Biomimetic antibacterial peptide, pharmaceutical composition, application of biomimetic antibacterial peptide in preparation of antibacterial drugs and design method of biomimetic antibacterial peptide

Abstract

The invention provides a bionic antibacterial peptide, a pharmaceutical composition, application of the bionic antibacterial peptide in preparation of antibacterial drugs and a design method of the bionic antibacterial peptide. A plurality of compounds are used for carrying out N-terminal modification on a polypeptide sequence consisting of 20 natural amino acids, so that a novel antibacterial peptide with a bionic function is generated, and the designed bionic antibacterial peptide is screened by adopting a method based on deep learning and experimental interaction. Compared with traditional drug screening, consumed time, manpower, material resources and financial resources are greatly reduced, the designed novel bionic antibacterial peptide has good biocompatibility, a functional group modified at the N end of the novel bionic antibacterial peptide can activate the self-assembly activity of polypeptide and help the polypeptide target a bacterial membrane, bacterial drug resistance is not likely to be caused, and the novel bionic antibacterial peptide has good application prospects. The drug-resistant bacteria can be effectively killed, the sterilization speed is higher, and potential and wide application prospects are achieved in clinical treatment of bacteria and drug-resistant bacteria infection diseases.

Description

Technical Field

[0001] The present application belongs to the field of biomedicine, and specifically relates to a biomimetic antimicrobial peptide, a pharmaceutical composition, and an application thereof in the preparation of an antimicrobial drug, as well as a design method for a biomimetic antimicrobial peptide. Background Art

[0002] Bacterial infection is one of the diseases that seriously threaten human life and health. The Lancet reported that in 2019, about 7.7 million people died from common bacterial infections, accounting for 13.6% of the total number of deaths worldwide, becoming the second leading cause of death in the world. Bacterial infections are mainly treated by the use of antibiotics, but the abuse and misuse of antibiotics have led to the gradual development of bacterial resistance and even multi-drug resistance, which has brought severe challenges to clinical anti-infection treatment. In 2019, about 4.95 million deaths worldwide were related to bacterial resistance, of which 1.27 million were directly attributed to the emergence of bacterial resistance. In the same year, the World Health Organization listed bacterial resistance as one of the top ten global health threats. Despite the increasingly serious problem of bacterial resistance, almost no new antibiotic categories have been approved in recent years. Faced with this situation, countries around the world have regarded the development of new antibacterial drugs as an important task in the field of biomedicine, and are trying their best to promote scientific and technological research on the prevention and control of bacterial resistance. The development of new antibacterial drugs is imminent.

[0003] In nature, the self-assembly phenomenon of proteins and peptides is widespread, and this phenomenon can enhance their activity and function. Studies have found that a depsipeptide (Clovibactin) can be isolated from soil bacteria. It self-assembles into advanced fibers and binds to the bacterial cell wall precursor pyrophosphate (PPi), thereby blocking the biosynthesis of the cell wall. This simple and unaffected killing mechanism makes Clovibactin less likely to develop drug resistance and has lower cytotoxicity. In the human body, Paneth cells in the small intestine can secrete a defensin that self-assembles into a nano-network structure in the intestine to resist bacterial invasion. Compared with traditional antimicrobial peptides such as Cathelicidin LL-37, Magainin 2, and Melittin, this type of antimicrobial peptide with self-assembly function has better biocompatibility, longer in vivo half-life, and is not easy to acquire drug resistance. It is a new type of antibacterial drug to replace antibiotics.

[0004] However, before a new antibiotic can be marketed, it must first go through the design and screening of lead compounds, the screening of lead compounds, and the optimization of lead compounds. This process can usually screen 1,000 to 10,000 drug molecules, but it takes 1-7 years and 5-10 million euros. In other words, traditional screening methods can only screen 1,000-10,000 lead compound molecules and optimize the lead compounds in a few years. The entire process consumes a lot of time, manpower, material resources, and financial resources. In addition, the antibiotics screened by traditional methods usually target the bacterial cell or affect a certain biological process, resulting in traditional antibiotics usually being active only against one or a class of bacteria, and traditional antibiotics have a slow bactericidal rate and weak biofilm removal ability. Bacteria gradually develop resistance to antibiotics, and the therapeutic effect of antibiotics on resistant bacteria is gradually declining.

[0005] It can be seen that the prior art has not yet found a design method that can efficiently and cost-effectively design and screen new bionic antimicrobial peptides in a short period of time, and no new bionic antimicrobial peptides with good biocompatibility, broad-spectrum antimicrobial activity, low resistance to drug resistance, faster sterilization speed, and stronger biofilm removal ability have been found. Summary of the invention

[0006] One object of the present application is to provide a method for designing biomimetic antimicrobial peptides, the method comprising: constructing a first data set comprising a plurality of polypeptide sequences without N-terminal modification, wherein the polypeptide sequences in the first data set consist of only 20 natural amino acids and are annotated with antimicrobial activity; performing N-terminal modification on polypeptide sequences selected from the first data set using a plurality of compounds to generate a plurality of self-assembling polypeptide sequences with N-terminal modification; performing a polypeptide drug sensitivity test on each of the generated self-assembling polypeptide sequences with N-terminal modification and annotating antimicrobial activity according to the test results, and taking each of the self-assembling polypeptide sequences with N-terminal modification annotated with antimicrobial activity as a second data set; pre-training a deep learning model using the first data set so that the pre-trained deep learning model can reconstruct the hidden feature representation of the polypeptide sequence and output the antimicrobial activity prediction result of the polypeptide sequence; and using the second data set to perform a peptide drug sensitivity test on each of the generated self-assembling polypeptide sequences with N-terminal modification and annotating antimicrobial activity according to the test results. The pre-trained deep learning model is incrementally transferred and trained, so that the deep learning model trained by incremental transfer can output the prediction results of the antibacterial activity of the self-assembling polypeptide sequence with N-terminal modification; the polypeptide sequences in the first data set are N-terminally modified using a variety of compounds to generate a third data set containing multiple self-assembling polypeptide sequences with N-terminal modification, and the antibacterial activity of each self-assembling polypeptide sequence with N-terminal modification in the third data set is predicted using the deep learning model trained by incremental transfer, and the self-assembling polypeptide sequence with N-terminal modification whose predicted antibacterial activity results are higher than a preset threshold is selected for synthesis and purification, and a polypeptide drug sensitivity test is carried out; the self-assembling polypeptide sequence with N-terminal modification that passes the polypeptide drug sensitivity test is sequentially subjected to in vitro experimental evaluation and therapeutic effect evaluation, and the self-assembling polypeptide sequence with N-terminal modification whose antibacterial activity is verified by the evaluation is used as the designed novel bionic antimicrobial peptide.

[0007] In some embodiments, the deep learning model includes a deep learning model of a feature representation unit, a pre-trained network and an incremental migration unit, wherein the incremental migration unit includes a noise enhancement module, an N-terminal modification encoding module and a transfer learning network; the deep learning model is pre-trained using the first data set so that the pre-trained deep learning model can reconstruct the hidden feature representation of the polypeptide sequence and output the antibacterial activity prediction result of the polypeptide sequence. Specifically, it includes: the feature representation unit performs input feature representation on each polypeptide sequence in the first data set, and the pre-trained network is trained and optimized based on the input feature representation and antibacterial activity annotation of the polypeptide sequence, and the pre-trained network outputs the hidden feature representation of the reconstructed polypeptide sequence and the antibacterial activity prediction result of the polypeptide sequence, wherein the antibacterial activity prediction result is the probability value of the polypeptide sequence having antibacterial activity; the pre-trained deep learning model is incrementally migrated using the second data set so that the incremental migration The trained deep learning model can output the prediction result of the antibacterial activity of the self-assembling polypeptide sequence with N-terminal modification, which specifically includes: using the data enhancement module to perform sequence-invariant noise enhancement on the input feature representation of the polypeptide sequence in the second data set generated by the feature representation unit, and inputting the input feature representation of the polypeptide sequence after noise enhancement into the pre-trained network to generate the hidden feature representation of the polypeptide sequence after noise enhancement; using the N-terminal modification encoding module to generate the N-terminal modification feature representation of the self-assembling polypeptide sequence in the second data set; inputting the spliced ​​N-terminal modification feature representation and the hidden feature representation of the polypeptide sequence after noise enhancement into the transfer learning network, and performing incremental migration training on the deep learning model in combination with the corresponding antibacterial activity annotation, and after the incremental migration training is completed, when the self-assembling polypeptide sequence with N-terminal modification is input into the deep learning model, the transfer learning network outputs the prediction result of the antibacterial activity of the self-assembling polypeptide sequence with N-terminal modification.

