins ≥80% and the homologous region coverage query sequence or target sequence length ≥80%, remove the 19 highly similar TadA homologous proteins, and obtain 4390 effective homologous proteins for subsequent analysis; Step 9, collect known experimentally active TadAs, first search related literatures on academic websites such as PubMed with the keywords (“TadA homologs” OR “TadA orthologs”) AND “base editor” AND (“screening” OR “mining”), and collect 73 ecTadA artificially evolved or modified variants and TadA homologous proteins by March 20, 2024, of which 54 proteins have editing activity and 19 proteins have no editing activity. Secondly, the inventors' research team randomly selected 77 from the 4390 effective homologous proteins for wet experiment synthesis and function verification, and found that 32 had editing activity and 45 had no editing activity. Combine these known active TadA homologous proteins, use the CD-HIT tool to cluster with sequence identity threshold 80%, and obtain 127 non-redundant known active TadA proteins, of which 63 have editing activity and 64 have no editing activity; Step 10, training the artificial intelligence-based TadA enzyme editing activity prediction model. For 127 TadA proteins with known editing activity, sequence feature embedding vectors are extracted using ESM1b and ESM2 models, model training and optimization are performed using the automated machine learning framework AutoGluon-TabularPredictor, the model with the highest F1 score is selected as the best model, and is recorded as activity_1d_esm1b.model and activity_1d_esm2.model respectively; the three-dimensional structure corresponding to each TadA protein sequence is predicted using ESMFold, and the structure feature embedding vector is extracted using the multi-modal protein deep representation learning model ProMEP, and the model training and optimization are performed using the automated machine learning framework AutoGluon-TabularPredictor, and the model with the highest F1 score is selected as the best model, and is recorded as activity_3d_promep.model. The prediction performance of the three models is evaluated using Leave-One-Out Cross-Validation (LOO-CV), and the Area Under the Receiver Operating characteristic Curve (ROC AUC) is used for summary display, as shown in Figure 2 FIG. 15, the AUC values of the three models are all ≥0.8, and the highest value is the prediction method based on structure features, reaching 0.88, which confirms the reliability of the prediction model; Step 11, referring to the TadA editing activity prediction model construction process, use the protein language model to obtain all 4390 effective TadA homologous proteins respectively, use ESM1b, ESM2 and ProMEP models to extract the sequence feature embedding vector (esm1b_embedding.vector, esm2_embedding.vector) and the structure feature embedding vector (promep_embedding.vector) of each protein, input esm1b_embedding.vector into activity_1d_esm1b.model model to obtain the prediction label 1, input esm2_embedding.vector into activity_1d_esm2.model model to obtain the prediction label 2, input promep_embedding.vector into activity_3d_promep.model model to obtain the prediction label 3, and the prediction labels 1, 2 and 3 are simultaneously used as the TadA enzyme candidates with editing activity (active) value and not belonging to the TadA proteins that have been experimentally verified, and 375 novel TadA enzyme candidates are obtained.

[0044] The novel single-stranded DNA deaminase mining method provided by the application is suitable for a protein database with a scale of 10 billion, and is actually applied to a natural microbial protein database with a scale of 590.05 million. The application mines 1323764 TadA homologous proteins from the protein database with a scale of 590.05 million. Further, 4390 novel TadA homologous proteins are obtained through sequence screening and sequence clustering. Based on an artificial intelligence model, 375 novel TadA enzyme candidates are obtained through activity prediction and screening. Compared with the methods reported in the literature, the application increases the exploration range of natural protein database by 389318.8 times (5904798022 / 15167), and expands the scale of TadA homologous proteins by 87.3 times (1323764 / 15167). The application greatly increases the exploration range of TadA enzyme diversity in the natural protein database, especially the search and mining of distant TadA homologous proteins. At the same time, the application provides a precise and efficient candidate range for experimental verification, and reduces the selection range by 3530 times (1323764 / 375), which proves the effectiveness of the application.

[0045] The novel single-stranded DNA deaminase mining method provided by the application can at least bring the following beneficial effects: Break through the limitations of traditional homology search: by constructing a deaminase family characteristic HMM spectrum, the recognition sensitivity of distant homologous proteins is significantly improved, the defects of traditional BLASTP method relying on local sequence similarity are overcome, potential functional deaminase sequences can be deeply mined from massive protein data, including natural or artificially generated distant homologues; Realize accurate prediction of functional activity: by integrating the deaminase editing activity prediction model, the screening standard is expanded from simple sequence similarity to functional activity prediction, which significantly reduces the blindness of candidate protein selection, avoids invalid experimental verification, and improves the mining efficiency of active deaminases; Improve the mining efficiency and coverage: combined with the deep search of HMM spectrum and the multi-level screening strategy of activity prediction model, the potential targets in the macrogenomic data or the artificial intelligence generated protein library can be systematically covered, high-throughput and automated analysis of massive data can be realized, and more high-quality candidate enzymes for developing novel single / dual functional base editors can be provided; Accelerate the optimization and application of base editors: by mining novel deaminases with unique editing characteristics (such as high specificity, low side editing / off-target effect), core elements for optimizing existing ABE / CBE or developing multifunctional editors (such as editors targeting A and C simultaneously) are provided, which promotes the application progress in the fields of gene therapy, molecular breeding, etc.; Technical compatibility and expansibility: the method can be flexibly adapted to the mining needs of different deaminase families (such as TadA, APOBEC, etc.), and can support functional screening combined with artificial intelligence generated protein database, which provides key technical support for the future artificial design-verification closed loop of deaminases.

