Systems and methods for enhancing the stability of synthetic proteins

Abstract

To provide a mutant protein with improved properties.SOLUTION: The invention provides a protein comprising a secBFP2 variant having one or more mutations at one or more residues selected from T18, S28, Y96, V124, T127, D151, N173 and R198, in relation to full-length wild-type secBFP2. The invention also provides: a nucleic acid molecule comprising a nucleotide sequence encoding the protein; and a composition and a kit which comprise the nucleic acid molecule.SELECTED DRAWING: None

Description

Technical Field

[0001] Related Applications This application claims the benefit of and priority to U.S. Provisional Patent Application No. 62 / 841,906, filed May 2, 2019, entitled "System and Method for Increasing Synthesized Protein Stab ility", the entirety of which is incorporated herein by reference is claimed.

[0002] Statement Regarding Federally Sponsored Research or Development This invention was made with government support under Grant No. R43 NS105463 awarded by the National Institutes of Health, and Grant No. F A9550-14-1-0089 awarded by the Air Force Offi ce of Scientific Research. The government has certain rights in this invention.

Background Art

[0003] Protein engineering is a transformative approach in biotechnology and biomedicine, with the goal of either imparting novel functionality to existing proteins or rendering proteins more durable in non-natural environments. A design consideration that affects both engineering approaches is the overall stability of the protein. In the former case, gain-of-function mutations that expand the role of a protein through rational design or directed evolution are often introduced at a thermodynamic cost. Since most natural proteins are only marginally stable, it has been shown that pre-selection stability enhancement promotes the evolvability of proteins, while ​​​​Therefore, functional mutations that destabilize proteins to the point of unfolding are likely to be overlooked. It is possible.

[0004] A major obstacle in converting useful naturally occurring biocatalysts to industrial applications is fundamentally This involves the adaptation of proteins to different environmental conditions, temperatures, and solvents. It also involves improving protein stability. Increasing production will alleviate many of these pressures and enable mass production with higher yields and lower costs. This can enable its expression. Therefore, stabilization requires many protein engineering efforts. It is essential for its success.

[0005] There are many ways to manipulate proteins, and generally speaking, all of them involve protein variants. How quickly and accurately can it be measured, and how efficiently can the protein variant status be measured? It represents a compromise between whether it can be sampled appropriately and whether it can be sampled. Mutagenic polymerase chain reaction (PC) Techniques such as R require minimal knowledge of the relationship between sequences and functions, Even so, to isolate large libraries of protein variants, high-slope technology is still required. It depends on the screen or selection. It uses structured data and computer computation approaches. By using these methods, the search space can be narrowed, and at the same time, the amount of downstream characteristic evaluation can be reduced. The tools are increasingly useful for proteins where the desired properties are particularly difficult to measure, especially on a large scale. It will become important. However, our understanding of the relationship between protein sequence / structure / function is incomplete. Different computer calculation tools for protein engineering are often completely different, or It offers even contradictory solutions. This is particularly relevant to properties such as stability and folding. However, these are often many small ones distributed throughout the entire protein sequence. This is the result of the interaction.

[0006] Typically, computer calculation methods are computationally intensive folding systems. By performing simulations, we identify residues that destabilize proteins. The level of detail involved in these simulations varies, drawing on quantum mechanics (MOE). Some methods go so far as to explain molecular interactions by using coarser-graining techniques (R Some use Osetta. Up to the first approximation, the coarse-graining approach is protein Search for gaps in the quality structure (RosettaVIP) or perform fast local free energy calculations (f Either perform oldX, or find the residues that are evolutionary outliers (PROSS) Identify the problematic residue using one of the following methods. Then, hydrophobic packing or evolutionary con By returning to the census, better compatible residues are proposed. Subsequently, the stability of the protein The qualitative effects of these substitutions are estimated through energy simulations of the mutants. This process involves identifying residues, proposing substitutions, refolding, and freeing up the entire structure. Energy calculations can take anywhere from a few hours to several days.