[0008] In some implementations, the amount of data in the second data set is much less than the amount of data in the first data set.

[0009] In some embodiments, the first data set is a public data set, which includes positive data and negative data, wherein the positive data is derived from the DBAASP database, and its polypeptide sequence has antibacterial activity; the negative data is derived from the Uniprot database, and its polypeptide sequence does not have antibacterial activity.

[0010] In some embodiments, the plurality of compounds specifically include: octanoic acid (C8-), lauric acid (C12-), hexadecanoic acid (C16-), phenylacetic acid (PHE-), 4-biphenylacetic acid (BIP-), diphenylacetic acid (DIP-), 2-naphthaleneacetic acid (NAP-), 9-anthracenecarboxylic acid (ANT-), 1-pyrenebutanoic acid (PYR-), cyclopropylacetic acid (C-PRO) and cyclohexylacetic acid (C-HEX).

[0011] In some embodiments, the generated self-assembling polypeptide sequences with N-terminal modifications are subjected to a polypeptide drug sensitivity test and the antimicrobial activity is annotated according to the test results, specifically including: when the polypeptide drug sensitivity test shows that the MIC is less than or equal to 100 μg / mL, annotating the self-assembling polypeptide sequence with N-terminal modification as having antimicrobial activity.

[0012] In some embodiments, the pre-trained network and the transfer learning network are constructed based on a Transformer network with a multi-head self-attention mechanism.

[0013] In some embodiments, the feature representation unit performs input feature representation on each polypeptide sequence in the first data set, specifically comprising: feature encoding 20 natural amino acids to generate 20 natural amino acid feature vectors {x j} j＝1,…,20 , where j is the number of the natural amino acid; when the length of the polypeptide sequence in the first data set is n, As the input feature representation of the polypeptide sequence, j1, j2, …, j n The method comprises: performing sequence-invariant noise enhancement on the input feature representation of the polypeptide sequence in the second data set generated by the feature representation unit using a data enhancement module, and inputting the noise-enhanced input feature representation of the polypeptide sequence into the pre-trained network after pre-training. Specifically, the method comprises:

[0014] According to formula (1), the characteristic vector x of each natural amino acid is calculated i The Euclidean distance d to the nearest neighbor of the remaining natural amino acid feature vector i :

[0015] d i =min({||x i -x j ||2} j＝1,…,20 ), formula (1)

[0016] Among them, ‖x i -x j ||2 represents x i With x j The Euclidean distance of

[0017] Each training step in the incremental transfer training is performed with x i Centered on Resample uniformly in the high-dimensional feature space with radius to obtain the equation (2) with x i The corresponding natural amino acid noise-enhanced feature vector x i ′:

[0018] x i ′=x i +ε i , formula (2)

[0019] Among them, ε i corresponds to x i The noise vector express Uniform distribution between

[0020] X′ in formula (3) is used as the input feature representation of the noise-enhanced polypeptide sequence corresponding to X:

[0021]

[0022] in, For the corresponding The noise vector For the corresponding The noise vector For the corresponding The noise vector Respectively Uniform distribution between .

[0023] Another object of the present application is to provide a biomimetic antimicrobial peptide having the structure shown below:

[0024]

[0025] Another object of the present application is to provide a pharmaceutical composition, which comprises the biomimetic antimicrobial peptide described in each embodiment of the present application, and an optional pharmaceutically acceptable carrier.

[0026] Another object of the present application is to provide an application of the biomimetic antimicrobial peptide described in each embodiment of the present application in the preparation of an antibacterial drug, or an application of the pharmaceutical composition described in each embodiment of the present application in the preparation of an antibacterial drug.

[0027] This application designs antimicrobial peptides with bionic functions based on deep learning and experimental interaction. Compared with the traditional drug molecule screening process, the entire process consumes less time, manpower, material resources and financial resources. In addition, the designed new bionic antimicrobial peptides are mainly composed of natural amino acids and have good biocompatibility. The functional groups modified at the N-terminus can activate the self-assembly activity of the polypeptide and help the polypeptide target the bacterial membrane. Therefore, it is not easy to induce bacterial resistance, and can effectively kill resistant bacteria with a faster sterilization speed. It has potential and broad application prospects in the treatment of clinical bacterial and drug-resistant bacterial infections. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In the drawings, which are not necessarily drawn to scale, the same reference numerals may describe similar parts in different views. The same reference numerals with letter suffixes or different letter suffixes may represent different instances of similar parts. The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the specification and claims, are used to illustrate the disclosed embodiments. When appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0029] Figure 1 A schematic flow chart of a method for designing a biomimetic antimicrobial peptide according to an embodiment of the present application is shown.

[0030] Figure 2 Eleven functional groups for modifying the N-terminus of a polypeptide sequence according to the examples of the present application are shown.

[0031] Figure 3(a) shows a schematic diagram of pre-training of a deep learning model according to an embodiment of the present application.

[0032] Figure 3(b) shows a schematic diagram of incremental migration training of a deep learning model according to an embodiment of the present application.

[0033] Figure 4 The chemical structural formula of polypeptide p45 according to an example of the present application is shown.

[0034] Figure 5 It is shown that the polypeptide p45 according to the embodiment of the present application has a stronger biological clearance ability than the traditional antibiotic ciprofloxacin, wherein: a shows the result of measuring the residual amount of biofilm after ciprofloxacin and polypeptide p45 treatment using the crystal violet method; b shows the result of observing the live / dead status of bacteria in the biofilm by combining SYTO 9 / PI staining and CLSM.

[0035] Figure 6The weight change (a) and survival rate (b) of mice infected with Salmonella typhimurium in the intestine after different treatments according to the examples of the present application are shown. PBS: PBS buffer; p45: polypeptide p45; Cip: ciprofloxacin. There are 5 mice in each group.

[0036] Figure 7 The contents of Salmonella Typhimurium in feces (a), small intestine (b), and colon (c) of mice infected with Salmonella Typhimurium in the intestine after different treatments according to the examples of the present application are shown. There are 5 mice in each group.

[0037] Figure 8 The following are HE, PAS, and AB-PAS stained sections of the small intestine of mice infected with Salmonella typhimurium after different treatments according to the embodiments of the present application. The black arrows are used to compare the appearance and length of the small intestine. The red box is used to magnify the morphology of the small intestinal villi, and the red arrows point to the area of ​​small intestinal villi lesions. Scale bar, 100 μm.

[0038] Fig. 9 The HE, PAS, and AB-PAS staining sections of the colon of mice infected with Salmonella typhimurium after different treatments according to the embodiments of the present application are shown. The red arrows point to the neutrophil infiltration area. Scale bar, 100 μm.

[0039] Fig.10 The following are the blood routine test results of mice after treatment in the acute toxicity and side effect study according to the embodiments of the present application.