[0046] The novel single-stranded DNA deaminase mining system provided by the application is described below, and the novel single-stranded DNA deaminase mining system described below can be correspondingly referred to the novel single-stranded DNA deaminase mining method described above.

[0047] Referring to Figure 3 The novel single-stranded DNA deaminase mining system provided by the application can include: A first data acquisition module for acquiring protein sequence data to be mined; A second data acquisition module for obtaining a deaminase family characteristic HMM spectrum; A deep search module for performing deaminase homologous protein sequence deep search according to the deaminase family characteristic HMM spectrum and the protein sequence data to be mined, and obtaining deaminase homologous protein sequence data in the protein sequence data to be mined; The preliminary screening module is configured to preliminarily screen the deaminase homologous protein sequence data; The final screening module is configured to predict the editing activity of the preliminarily screened deaminase homologous protein sequence data by using a deaminase editing activity prediction model, and further screen the preliminarily screened deaminase homologous protein sequence data to obtain protein sequence data capable of serving as a candidate novel single-stranded DNA deaminase by combining with a preset screening condition.

[0048] Figure 4 An example of a schematic diagram of the physical structure of an electronic device is shown in Figure 4 As shown, the electronic device can include a processor 810, a communications interface 820, a memory 830, and a communications bus 840, wherein the processor 810, the communications interface 820, and the memory 830 complete mutual communication through the communications bus 840. The processor 810 can invoke a logical instruction in the memory 830 to perform the following steps: Obtain protein sequence data to be mined; Obtain a deaminase family characteristic HMM spectrum; According to the deaminase family characteristic HMM spectrum and the protein sequence data to be mined, perform a deep search for deaminase homologous protein sequence data in the protein sequence data to be mined, and obtain deaminase homologous protein sequence data; Preliminarily screen the deaminase homologous protein sequence data; Preliminarily screen the deaminase homologous protein sequence data;

[0049] In addition, the logical instruction in the memory 830 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the method described in various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0050] In another aspect, the present application also provides a computer program product comprising a computer program, the computer program being stored in a non-transitory computer-readable storage medium, and the computer program being executable by a processor to cause a computer to perform the following steps: obtaining protein sequence data to be mined; obtaining a HMM profile characteristic of a deaminase family; performing a deep search for deaminase homologous protein sequences according to the HMM profile characteristic of the deaminase family and the protein sequence data to be mined, to obtain deaminase homologous protein sequence data from the protein sequence data to be mined; performing a preliminary screening on the deaminase homologous protein sequence data; performing editing activity prediction on the deaminase homologous protein sequence data after the preliminary screening by a deaminase editing activity prediction model, and further screening, in the deaminase homologous protein sequence data after the preliminary screening, to obtain protein sequence data capable of being used as a candidate for a novel single-stranded DNA deaminase, in combination with a preset screening condition.

[0051] In yet another aspect, the present application also provides a non-transitory computer-readable storage medium having a computer program stored thereon, the computer program being executable by a processor to cause a computer to perform the following steps: obtaining protein sequence data to be mined; obtaining a HMM profile characteristic of a deaminase family; performing a deep search for deaminase homologous protein sequences according to the HMM profile characteristic of the deaminase family and the protein sequence data to be mined, to obtain deaminase homologous protein sequence data from the protein sequence data to be mined; performing a preliminary screening on the deaminase homologous protein sequence data; performing editing activity prediction on the deaminase homologous protein sequence data after the preliminary screening by a deaminase editing activity prediction model, and further screening, in the deaminase homologous protein sequence data after the preliminary screening, to obtain protein sequence data capable of being used as a candidate for a novel single-stranded DNA deaminase, in combination with a preset screening condition.

[0052] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., they can be located in one place, or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the present embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0053] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0054] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A novel method for discovering single-stranded DNA deaminases, characterized in that, include: Obtain the protein sequence data to be mined; The characteristic HMM profiles of the deaminase family were obtained; Based on the characteristic HMM profiles of the deaminase family and the protein sequence data to be mined, a deep search for deaminase homologous protein sequences is performed to obtain deaminase homologous protein sequence data from the protein sequence data to be mined. Preliminary screening of deaminase homologous protein sequence data; By using a deaminase editing activity prediction model, the editing activity of deaminase homologous protein sequence data after preliminary screening is predicted. Combined with preset screening conditions, the protein sequence data that can be used as candidate novel single-stranded DNA deaminases are further screened from the preliminary screening deaminase homologous protein sequence data.