[0007] Machine learning eliminates the need for prior knowledge or time-consuming manual work about the characteristics of specific proteins. It is an attractive alternative because it does not require inspection or the assignment of characteristics to individual structures. In recent years, Torng and Altman (Tor, incorporated herein by reference) ng et al.,“3D deep convolutional neural networks for amino acid environment simi larity analysis," BMC Bioinformatics, 18:3 02, 2017) describes a general framework for applying a three-dimensional convolutional neural network (3DCNN) to protein structure analysis by predicting the identity of amino acids provided with information about their surrounding protein microenvironments . This neural network achieves a prediction accuracy of 42% in the assignment of amino acids to wild-type sequences , which is superior to other computational methods that rely on the identification of pre-assigned structure-based features . Furthermore, given the structural data of the model protein T4 lysozyme , the 3D CNN typically predicts wild-type residues at locations where mutations are known to be destabilizing , and when provided with the structures of these known destabilizing mutants, it shows a strong preference for wild-type residues . SUMMARY OF THE INVENTION

[0008] Considering that the proteome must simultaneously exhibit several unrelated or even conflicting phenotypes such as folding shape, stability, catalysis, and binding specificity , it is reasonable that amino acids that are structural outliers at positions distant from the active site can affect folding and stability but do not affect function . Accordingly , there is a need in the art for improved protein engineering techniques that utilize artificial intelligence to learn the consensus microenvironments of different amino acids and scan the entire structure to identify residues that deviate from the structural consensus . These residues, which are considered to have a low probability of being wild-type, are thought to be loci of instability , and are thus targets for mutagenesis and stability engineering ​​​​It is a good candidate. Implementations of the systems and methods discussed herein are such modifications We provide advanced protein engineering technologies.

[0009] In one embodiment, training a neural network to improve the properties of a protein The computer implementation method involves collecting a set of amino acid sequences from a database, and Compiling a set of three-dimensional crystal structures that have a chemical environment for a set of no acids and , translating the chemical environment into a voxelized matrix, and sub-voxelized matrix Training a neural network in a set, and using a neural network to target a Identifying candidate residues that mutate in the protein, and using a neural network to identify candidate residues This includes identifying predicted amino acid residues to be substituted and producing mutant proteins. The protein shows improved properties compared to the target protein. In one embodiment, the method involves hydrogen Position, partial charge, beta factor, secondary structure, aromaticity, electron density, polarity, and combinations thereof The spatial arrangement of features selected from a group consisting of combinations is determined by a small number of the three-dimensional crystal structures. This further includes the step of adding to one.

[0010] In one embodiment, the method adjusts a set of amino acid sequences to reflect their intrinsic frequencies. This further includes the following: In one embodiment, the method involves taking amino acids from random positions in the sequence. This further includes sampling at least 50% of the amino acids in the set of acid sequences. In one embodiment, the method involves a second subset of a three-dimensional crystal structure or voxelized matrix. This involves training a second independent neural network and both neural networks This further includes identifying candidate residues and predicted residues based on the results of the network. In one embodiment, the properties are stability, maturity, folding, or a combination thereof. That is so.

[0011] In another aspect, a system for improving the properties of a protein comprises a processor and instructions It comprises a stored, non-temporary, computer-readable medium, and instructions are executed by the processor. When this is done, the step of providing a target protein containing the sequence of residues, and each three-dimensional model In contrast, it provides a set of three-dimensional models of the amino acids and a set of protein property values. The steps include: estimating a set of parameters at various points in each three-dimensional model; and estimating a set of parameters at various points in each three-dimensional model. And, with a three-dimensional model, parameters, and protein properties, the neural network The training steps involve a neural network to identify candidate residues that mutate in the target protein. The steps involve identifying candidate residues and using a neural network to predict the amino acid residues to be substituted. The steps include identifying a base and producing a mutant protein, and the mutant protein The protein shows improved properties compared to the target protein.

[0012] In one embodiment, the protein property is stability. In one embodiment, the step is Recompile at least one amino acid sequence of the folded amino acid sequence This includes generating an updated three-dimensional model. In one embodiment, the step is to reconvert Before piling, at least one amino acid sequence of the folded amino acid sequence, This includes adding the spatial arrangement of features.