[0040] Fig.11 The graph shows the blood biochemical results of mice after treatment in the acute toxicity study according to the embodiments of the present application.

[0041] Fig.12 Shown are H&E stained sections of the heart, liver, spleen, lung, kidney, small intestine, and colon of mice after treatment in the acute toxicity study according to the embodiments of the present application.

[0042] Fig.13 The following are the blood routine results of mice after treatment in the chronic toxic and side effect study according to the embodiments of the present application.

[0043] Fig.14 The graph shows the blood biochemical results of mice after treatment in the chronic toxic side effect study according to the embodiments of the present application.

[0044] Fig.15 Shown are H&E stained sections of the heart, liver, spleen, lung, kidney, small intestine, and colon of mice after treatment in the chronic toxicity study according to the embodiments of the present application.

[0045] Fig.16The figures show the results of 16srRNA detection and analysis of mouse fecal homogenate in an in vivo therapeutic effect study according to an embodiment of the present application to explore the recovery of intestinal flora.

[0046] In the above figure, ns means no significant difference, * means p value is less than 0.05, ** means p value is less than 0.01, *** means p value is less than 0.001. The smaller the p value, the more significant the difference between the two groups of data. DETAILED DESCRIPTION

[0047] The terms used herein are only used to explain the specific embodiments and are not used to limit the present application. A singular expression includes its plural expression unless it is explicitly stated or it is obvious from the context that it is not intended to do so.

[0048] Although various modifications can be made to the application and the application can have various forms, specific examples will be described in detail and explained below. However, it should be understood that these are not intended to limit the application to specific disclosures, and the application includes all modifications, equivalents or replacements thereof without departing from the spirit and technical scope of the application.

[0049] The present application will be further described in detail below through examples, but the scope of the present application is not limited to the examples.

[0050] In a first aspect, the present application provides a method for designing a biomimetic antimicrobial peptide. Figure 1 A schematic flow chart of a method for designing a biomimetic antimicrobial peptide according to an embodiment of the present application is shown.

[0051] like Figure 1 As shown, first in step 101, a first data set comprising a plurality of polypeptide sequences without N-terminal modification is constructed, wherein the polypeptide sequences in the first data set are composed of only 20 natural amino acids and are annotated with antibacterial activity.

[0052] In the present application, a polypeptide sequence marked as having antibacterial activity can also become positive data, while a polypeptide sequence marked as not having antibacterial activity is called negative data. In some embodiments, the above-mentioned first data set includes both positive data and negative data. As an example only, the first data set is a public data set, in which the positive data is derived from the DBAASP database, and the negative data is derived from the Uniprot database. In other embodiments, the positive data and negative data in the first data set can also come from other databases, as long as the antibacterial activity of the selected polypeptide sequence composed of 20 kinds of natural amino acids without N-terminal modification is verified by clear experiments. In some embodiments, the antibacterial activity can be marked by the test results of the polypeptide drug sensitivity test on the polypeptide sequence. More specifically, for example, when the polypeptide drug sensitivity test shows that the MIC (minimum inhibitory concentration) is less than or equal to 100 μg / mL, the polypeptide sequence is marked to have antibacterial activity. On the contrary, when the test result shows that the MIC is greater than 100 μg / mL, the polypeptide sequence is marked to have no antibacterial activity. In other embodiments, the in vitro antibacterial activity experimental screening method can also use the inhibition ring test instead of the MIC test to evaluate the antibacterial activity of the biomimetic self-assembling polypeptide, which is not limited in the present application.

[0053] Next, in step 102, for example, a polypeptide sequence can be selected from the first data set, and the selected polypeptide sequence can be N-terminally modified using a variety of compounds to generate a variety of self-assembling polypeptide sequences with N-terminal modifications (hereinafter also referred to as self-assembling polypeptide sequences); each generated self-assembling polypeptide sequence with N-terminal modification is subjected to a polypeptide drug sensitivity test and antibacterial activity is annotated according to the test results, and each self-assembling polypeptide sequence with N-terminal modification annotated with antibacterial activity is used as a second data set. Among them, the polypeptide sequence selected from the first data set can be originally antibacterial (positive data) or originally not antibacterial (negative data), which has no effect on the applicability of the design method described in the embodiment of the present application, and the present application does not limit this.

[0054] In the research practice of the present application, it was found that by using specific compounds to modify the N-terminus of a polypeptide sequence composed of natural amino acids, the self-assembly function of the polypeptide sequence can be greatly enhanced, so that this type of self-assembling polypeptide sequence containing N-terminal modification not only has good antibacterial properties, but also compared with traditional antimicrobial peptides such as Cathelicidin LL-37, Magainin2, Melittin, etc., it may also have better biocompatibility, longer in vivo half-life and is not easy to acquire drug resistance, and is expected to obtain a new antibacterial drug to replace traditional antibiotics.

[0055] Therefore, the present application creatively proposes to use eleven compounds to perform N-terminal modification on several non-N-terminal modified polypeptide sequences in the first data set, thereby generating a variety of self-assembling polypeptide sequences with N-terminal modifications. Just as an example, for example, eleven self-assembling polypeptide sequences containing different N-terminal modifications can be generated for each polypeptide sequence selected from the first data set, and each self-assembling polypeptide sequence containing N-terminal modifications can be tested for polypeptide drug sensitivity, and antibacterial activity can be annotated according to the test results, and each self-assembling polypeptide sequence with N-terminal modification annotated with antibacterial activity can be used as the second data set.

[0056] Figure 2 Eleven functional groups for modifying the N-terminus of a polypeptide sequence according to the examples of the present application are shown. Figure 2 As shown, the eleven compounds (functional groups) specifically include: octanoic acid (C8-), lauric acid (C12-), hexadecanoic acid (C16-), phenylacetic acid (PHE-), 4-biphenylacetic acid (BIP-), diphenylacetic acid (DIP-), 2-naphthylacetic acid (NAP-), 9-anthracenecarboxylic acid (ANT-), 1-pyrenebutanoic acid (PYR-), cyclopropylacetic acid (C-PRO) and cyclohexylacetic acid (C-HEX).

[0057] More specifically, for example, the peptide can be first synthesized by the classical Fmoc peptide solid phase synthesis method (SPPS), and then Figure 2 The various compounds shown are used to modify the N-terminus of the synthesized polypeptide sequence, and then purified by reverse high pressure liquid chromatography (RP-HPLC), and finally freeze-dried to obtain the final product. The synthesized polypeptide sequence (i.e., polypeptide molecule) containing the N-terminal modification can be detected for molecular weight and purity using mass spectrometry (MS) and liquid chromatography (HPLC). In some embodiments, for example, Escherichia coli (Gram-negative bacteria) and Staphylococcus aureus (Gram-positive bacteria) can be used to carry out polypeptide drug sensitivity testing. Specifically, 100 μL of polypeptide solutions of different concentrations (200 μg / mL, 100 μg / mL, 50 μg / mL, 25 μg / mL, 12.5 μg / mL) are added to a 96-well plate, and then an equal volume concentration of 106 CFU / mL of bacterial solution is added. After 18 hours, the 96-well plate is observed by the naked eye, and the hole with a clear solution and the highest concentration is the minimum inhibitory concentration (MIC) of the polypeptide. The antibacterial activity of the polypeptide molecule was evaluated according to the results (MIC ≤ 100 μg / mL was positive and had antibacterial activity), and then each self-assembling polypeptide sequence with N-terminal modification annotated with antibacterial activity was used as the second data set.

[0058] Then, in step 103, the deep learning model is pre-trained using the first data set, so that the pre-trained deep learning model can reconstruct the hidden feature representation of the polypeptide sequence and output the antibacterial activity prediction result of the polypeptide sequence.