2. The novel single-stranded DNA deaminase discovery method according to claim 1, characterized in that, The protein sequence data to be mined includes any one or any combination of the following: amino acid sequence data of natural microbial proteins obtained by microbial culture sequencing or metagenomic sequencing, and deaminase protein sequence data artificially designed and generated using artificial intelligence methods.

3. The novel single-stranded DNA deaminase discovery method according to claim 1, characterized in that, The obtained characteristic HMM profile of the deaminase family includes: Download the HMM profiles of deaminase protein families from a protein database; Obtain amino acid sequence data of deaminases known to have deamination activity; Based on the known amino acid sequence data of deaminases with deamination activity and the HMM profiles of the deaminase protein family, a multiple sequence alignment file was constructed and a characteristic HMM profile of the deaminase family was established.

4. The novel single-stranded DNA deaminase discovery method according to claim 1, characterized in that, The process involves performing a deep search for deaminase homologous protein sequences based on the characteristic HMM profiles of the deaminase family and the protein sequence data to be mined. This search yields deaminase homologous protein sequence data from the protein sequence data to be mined, including: Based on the characteristic HMM profiles of the deaminase family and the protein sequence data to be mined, an E-value threshold of 1e-10 was set to control false positives. A deep search for deaminase homologous protein sequences was performed, and deaminase homologous protein sequence data were obtained from the protein sequence data to be mined.

5. The novel single-stranded DNA deaminase discovery method according to claim 1, characterized in that, The preliminary screening of deaminase homologous protein sequence data includes any one or any combination of the following: Filter out protein sequence data with anomalous translation expression; Filter out protein sequence data whose length is outside the preset length range; Filter out protein sequence data that do not contain HAE or C**C motifs, where * represents any legal amino acid residue; Redundancy removal was performed using a sequence consistency threshold of 100% and a shorter sequence coverage ratio of 100% as thresholds, retaining non-redundant protein sequence data. Clustering was performed using a sequence consistency threshold of 50% and a shorter sequence coverage rate of 90% as thresholds; Protein sequence data that are ≥80% identical to publicly available single-stranded DNA deaminase sequences and whose homologous regions cover ≥80% of the query or target sequence length are filtered out.

6. The novel single-stranded DNA deaminase discovery method according to any one of claims 1-5, characterized in that, The method involves using a deaminase editing activity prediction model to predict the editing activity of pre-screened deaminase homologous protein sequence data, and further screening the pre-screened deaminase homologous protein sequence data to obtain protein sequence data that can serve as candidate novel single-stranded DNA deaminases, based on preset screening conditions. Based on the preliminary screening of deaminase homologous protein sequence data, the sequence feature embedding vector of the preliminary screening of deaminase homologous protein sequence data is obtained through a protein sequence language model. The three-dimensional structure of the preliminary screening of deaminase homologous proteins is predicted using a protein structure prediction tool. The structural feature embedding vector of the preliminary screening of deaminase homologous proteins is obtained through a large protein structure perception model. Based on the sequence feature embedding vectors and structural feature embedding vectors of the deaminase homologous protein sequence data after preliminary screening, the deaminase editing activity prediction model is used to obtain the deaminase editing activity prediction results based on sequence features and the deaminase editing activity prediction results based on structural features. Based on the prediction results of deaminase editing activity based on sequence features and the prediction results of deaminase editing activity based on structure features, and combined with preset screening conditions, protein sequence data that can be used as candidate novel single-stranded DNA deaminases are further screened from the deaminase homologous protein sequence data after the initial screening. Among them, the deaminase editing activity prediction model is trained using a machine learning framework based on sample data of deaminase protein sequences with known editing activities.

7. The novel single-stranded DNA deaminase discovery method according to claim 6, characterized in that, The process involves further screening the pre-screened deaminase homologous protein sequence data based on sequence feature-based and structure feature-based deaminase editing activity prediction results, combined with preset screening conditions, to obtain protein sequence data that can serve as candidate novel single-stranded DNA deaminases. This includes: In the initial screening of deaminase homologous protein sequence data, both the sequence-based and structural-based prediction results of deaminase editing activity were found to be deaminase homologous protein sequence data with editing activity, which were then used as candidate novel single-stranded DNA deaminase protein sequence data.

8. A novel single-stranded DNA deaminase discovery system, characterized in that, include: The first data acquisition module is used to: acquire protein sequence data to be mined; The second data acquisition module is used to: obtain the characteristic HMM spectrum of the deaminase family; The deep search module is used to: perform a deep search for deaminase homologous protein sequences based on the characteristic HMM profiles of the deaminase family and the protein sequence data to be mined, and obtain the deaminase homologous protein sequence data in the protein sequence data to be mined. The initial screening module is used for: preliminary screening of deaminase homologous protein sequence data; The final screening module is used to: predict the editing activity of deaminase homologous protein sequence data after preliminary screening using a deaminase editing activity prediction model, and further screen the deaminase homologous protein sequence data after preliminary screening to obtain protein sequence data that can be used as candidate novel single-stranded DNA deaminases in combination with preset screening conditions.

9. 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 novel single-stranded DNA deaminase discovery method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the novel single-stranded DNA deaminase discovery method as described in any one of claims 1 to 7.

Images