[0013] In another aspect, the present invention relates to full-length wild-type secBFP2, T18, S28, Choose from Y96, S114, V124, T127, D151, N173, and R198. Includes a secBFP2 variant having one or more mutations in another residue. , relating to proteins. In one embodiment, the protein is represented by SEQ ID NOs: 2 to 28. It contains a secBFP2 variant that includes one amino acid sequence. In one embodiment, se The cBFP2 variant is a variant of one of the amino acid sequences from SEQ ID NOs: 2 to 28. Includes Ant. In one embodiment, the secBFP2 variant is sequence number 2 to sequence number 2. It contains a fusion protein containing one of the 8 amino acid sequences. In one embodiment, BFP is It contains a fragment of one amino acid sequence from sequence numbers 2 to 28.

[0014] In another aspect, the present invention encodes a protein containing the secBFP2 variant. This relates to nucleic acid molecules containing a cleotide sequence. In one embodiment, the nucleotide sequence is sequence number The amino acid sequences described in numbers 2 to 28, their variants, and their fusion proteins, or encodes a fragment thereof. In one embodiment, the molecule is a plasmid. Then, the molecule is an expression vector. In one embodiment, the nucleic acid molecule is a heterologous protein molecule. The present invention further includes multiple cloning sites for insertion of a sequence. In another embodiment, the present invention is A composition containing the above protein, a composition containing the above nucleic acid molecule, a composition containing the above protein The kit contains the nucleic acid molecules mentioned above. [Brief explanation of the drawing]

[0015] A patent or application file must include at least one drawing, which must be in color. A copy of this patent or the publication of the patent application, including the drawings (or drawings), is available for payment of the claim and any applicable fees. Provided by the office after payment.

[0016] The aforementioned objectives and features, as well as other objectives and features, are not permitted in this description and in understanding the present invention. Included to provide and evident from the following attached drawings which constitute part of this specification Therefore, in the drawing, similar numbers represent the same elements. [Figure 1A] This is a diagram illustrating the implementation of a computer-implemented neural network for improving the properties of synthetic proteins. [Figure 1B] This is a flowchart illustrating the implementation of a method for determining the amino acid residue at the center of the microenvironment. [Figure 1C] This is a flowchart for implementing a method to increase the properties of synthetic proteins during testing. [Figure 1D] This is a block diagram of a neural network implementation for increasing synthetic protein properties during training. [Figure 1E] This is a block diagram of an implementation of a convolutional neural network for improving the properties of synthetic proteins. [Figure 2A] This graph shows experimental results of the implementation of methods and systems for increasing the properties of synthetic proteins. [Figure 2B] This is another graph showing experimental results of the implementation of methods and systems for increasing the properties of synthetic proteins. [Figure 3A] This is another graph showing experimental results of the implementation of methods and systems for increasing the properties of synthetic proteins. [Figure 3B] This is a photograph of a protein synthesized using modifications suggested by the implementation of a system for enhancing the properties of the synthesized protein. [Figure 4A] This is another graph showing experimental results of the implementation of methods and systems for increasing the properties of synthetic proteins. [Figure 4B]This is a diagram of suggested protein modifications, indicated by the implementation of a system to enhance the properties of synthetic proteins. [Figure 5] This is a set of photographs showing experimental results from the implementation of a system to enhance the properties of synthetic proteins. [Figure 6] This graph shows the experimental results of implementing a system to enhance the properties of synthetic proteins. [Figure 7] This graph shows the experimental results of implementing a system to enhance the properties of synthetic proteins. [Figure 8] This graph shows the fluorescence magnification changes of 17 blue fluorescent protein variants relative to the wild-type protein. [Figure 9] This graph shows the change in fluorescence intensity of a blue fluorescent protein variant relative to the wild-type protein. [Figure 10] This provides exemplary images of the fluorescence of the blue fluorescent protein variant "Bluebonnet," including the S28A, S114T, N173H, and T127L mutations, compared to the parent protein and other blue fluorescent proteins. [Figure 11A] A block diagram showing the implementation of a system for increasing the properties of synthetic proteins. [Figure 11B] A block diagram showing the implementation of a system for increasing the properties of synthetic proteins. [Modes for carrying out the invention]