[0059] In some embodiments, the goal of pre-training can be set into two parts, one part is used to accurately reconstruct the feature representation of the polypeptide sequence input into the deep learning model, and the other part is used to accurately predict the antibacterial activity of the polypeptide sequence without N-terminal modification. In this way, after pre-training using tens of thousands of polypeptide sequence data with antibacterial activity annotations in the first data set, the deep learning model can not only accurately predict the antibacterial activity of the polypeptide sequence without N-terminal modification, but also accurately identify and reconstruct the feature representation of the polypeptide sequence input into the model.

[0060] Next, in step 104, the pre-trained deep learning model is further trained with the second data set through incremental migration, so that the deep learning model trained through incremental migration can output the prediction results of the antibacterial activity of the self-assembling polypeptide sequence with N-terminal modification.

[0061] In the embodiments of the present application, through incremental transfer training, the finally trained deep learning model can accurately predict whether the self-assembling polypeptide sequence after various N-terminal modifications corresponding to the polypeptide sequence has antibacterial activity based on the accurate reconstruction of the feature representation of the polypeptide sequence without N-terminal modification.

[0062] Finally, in step 105, the polypeptide sequences in the first data set are modified at the N-terminus by using a plurality of compounds to generate a third data set containing a plurality of self-assembling polypeptide sequences with N-terminus modifications, and then the deep learning model trained by incremental migration is used to predict the antimicrobial activity of each self-assembling polypeptide sequence with N-terminus modifications in the third data set. For example, in the case where each polypeptide sequence in the first data set is modified at the N-terminus by using each of the eleven functional groups as described above, the number in the third data set will be eleven times the amount of data in the first data set, but the deep learning model usually runs on a high-performance processor, so the execution speed is very fast, and despite the huge amount of data, the antimicrobial activity of each self-assembling polypeptide sequence with N-terminus modifications in the third data set can still be quickly known.

[0063] On this basis, in step 105, the N-terminally modified self-assembling polypeptide sequences with predicted antibacterial activity results higher than a preset threshold can be further selected for synthesis and purification, and a polypeptide drug sensitivity test can be carried out. The N-terminally modified self-assembling polypeptide sequences that have passed the polypeptide drug sensitivity test (that is, the test results show that they have antibacterial activity) are sequentially subjected to in vitro experimental evaluation and therapeutic effect evaluation, and the N-terminally modified self-assembling polypeptide sequences whose antibacterial activity has been verified by the evaluation are used as the designed novel bionic antibacterial peptides.

[0064] In some embodiments, the prediction result of antibacterial activity is the probability value of the polypeptide sequence / self-assembling polypeptide sequence having antibacterial activity (that is, the value range is [0,1]), therefore, the above-mentioned preset threshold value can be determined based on the probability value, for example. In other embodiments, the probability values ​​of the predicted results of the antibacterial activity of each self-assembling polypeptide sequence can be arranged in order of size before being determined, so as to ensure that a suitable number of self-assembling polypeptide sequences with higher antibacterial activity are selected, so that subsequent synthesis and purification, drug sensitivity testing, in vitro experimental evaluation, treatment effect evaluation and other processes have an acceptable workload and an acceptable time cost. In an embodiment of the present application, for example, 140 self-assembling polypeptide sequences with the highest antibacterial activity can be selected for subsequent processes.

[0065] More specifically, drug sensitivity tests can be performed on Gram-negative bacteria (Escherichia coli and Salmonella typhimurium) and Gram-positive bacteria (Staphylococcus aureus and Listeria monocytogenes). In vitro experimental evaluation and therapeutic effect evaluation will be described in detail in the examples below.

[0066] According to the design method of the biomimetic antimicrobial peptide in the embodiment of the present application, the designed novel biomimetic antimicrobial peptide is obtained by N-terminal modification on the basis of natural amino acids, and has good biocompatibility. The functional group modified at the N-terminus can activate the self-assembly activity of the polypeptide and help the polypeptide target the bacterial membrane. Therefore, it is not easy to induce bacterial resistance, and can effectively kill resistant bacteria with a faster sterilization speed. It has potential and broad application prospects in the treatment of clinical bacterial and drug-resistant bacterial infection diseases.

[0067] In addition, due to the use of deep learning and test experiment interaction methods in the design process, the entire process consumes less time, manpower, material resources and financial resources compared to the traditional drug molecule screening process. Moreover, the deep learning model is divided into two stages: pre-training and incremental migration training, so that the deep learning model always predicts the antibacterial activity of the self-assembling polypeptide sequence on the basis of accurate reconstruction of the characteristic representation of the polypeptide sequence. Therefore, the prediction result is more accurate and reasonable, avoiding the subsequent large amount of experimental verification work being wasted on sequences that do not actually have antibacterial activity. Actual conditions have proved that the bionic antimicrobial peptide design method in this application can quickly screen a polypeptide sequence database of the order of 100 million, and combined with experimental verification and screening, it can screen up to 1011 polypeptide molecules within 4 weeks, and successfully design a bionic antimicrobial peptide with practical medicinal value, the structure of which is shown below:

[0068] Various experimental verifications of this novel biomimetic antimicrobial peptide will be described in detail later.

[0069] Table 1 compares the biomimetic antimicrobial peptide design method of the present application with the traditional antibiotic screening method.

[0070] Table 1 Comparison of the biomimetic antimicrobial peptide design method of this application and the traditional antibiotic screening method

[0071]

[0072] The composition and training process of the deep learning network according to the embodiment of the present application are described in detail below in conjunction with Figure 3 (a) and Figure 3 (b). The deep learning model in the present application may include three main components: a feature representation unit, a pre-trained network, and an incremental migration unit, wherein the incremental migration unit further includes a noise enhancement module, an N-terminal modification encoding module, and a transfer learning network. As an example only, the pre-trained network and the transfer learning network can be constructed based on a Transformer network with a multi-head self-attention mechanism, but not limited to this, the pre-trained network and the transfer learning network can also adopt any other neural network structure used.

[0073] FIG3( a) shows a schematic diagram of pre-training of a deep learning model according to an embodiment of the present application. As shown in FIG3( a), only the feature representation unit and the pre-training network are involved in the pre-training of the deep learning model. In some embodiments, the first data set may contain tens of thousands of polypeptide sequence data with antimicrobial activity annotations. For example, the first data set may be divided into a training data set, a validation data set, and a test data set in a ratio of 8:1:1. Specifically, each polypeptide sequence in the training data set is input into the feature representation unit, and the feature representation unit performs input feature representation on the input polypeptide sequence in a unified encoding method; then, the input feature representation of the polypeptide sequence is input into the pre-training network. In the training optimization process of pre-training, for example, the sum of the L2 sequence reconstruction error of the model and the binary cross entropy of the antimicrobial activity prediction result can be used as the loss function, and the AdaM optimizer is used to update the network weights and completely traverse the training data in the training data set 100 times. Among them, the L2 sequence reconstruction error refers to the error between the hidden feature representation of the reconstructed polypeptide sequence and the input feature representation of the polypeptide sequence in the training data set, and the binary cross entropy of the antimicrobial activity prediction result refers to the binary cross entropy between the antimicrobial activity prediction result of the polypeptide sequence without N-terminal modification output by the deep learning model during pre-training and the antimicrobial activity annotation of the polypeptide sequence. Since the loss function of pre-training consists of the above two parts, the deep learning model after pre-training can not only accurately output the prediction result of the antimicrobial activity of the polypeptide sequence without N-terminal modification by the pre-training network, wherein the antimicrobial activity prediction result can be, for example, the probability value of the polypeptide sequence having antimicrobial activity, but also the feature representation of the polypeptide sequence input into the deep learning model can be accurately identified and reconstructed by the pre-training network.

[0074] Since the antimicrobial peptides intended to be designed in the present application are not polypeptide sequences composed only of natural amino acids, but are self-assembling antimicrobial peptides with N-terminal modifications, after pre-training the deep learning model using the polypeptide sequences in the first data set that do not contain N-terminal modifications, further supplementary training is required to enable it to accurately predict the antimicrobial activity of self-assembling polypeptide sequences containing N-terminal modifications.

[0075] FIG3( b) shows a schematic diagram of incremental migration training of a deep learning model according to an embodiment of the present application. As shown in FIG3( b), in the incremental migration training, each component of the deep learning model must participate in the training. It is worth noting that since the self-assembling polypeptide sequence containing N-terminal modification in the second data set needs to be tested for polypeptide drug sensitivity in order to annotate the antibacterial activity according to the test results, the workload is large, so the amount of data is much smaller than the amount of data in the first data set. In this case, the impact of the small-volume second data set on the deep learning model is easily diluted by the impact of the large-volume first data set. In other words, if the polypeptide sequence data in the second data set is not processed and is directly used for incremental migration training of the deep learning model, then after the incremental migration training, the deep learning model may only be able to accurately predict whether the self-assembling polypeptide sequence containing N-terminal modification corresponding to the trained polypeptide sequence has antibacterial activity, but the antibacterial activity of the self-assembling polypeptide sequence containing N-terminal modification generated based on other polypeptide sequences cannot be well predicted.

[0076] In order to solve the above technical problems, the present application creatively proposes that in the incremental migration training stage, as shown in FIG3(b), a data enhancement module is added between the feature representation unit and the pre-trained network, and the input feature representation of the polypeptide sequence in the second data set generated by the feature representation unit is subjected to sequence-invariant noise enhancement by using the data enhancement module, and then the input feature representation of the polypeptide sequence after noise enhancement is input into the pre-trained network after pre-training. Moreover, the noise enhancement of the input polypeptide sequence by the above data enhancement module is performed with each training step in the incremental migration training, and the noise enhancement method changes randomly each time. On the one hand, this method greatly expands the amount of data used for incremental migration training, making up for the defect that the amount of data in the second data set is much smaller than that in the first data set, and at the same time, no additional storage or time requirements are generated due to data enhancement.

[0077] An implementation method of the feature representation unit and the data enhancement module is introduced below by way of example.

[0078] In some embodiments, the feature representation unit is used to represent the input features of the polypeptide sequence (specifically, the polypeptide sequence that does not contain N-terminal modification and is composed of only 20 natural amino acids). It is easy to understand that in the pre-training stage, the feature representation unit performs input feature representation on each polypeptide sequence in the first data set, and in the incremental migration training stage, similarly, the feature representation unit is used to represent the input features of the polypeptide sequence portion composed of natural amino acids (i.e., does not contain compounds for N-terminal modification) in each self-assembled polypeptide sequence in the second data set. The specific feature representation method is as follows.

[0079] First, the 20 natural amino acids are feature encoded using an appropriate encoding method such as One-Hot Encoding to generate 20 natural amino acid feature vectors {x j} j＝1,…,20 , where j is the number of the natural amino acid; when the length of the polypeptide sequence in the first data set (or the second data set) is n, As the input feature representation of the polypeptide sequence, j1, j2, …, j n is the number of each natural amino acid in the polypeptide sequence. As an example only, the natural amino acids can be numbered according to the numbering method commonly used in the art, for example, the single-letter abbreviation of alanine is A, and its number is 1; the single-letter abbreviation of cysteine ​​is C, and since there is no natural amino acid with a single-letter abbreviation B, its number is 2, and so on. In other embodiments, feature encoding can be performed for the 20 natural amino acids according to other numbering and encoding methods, and corresponding input feature representations can be generated for polypeptide sequences of different lengths and compositions.

[0080] Different from the pre-training stage, in the incremental transfer training stage, the input feature representation of the peptide sequence output by the feature representation unit is not directly input into the pre-training network, but input into the data enhancement module, which performs sequence-invariant noise enhancement on the input feature representation of the peptide sequence. The specific method is as follows:

[0081] According to formula (1), the characteristic vector x of each natural amino acid is calculated i The Euclidean distance d to the nearest neighbor of the remaining natural amino acid feature vector i :

[0082] d i =min({||x i -x j ||2} j＝1,…,20 ), formula (1)

[0083] Among them, ||x i -x j ||2 represents x i With x j The Euclidean distance of

[0084] Each training step in the incremental transfer training is performed with x i Centered on Resample uniformly in the high-dimensional feature space with radius to obtain the equation (2) with x i The corresponding natural amino acid noise-enhanced feature vector x i ′:

[0085] x′ i =x i +ε i , formula (2)

[0086] Among them, ε i corresponds to x i The noise vector express Uniform distribution between

[0087] X′ in formula (3) is used as the input feature representation of the noise-enhanced polypeptide sequence corresponding to X:

[0088]

[0089] in, To correspond to The noise vector To correspond to The noise vector To correspond to The noise vector Respectively Uniform distribution between .

[0090] According to the above method, the discrete sequence feature space can be continuous while retaining the representation of the sequence itself, that is, the sequence feature after noise addition can be approximately considered as an augmentation representation of the original sequence. In the present application, it means that there is no need to re-verify the antibacterial activity of the self-assembling polypeptide sequence after noise addition. It can be seen that the present application proposes an innovative method that can significantly enhance the features of discrete sequence data without increasing manpower, material resources and time costs, or even without increasing the amount of training data. The input feature representation of the polypeptide sequence after noise addition and enhancement is input into the pre-training network, and the pre-training network can use its ability to reconstruct the hidden feature representation of the polypeptide sequence learned in the pre-training stage to generate the corresponding hidden feature representation for the self-assembling polypeptide sequence after noise addition and enhancement.

[0091] At the same time, the eleven compounds used for N-terminal modification of the self-assembling polypeptide sequence in the second data set are input into the N-terminal modification encoding module to generate the N-terminal modification feature representation, and then the N-terminal modification feature representation is spliced ​​with the hidden feature representation of the polypeptide sequence after noise enhancement, and then the N-terminal modification feature representation is input into the transfer learning network as the feature representation of the self-assembling polypeptide sequence with N-terminal modification after noise enhancement, and the deep learning model including the pre-training network and the transfer learning network is jointly trained for incremental migration in combination with the corresponding antibacterial activity annotation. In the training process of incremental migration, similarly, the data in the second data set can also be divided into a training data set, a validation data set and a test data set in a ratio of 8:1:1, and then, only the binary cross entropy of the antibacterial activity prediction result is used as the loss function, and the noise vector used to enhance the sequence representation is resampled at each training step, and the network weight is updated with the AdaM optimizer and the training data set is completely traversed 50 times. After each incremental migration learning training, the deep learning model and the hyperparameters used in the training process are adjusted according to the performance parameters of the deep learning model (such as accuracy, precision and recall), and finally the best performing deep learning model is obtained.

[0092] After the incremental transfer training is completed, the deep learning model has the ability to predict the antibacterial activity of the self-assembling polypeptide sequence containing the N-terminal modification. Therefore, when the self-assembling polypeptide sequence with the N-terminal modification is input into the deep learning model, the transfer learning network will output the prediction result of the antibacterial activity of the self-assembling polypeptide sequence. Consistent with the above, the prediction result of the antibacterial activity here can also be the probability value of the self-assembling polypeptide sequence having antibacterial activity.