[0017] The drawings and description of this invention are simplified to illustrate elements relevant to a clear understanding of the invention. At the same time, to clarify, many of the related systems and methods are found in It should be understood that other elements are excluded. Those skilled in the art will understand when implementing the present invention. Recognizing that other elements and / or steps are desirable and / or necessary Obtain. However, such elements and steps are well known in the art. Furthermore, such elements and st No such considerations are provided herein. Disclosures herein are known to those skilled in the art. This covers all such changes and modifications to elements and methods.

[0018] Unless otherwise defined, all technical and scientific terms used herein are defined in this document. This specification has the same meaning as that generally understood by those skilled in the art to which the invention pertains. Any method or material similar or equivalent to those described in this document may be used in the practice or testing of the present invention. While these may be used in experiments, illustrative methods and materials are described.

[0019] When used herein, each of the following terms has the meaning associated with it in this section. It has.

[0020] In this specification, the articles "a" and "an" refer to one of the grammatical objects of an article, "T" is used to refer to two or more (i.e., at least one). For example, "Require "Element" refers to one or more elements.

[0021] When used herein to refer to measurable values ​​such as quantity or duration, "about" This refers to a variation of ±20%, ±10%, ±5%, ±1%, and ±0.1% from the specified value. It is intended to include dynamics, and therefore variability is appropriate.

[0022] The terms "nucleic acid molecule" or "polynucleotide" refer to single-stranded or double-stranded forms. This refers to a deoxyribonucleotide or ribonucleotide polymer in any of the following: Unless otherwise indicated, they function in a manner similar to naturally occurring nucleotides. It is possible to include polynucleotides containing known analogues of naturally occurring nucleotides. When a nucleic acid molecule is represented by a DNA sequence, this means that "U" (uridine) is "T It is understood that this also includes RNA molecules that have a corresponding RNA sequence that replaces (thymidine). It will likely happen.

[0023] The term "recombinant nucleic acid molecule" refers to a molecule containing two or more linked polynucleotide sequences. This refers to nucleic acid molecules that do not exist in nature. Recombinant nucleic acid molecules are those created using recombination methods, particularly genetic engineering techniques. Recombinant nucleic acids may be produced by or by chemical synthesis. The molecules discussed herein are fused to fusion proteins, for example, the polypeptide of interest. The systems and methods used to encode fluorescent protein variants suggested by This is possible. The term "recombinant host cell" refers to a cell that contains recombinant nucleic acid molecules. Therefore, recombinant host cells contain "genes" that are not found in the cell's native (non-recombinant) form. Polypeptides can be expressed from this.

[0024] References to polynucleotides that "encode" polypeptides refer to the transcription of polynucleotides. And this means that polypeptides are produced during the translation of the mRNA produced therefrom. A coding polynucleotide is a coding polynucleotide whose nucleotide sequence is identical to that of mRNA. Such coding polynucleotides are considered to include both the chain and its complementary chain. It is recognized that it is considered to contain a degenerate nucleotide sequence encoding the same amino acid residue. It will be done. The nucleotide sequence encoding the polypeptide contains introns. It may include polynucleotides and coding exons.