[0093] According to the above-mentioned design method of biomimetic antimicrobial peptide, a new type of biomimetic antimicrobial peptide can be designed. Therefore, according to the embodiment of the present application, a biomimetic antimicrobial peptide is also provided, which has the following structure:

[0094]

[0095] According to an embodiment of the present application, a pharmaceutical composition is also provided, which comprises the biomimetic antimicrobial peptide according to the embodiment of the present application, and an optional pharmaceutically acceptable carrier.

[0096] According to the embodiments of the present application, the use of the biomimetic antimicrobial peptide or the pharmaceutical composition in the present application in the preparation of antimicrobial drugs is also provided.

[0097] Hereinafter, the experimental materials, instruments and experimental methods involved in the design method of the biomimetic antimicrobial peptide of the present application, or the biomimetic antimicrobial peptide of the present application, the pharmaceutical composition containing the biomimetic antimicrobial peptide of the present application, and the biomimetic antimicrobial peptide / pharmaceutical composition of the present application in the process of preparing antimicrobial drugs will be described in detail through some examples. However, the examples provided herein are for illustrative purposes only and are not intended to limit the present application.

[0098] Unless otherwise specified, the experimental methods used in the following examples are conventional methods.

[0099] Unless otherwise specified, the materials and reagents used in the following examples can be obtained from commercial sources.

[0100] Experimental materials and instruments

[0101] Culture medium, RMPI 1640, purchased from ThermoFisher Scientific, sterile;

[0102] Fetal bovine serum was purchased from ThermoFisher Scientific, sterile;

[0103] 2-Cl resin was purchased from Tianjin Nankai Hecheng Technology Co., Ltd., 1.1 mmol / mg;

[0104] N,N-Diisopropylethylamine (DIPEA), purchased from Sigma-Aldrich, purity 99%;

[0105] Trifluoroacetic acid (TFA), purchased from Sigma-Aldrich, purity 99%;

[0106] Triisopropylsilane (TIS) was purchased from Sigma-Aldrich with a purity of 99%;

[0107] Gastric epithelial cells (GES-1) were purchased from Hunan Fenghui Biotechnology Co., Ltd.;

[0108] Octanoic acid (C8-) and cyclohexylacetic acid (C-HEX-) were purchased from Shanghai Haohong Biopharmaceutical Technology Co., Ltd. with a purity of 98%;

[0109] Lauric acid (C12-), purchased from Shanghai Yuanye Biotechnology Co., Ltd., purity 98%;

[0110] Hexadecanoic acid (C16-), purchased from Shanghai Sane Chemical Technology Co., Ltd., purity 98%;

[0111] 4-Biphenylacetic acid (BIP-), 2-naphthylacetic acid (NAP-), and 1-pyrenebutyric acid (PYR-) were purchased from Shanghai Aladdin Biochemical Technology Co., Ltd. with a purity of 98%;

[0112] Diphenylacetic acid (DIP-), 9-anthracenecarboxylic acid (ANT-), and cyclopropylacetic acid (C-PRO-) were purchased from Shanghai MacLean Biochemical Technology Co., Ltd. with a purity of 98%;

[0113] Phenylacetic acid (PHE) was purchased from Shanghai Sinopharm Group with a purity of 98%;

[0114] L-configuration amino acids were purchased from Jier Biochemical (Shanghai) Co., Ltd. with a purity of 98%;

[0115] Varioskan LUX multifunctional microplate reader, purchased from Thermo Fisher Scientific;

[0116] Escherichia coli (ATCC 25922), Staphylococcus aureus (ATCC 25923), Enterococcus faecium (ATCC 51559), Enterococcus faecalis (ATCC 51575, ATCC 51299) from the American Cell Culture Collection (ATCC);

[0117] Listeria monocytogenes (L. monocytogenes, CMCC 54004), from the China Medical Culture Collection Center (CMCC);

[0118] E. cancerogenus (BNCC 363037), Staphylococcus epidermidis (S. epidemidis, BNCC330867), Acinetobacter baumannii (A. baumannii, BNCC 254392), Escherichia coli (E. coli, BNCC 186732), from BNCC;

[0119] Salmonella typhimurium (S.typhimurium, SL 1344) and Staphylococcus aureus (S.aureus, USA300) were obtained from the laboratory bacterial strain library;

[0120] SYTO 9 / PI live bacteria / dead bacteria double staining kit was purchased from Beijing Langbolide Trading Co., Ltd.;

[0121] Hematoxylin and eosin combined staining (H&E) kit was purchased from Wuhan Saiweier Biotechnology Co., Ltd.;

[0122] PAS and AB-PAS staining kits were purchased from Wuhan Saiweier Biotechnology Co., Ltd.

[0123] Example of experimental method

[0124] Using deep learning models and the interaction between deep learning and test experiments, the process of efficiently, cheaply and accurately predicting the antibacterial activity of each self-assembling peptide sequence with N-terminal modification in the third dataset has been combined Figure 1 , Figure 2 3(a) and 3(b) are described in detail. The sensitivity test method of peptide drugs is also described in the previous text. Figure 1 The synthesis of the self-assembling polypeptide sequence with N-terminal modification involved in step 102 and step 105, in vitro experimental evaluation, therapeutic effect evaluation, and finally obtaining the experimental process of the self-assembling polypeptide sequence with N-terminal modification with medicinal value obtained in the present application are illustrated by way of example.

[0125] Example of synthesizing biomimetic antimicrobial peptides using Fmoc-solid phase synthesis method

[0126] (1) Weigh 0.5 mmol of 2-Cl resin into a solid phase synthesis tube, add 10 mL of anhydrous dichloromethane (DCM), and shake on a shaker for 10 min to ensure that the 2-Cl resin is fully swollen;

[0127] (2) Use an ear bulb to squeeze out the DCM from the solid phase synthesis tube;

[0128] (3) Dissolve 1 mmol of Fmoc-protected amino acid in 10 mL of anhydrous DCM, add 2 mmol of N,N-diisopropylethylamine (DIPEA), and then transfer to a solid phase synthesis tube and react at room temperature for 1 hour. Remove the reaction solution in the solid phase synthesis tube with an ear bulb, and then wash with anhydrous DCM, using 10 mL of DCM each time, washing for 1 minute, and washing 5 times in total. After washing, add 20 mL of the prepared blocking reaction solution (anhydrous DCM: DIPEA: methanol = 17:1:2) and react at room temperature for 15 minutes;

[0129] (4) Use an ear bulb to remove the reaction solution in the solid phase synthesis tube, and wash it with anhydrous DCM, using 10 mL of DCM each time, washing for 1 min, and washing for a total of 5 times;

[0130] (5) washing with N,N-dimethylformamide (DMF), using 10 mL of DMF each time for 1 min, for a total of 5 times, adding 10 mL of a 20% by volume piperidine DMF solution, reacting for 30 min, and then washing with DMF, using 10 mL of DMF each time for 1 min, for a total of 5 times;

[0131] (6) Add 3 mmol of the second Fmoc-protected amino acid, 4.5 mmol of HBTU, and 6 mmol of DIPEA in 20 mL of DMF, add the prepared reaction solution to the solid phase synthesis tube, and react for 2 h;

[0132] (7) Repeat the steps (5) and (6) to sequentially add amino acids or N-terminal modification groups in the peptide sequence; then wash with DMF 5 times and then with DCM 5 times;

[0133] (8) A solution containing 95% TFA, 2.5% TIS, and 2.5% H2O by volume was added to a solid phase synthesis tube and reacted for 2 h. The reaction solution was discharged and collected using an ear bulb, and the solvent was removed by rotary evaporation to obtain a crude product, which was then purified using HPLC.

[0134] Examples of in vitro evaluations

[0135] In vitro experimental evaluation of novel biomimetic antimicrobial peptides may include, for example, in vitro cytotoxicity experiments, in vitro hemolytic experiments, in vitro broad-spectrum antimicrobial experiments, biofilm clearance experiments, etc.