[0025] The term "expression regulatory sequence" refers to the transcription or translation of a polynucleotide, or its This refers to a nucleotide sequence that regulates the localization of operablely linked polypeptides. Your sequence is an expression control sequence that transcribes nucleotide sequences and, if necessary, translates them (i.e., Identifying cells that encode transcriptional or translational regulatory elements (or polypeptides, respectively), or polypeptides. When controlling or adjusting the localization to a section, it is "operably connected". Therefore The expression regulatory sequences are promoters, enhancers, transcription terminators, and start codons. ATG), intron resection and splicing for maintaining the correct leading frame A signaling signal, stop codon, ribosome binding site, or polypeptide is targeted at a specific location. Targeting sequences, for example, cellular compartmentalization signals (which involve polypeptides in the cytosol and nucleus). , plasma membrane, endoplasmic reticulum, mitochondrial membrane or matrix, chloroplast membrane or chloroplast It can target the lumen, intermediate trans-Golgi flat sac, lysosome, or endosome. It is possible that.... As a cell compartmentalization domain, for example, human type II membrane anchor protein... Amino acid residues 1-81 of chlorogalactosyltransferase, or cytochrome c oxy Peptides containing amino acid residues 1-12 of the presequence of sidase subunit IV are listed. (Also, Hancock et al., EMBO J.10:4033-40) 39,1991, Buss et al.,Mol.Cell.Biol.8:3960 See also U.S. Patent Nos. 3963,1988 and 5,776,689 (each of which is (and incorporated herein by reference).

[0026] When used to describe chimeric proteins, "operably linked" or This is a term that means "operationally coupled" or "operably connected," or similar. These refer to polypeptide sequences that are in physical and functional relationships with one another. In one embodiment, the function of the polypeptide components of the chimeric molecule is compared to the functional activity of the individual components. No change in comparison. For example, as suggested by the systems and methods discussed herein. The fluorescent protein can be fused to the target polypeptide. In this case, the fusion molecule It retains its fluorescence, and the target polypeptide retains its original biological activity. Preferred. In some embodiments of the systems and methods considered herein, a fluorescent The activity of either the protein or the target protein is compared to their activity alone. Such fusion can be reduced. Furthermore, such fusion is a system and method considered herein. It can be used together with

[0027] The term "labeling" can refer to, for example, visual inspection, spectroscopy, or photochemical or biochemical reactions. A composition that can be detected by an immunochemical reaction or a chemical reaction, with or without the use of equipment. To point to. Useful labels include, for example, phosphorus-32, fluorescent dyes, fluorescent proteins, and highly electron-tight. Reagents, enzymes (such as those commonly used in ELISA), small molecules, for example, biotin Antiserum or antibodies that may be digoxigenin or monoclonal antibodies are available. Other haptens or peptides may be mentioned. Systems and methods discussed herein. The fluorescent protein variants suggested by the implementation are proteins that are themselves detectable. It is a quality, but nevertheless, by means other than its own fluorescence, for example, radioactive By incorporating a nuclide label or peptide tag into a protein, for example, in the expression of that protein Detectable by facilitating the isolation of specific and expressed proteins within the sample. It will be recognized that the system discussed herein can be labeled in such a way. Labels useful for the implementation of the method generally include radioactive signals, fluorescent light, and enzyme activity. It generates measurable signals such as, for example, fluorescence in the sample. It can be used to quantify the amount of protein variants.

[0028] The term "polypeptide" or "protein" refers to a polypeptide of two or more amino acid residues. These terms refer to amino acids, which are naturally occurring amino acids, where one or more amino acid residues are present. Amino acid polymers, which are artificial chemical analogs of amino acids, as well as naturally occurring amino acid polymers. The term "recombinant protein" applies to proteins derived from recombinant DNA molecules. Proteins produced by the expression of nucleotide sequences that encode the amino acid sequence of a protein. It refers to.

[0029] The terms "isolated" or "purified" refer to substances in their natural state as they exist in nature. Refers to a substance that substantially or essentially does not contain its normally associated components. Purity or homogeneity. The properties are generally determined by methods such as polyacrylamide gel electrophoresis and high-performance liquid chromatography. It is determined using analytical chemistry techniques. The polynucleotide or polypeptide is then prepared. A species is considered isolated if it is the dominant species present in a substance. Generally speaking, an isolated tan Proteins or nucleic acid molecules represent more than 80% of the high-molecular-weight species present in the preparation, and in many cases, This represents over 90% of all existing polymer species, and usually over 95% of polymer species, and especially, When tested using conventional methods for determining the purity of such molecules, they are detected. A polypeptide or polynucleotide purified to the point of essential homogeneity, such as being the only species of its kind. It's Ochido.