[0136] (1) In vitro cytotoxicity assay

[0137] Gastric epithelial cells (GES-1) were inoculated in 96-well plates at a density of 10,000 per well. After the cells adhered to the wall and grew, peptide solutions of different concentrations (1000 μg / mL, 500 μg / mL, 250 μg / mL, 125 μg / mL, 63 μg / mL, 31 μg / mL) were prepared in RMPI 1640 medium. A control group (medium, medium + cells) was set up on each plate. After incubation for 24 hours, the 50% cell lethal concentration (EC50) was detected and calculated by the MTT method. The selectivity coefficient for gastric epithelial cells was calculated as follows: SI_E represents the selectivity coefficient (cytotoxicity), EC50 represents the concentration of the peptide that causes 50% cell death, and MIC represents the minimum inhibitory concentration.

[0138] (2) In vitro hemolytic test

[0139] Rabbit red blood cells (rRBC) were diluted to 8% and 100 μL was added to each well of a 96-well plate. Then 100 μL of peptide solutions of different concentrations (1000 μg / mL, 500 μg / mL, 250 μg / mL, 125 μg / mL, 63 μg / mL, 31 μg / mL) were added. A control group (cell suspension + 1% Triton X-100, cell suspension + PBS) was set up on each plate. After incubation for 3 hours, OD570 was read by an ELISA reader and the 50% hemolysis concentration (HC50) was calculated. The selectivity coefficient for mouse red blood cells was calculated as follows: SI_H represents the selectivity coefficient (mouse red blood cells), HC50 represents the concentration of the peptide that causes 50% hemolysis, and MIC represents the minimum inhibitory concentration. SI_E represents the selectivity coefficient (human gastric epithelial cells), EC50 represents the concentration of the peptide that causes 50% cell death, and MIC represents the minimum inhibitory concentration. The experimental results are shown in Table 2.

[0140] Table 2 MIC values ​​of candidate peptides against Salmonella typhimurium, 50% hemolytic concentration (HC50) for mouse red blood cells, 50% lethal concentration (EC50) for human gastric mucosal epithelial cells, and selectivity coefficients (SI_H (mouse red blood cells), SI_E (human gastric mucosal epithelial cells)), unit: μg / mL.

[0141]

[0142] As shown in Table 2, in further in vitro experimental screening, peptide p45 showed the best biocompatibility (highest selectivity) and the strongest antibacterial activity (lowest MIC value). At the same time, the MIC value of p45 without N-terminal modification was greater than 100 μg / mL, proving that without N-terminal modification, the peptide lost its antibacterial activity. The structure of the best self-assembling antibacterial peptide p45 screened out is shown in Figure 4 shown.

[0143] (3) In vitro broad-spectrum antibacterial activity test

[0144] As shown in Table 3, the MIC values ​​of the verified self-assembled antimicrobial peptides against various bacteria (E. coli, S. aureus, S. typhimurium, L. monocytogenes, E. faecium, E. faecalis, E. cancerogenus, S. epidemidis, A. baumannii, etc.) were measured by MIC experiments to evaluate their broad-spectrum antibacterial properties.

[0145] Table 3 Broad-spectrum antibacterial activity of the verified self-assembling antimicrobial peptides against a variety of bacteria

[0146]

[0147] Finally, by comparing the above results and the bacterial selectivity coefficients (EC50 / MIC, HC50 / MIC), the biomimetic antimicrobial peptides with good biocompatibility and broad-spectrum antibacterial activity were screened out.

[0148] (4) Biofilm removal experiment

[0149] The Salmonella typhimurium biofilm was incubated on the surface of the culture dish, and 1 mL of PBS buffer, ciprofloxacin (concentration: 20×MIC, 20 μg / mL) and p45 (concentration: 20×MIC, 125 μg / mL) peptide were added for 3 hours, and then the residual biofilm content was detected using the crystal violet test method. SYTO 9 / PI (live / dead) staining combined with confocal microscopy was used to observe the live and dead status of the treated biofilm. Figure 5 As shown, peptide p45 has stronger bio-clearance ability than the traditional antibiotic ciprofloxacin.

[0150] Example of evaluating the therapeutic effect of novel biomimetic antimicrobial peptides

[0151] (1) In vivo therapeutic effect study

[0152] A mouse model of intestinal infection was constructed using Salmonella typhimurium. Female BALB / c mice aged 6-8 weeks (purchased from Shanghai Jihui Experimental Animal Breeding Co., Ltd., ethics approval number: AP#22-025-WHM, with a breeding environment temperature of 20-26°C, a relative humidity of 50-60%, and a light cycle of 12 hours of light / 12 hours of darkness) were selected. After fasting for 4 hours, each mouse was intraperitoneally injected with 20 mg of streptomycin sulfate. After resuming the diet for 20 hours, fasting for another 4 hours, and then 100 μL of Salmonella typhimurium solution (10 9 CFU / mL, PBS) was gavaged to each mouse. After 2 days of diet recovery, the mice were randomly divided into groups (5 per group) for subsequent experiments. Each group of infected mice was intraperitoneally injected with 200 μL PBS buffer, ciprofloxacin and peptide p45 at a dose of 30 mg / kg. The survival and weight changes of each group of mice were recorded every day ( Figure 6 On the fifth day after treatment, fresh feces of mice were collected and the mice were killed. The small intestine and colon tissues were taken, homogenized, plated, cultured, counted and the contents of Salmonella typhimurium in feces, small intestine and colon were calculated ( Figure 7 Then, the H&E and PAS staining sections of the small intestine and colon of the mice were compared and observed ( Figure 8 , Fig. 9) to study the recovery of the intestine after treatment (inflammation, small intestinal villi, intestinal cells, etc.). At the same time, Wuhan Sewell Biotechnology Co., Ltd. was commissioned to perform 16s rRNA detection and analysis on mouse fecal homogenate to explore the recovery of intestinal flora. The experimental results are as follows Fig.16 As shown, it can be seen that the bacterial flora after p45 treatment and after antibiotic cip treatment are closer, that is, the therapeutic effect of the present invention is equivalent to that of the traditional antibiotic cip.

[0153] (2) Toxic and side effect studies

[0154] Female BALB / c mice aged 6-8 weeks were selected and randomly divided into groups (5 mice in each group, a total of 6 groups) for toxicity and side effect research. The same treatment regimen as in the treatment experiment was used to inject healthy mice with biomimetic antimicrobial peptides and PBS solution. Blood was collected from 3 groups 24 hours after treatment for blood routine tests (red blood cells (RBC), mean corpuscular volume (MCV), platelets (PLT), white blood cells (WBC), neutrophils (NEUT), monocytes (MONO)) ( Fig.10 ) and blood biochemistry (alanine aminotransferase (ALT), aspartate aminotransferase (AST), blood urea nitrogen (BUN), uric acid (UA), creatinine (CREA)) Fig.11 ) tests and collected data from different organs (heart, liver, spleen, lung, kidney, small intestine, colon) ( Fig.12 ) were scanned with H&E staining, and the acute toxicity and side effects of the treatment were analyzed based on the results. The survival rate and weight changes of the remaining three groups were recorded after treatment, and the same blood routine test was performed on the 7th day ( Fig.13 )、Blood Biochemistry( Fig.14 ), tissue staining section experiment ( Fig.15 ) to analyze the chronic toxicity and side effects of treatment options.

[0155] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art can use other embodiments when reading the above description. In addition, in the above-mentioned specific embodiments, various features can be grouped together to simplify the application. This should not be interpreted as an intention that a disclosed feature that does not require protection is necessary for any claim. On the contrary, the subject matter of the present invention may be less than all the features of a specific disclosed embodiment. Thus, the claims are incorporated into the specific embodiments as examples or embodiments, wherein each claim is independently used as a separate embodiment, and it is considered that these embodiments can be combined with each other in various combinations or arrangements. The scope of the present invention should be determined with reference to the attached claims and the full scope of equivalent forms granted by these claims.

Claims

1. A method for designing a biomimetic antimicrobial peptide, characterized in that: The design method comprises: Constructing a first data set comprising a plurality of polypeptide sequences without N-terminal modification, wherein the polypeptide sequences in the first data set consist of only 20 natural amino acids and are annotated with antibacterial activity; Using a plurality of compounds to perform N-terminal modification on the polypeptide sequences selected from the first data set to generate a plurality of self-assembling polypeptide sequences with N-terminal modification; performing a polypeptide drug sensitivity test on each of the generated self-assembling polypeptide sequences with N-terminal modification and annotating the antibacterial activity according to the test results, and using each of the self-assembling polypeptide sequences with N-terminal modification annotated with antibacterial activity as a second data set; Pre-training a deep learning model using the first data set, so that the pre-trained deep learning model can reconstruct the hidden feature representation of the polypeptide sequence and output the prediction result of the antibacterial activity of the polypeptide sequence; Using the second data set to perform incremental migration training on the pre-trained deep learning model, so that the deep learning model trained by incremental migration can output the prediction result of the antibacterial activity of the self-assembling polypeptide sequence with N-terminal modification; The peptide sequences in the first data set are modified at the N-terminus using a variety of compounds to generate a third data set containing multiple self-assembling peptide sequences with N-terminus modifications, and the antibacterial activity of each self-assembling peptide sequence with N-terminus modifications in the third data set is predicted using the deep learning model trained by incremental migration. The self-assembling peptide sequences with N-terminus modifications whose predicted antibacterial activity results are higher than a preset threshold are selected for synthesis and purification, and peptide drug sensitivity testing is carried out. The self-assembling peptide sequences with N-terminus modifications that pass the peptide drug sensitivity test are sequentially subjected to in vitro experimental evaluation and therapeutic effect evaluation, and the self-assembling peptide sequences with N-terminus modifications whose antibacterial activity has been verified by the evaluation are used as the designed new bionic antimicrobial peptides.

2. The design method according to claim 1, characterized in that: The deep learning model includes a deep learning model of a feature representation unit, a pre-trained network and an incremental migration unit, wherein the incremental migration unit includes a noise enhancement module, an N-terminal modification encoding module and a migration learning network; Pre-training the deep learning model using the first data set so that the pre-trained deep learning model can reconstruct the hidden feature representation of the polypeptide sequence and output the antibacterial activity prediction result of the polypeptide sequence specifically includes: inputting feature representation of each polypeptide sequence in the first data set by the feature representation unit, training and optimizing the pre-trained network based on the input feature representation and antibacterial activity annotation of the polypeptide sequence, and outputting the reconstructed hidden feature representation of the polypeptide sequence and the antibacterial activity prediction result of the polypeptide sequence by the pre-trained network, wherein the antibacterial activity prediction result is the probability value of the polypeptide sequence having antibacterial activity; The method uses the second data set to perform incremental transfer training on the pre-trained deep learning model so that the deep learning model trained by incremental transfer can output the prediction result of the antibacterial activity of the self-assembling polypeptide sequence with N-terminal modification. Specifically, the method includes: using the data enhancement module to perform sequence-invariant noise enhancement on the input feature representation of the polypeptide sequence in the second data set generated by the feature representation unit, and inputting the input feature representation of the polypeptide sequence after noise enhancement into the pre-trained network to generate the hidden feature representation of the polypeptide sequence after noise enhancement; using the N-terminal modification encoding module to generate the N-terminal modification feature representation of the self-assembling polypeptide sequence in the second data set; inputting the spliced ​​N-terminal modification feature representation and the hidden feature representation of the polypeptide sequence after noise enhancement into the transfer learning network, and performing incremental transfer training on the deep learning model in combination with the corresponding antibacterial activity annotation; and after the incremental transfer training is completed, when the self-assembling polypeptide sequence with N-terminal modification is input into the deep learning model, the transfer learning network outputs the prediction result of the antibacterial activity of the self-assembling polypeptide sequence with N-terminal modification.

3. The design method according to claim 1 or 2, characterized in that: The amount of data in the second data set is much less than the amount of data in the first data set.

4. The design method according to claim 1 or 2, characterized in that: The first data set is a public data set, which includes positive data and negative data, wherein: The positive data came from the DBAASP database, and its peptide sequence had antibacterial activity; Negative data came from the Uniprot database, and its peptide sequence had no antibacterial activity.

5. The design method according to claim 1 or 2, characterized in that: The various compounds specifically include: octanoic acid (C8-), lauric acid (C12-), hexadecanoic acid (C16-), phenylacetic acid (PHE-), 4-biphenylacetic acid (BIP-), diphenylacetic acid (DIP-), 2-naphthylacetic acid (NAP-), 9-anthracenecarboxylic acid (ANT-), 1-pyrenebutyric acid (PYR-), cyclopropylacetic acid (C-PRO) and cyclohexylacetic acid (C-HEX).

6. The design method according to claim 1 or 2, characterized in that: The generated self-assembling polypeptide sequences with N-terminal modifications are subjected to polypeptide drug sensitivity tests and antimicrobial activity annotations are performed based on the test results, specifically including: When the peptide drug sensitivity test shows that the MIC is less than or equal to 100 μg / mL, the self-assembling peptide sequence with N-terminal modification is marked as having antibacterial activity.

7. The design method according to claim 2, characterized in that: The pre-trained network and the transfer learning network are constructed based on a Transformer network with a multi-head self-attention mechanism.

8. The design method according to claim 2, characterized in that: The feature representation unit performs input feature representation on each polypeptide sequence in the first data set, specifically including: The 20 natural amino acids are feature encoded to generate 20 natural amino acid feature vectors {x j } j＝1,…,20 , where j is the number of the natural amino acid; when the length of the polypeptide sequence in the first data set is n, As the input feature representation of the polypeptide sequence, j1, j2, …, j n is the number of each natural amino acid in the polypeptide sequence; The method of performing sequence-invariant noise enhancement on the input feature representation of the polypeptide sequence in the second data set generated by the feature representation unit using the data enhancement module, and inputting the noise-enhanced input feature representation of the polypeptide sequence into the pre-trained pre-trained network specifically includes: According to formula (1), the characteristic vector x of each natural amino acid is calculated i The Euclidean distance d to the nearest neighbor of the remaining natural amino acid feature vector i : d i =min({||x i -x j ||2} j＝1,…,20 ), formula (1) Among them, ||x i -x j ||2 represents x i With x j The Euclidean distance of Each training step in the incremental transfer training is performed with x i Centered on Resample uniformly in the high-dimensional feature space with radius to obtain the equation (2) with x i The corresponding natural amino acid noise-enhanced feature vector x′ i : x′ i =x i +ε i , formula (2) Among them, ε i corresponds to x i The noise vector express Uniform distribution between X′ in formula (3) is used as the input feature representation of the noise-enhanced polypeptide sequence corresponding to X: in, To correspond to The noise vector To correspond to The noise vector To correspond to The noise vector Respectively Uniform distribution between .

9. A biomimetic antimicrobial peptide having the structure shown below:

10. A pharmaceutical composition comprising the biomimetic antimicrobial peptide according to claim 9, and optionally a pharmaceutically acceptable carrier.

11. Use of the biomimetic antimicrobial peptide according to claim 9 or the pharmaceutical composition according to claim 10 in the preparation of antimicrobial drugs.

Images