[0030] The term "naturally occurring" refers to proteins, nucleic acid molecules, cells, or occurrences in nature. It is used to refer to other substances. For example, polypeptides present in living organisms, including viruses. A nucleotide or polynucleotide sequence. Naturally occurring substances are like those found in nature. It can be a form, and for example, it can be modified by human hands to be an isolated form.

[0031] The term "antibody" refers to an immunoglobulin gene(s) or its antigen-binding fragment. Therefore, it refers to polypeptides that are essentially encoded, and they are specific to the analyte (antigen). It binds and recognizes it. Examples of recognized immunoglobulin genes include kappa, lambda, Alpha, gamma, delta, epsilon, and muon constant region genes, as well as countless others Examples include immunoglobulin variable region genes. Antibodies exist as complete immunoglobulins. Similarly, the antigen-binding fragments of the antibodies are characterized in a way that allows them to be digested by peptidases. Therefore, antibodies can be produced, or recombinant DNA methods can be used. Examples of antigen-binding fragments include Fv, Fab', and F(ab)'2 fragments. As used herein, the term "antibody" refers to the entire antibody produced by modification. Either an antibody fragment that is synthesized using recombinant DNA, or an antibody fragment that is de novo synthesized using recombinant DNA. The term "immunoassay" refers to a method that uses antibodies to specifically bind to an analyte. This refers to (i). Immunoassay is a method that uses the specific binding properties of a particular antibody to isolate and target an analyte. It is characterized by chemicalization and / or quantification.

[0032] Used in relation to two or more polynucleotide sequences or two or more polypeptide sequences. In this case, the term "identical" means that they are the same when aligned for the greatest match. It refers to residues in a sequence. When the degree of sequence identity is used in relation to polypeptides, Otherwise, the positions of one or more residues that are not identical can differ due to conservative amino acid substitutions. The first amino acid residue has similar charge or similar chemical properties such as hydrophobic or hydrophilic properties. Substituted in place of another amino acid residue having the desired properties, and therefore the function of the polypeptide. It is recognized that the properties are not changed. When the polypeptide sequence differs by a conservative substitution, The sequence identity percentage can be adjusted upward to compensate for the conservative nature of substitutions. Such adjustments, for example, can result in a partial mismatch rather than a complete mismatch for a conservative substitution. This is done by scoring as a "C" and thereby increasing the proportion of sequence identity. Therefore, for example, the same amino acid may be given a score of 1, and non-conservative substitutions may occur. However, if a score of zero is given, a conservative permutation is given a score between zero and 1. The scoring of conservative substitutions is, for example, by Meyers and Miller, Co. mp.Appl.Biol.Sci.4:11-17,1988, Smith and Waterman, Adv. Appl. Math. 2:482, 1981, Needle man and Wunsch, J. Mol. Biol. 48:443, 1970, Pe. arson and Lipman,Proc.Natl.Acad.Sci.,USA 85:2444 (1988), Higgins and Sharp, Gene 73 :237-244, 1988, Higgins and Sharp, CABIOS 5 :151-153;1989, Corpet et al., Nucl. Acids R es.16:10881-10890,1988, Huang,et al.,Comp .Appl.Biol.Sci.8:155-165,1992, Pearson et. al., Meth. Mol. Biol., 24:307-331, 1994 (These Each is calculated using the algorithms discussed (which are incorporated herein by reference). Alignment can also be done by simple visual inspection and manual alignment of the arrangement. It can be implemented.

[0033] When used in relation to a specific polynucleotide sequence, it is referred to as a "conservatively modified mutation." The term refers to different polynucleotides that encode the same or essentially the same amino acid sequence. Ocside sequences, or polynucleotides, refer to amino acid sequences that are essentially identical. It does not code for sequences. Due to the degeneracy of the gene code, a large number of functionally identical polynucleotides However, it encodes any given polypeptide. For example, codons CGU, CGC, CGA, CGG, AGA, and AGG all encode the amino acid arginine. Therefore At any position where arginine is specified by the codon, the codon is encoded poly The peptide can be modified to any of the corresponding codons described without altering the peptide itself. Nucleotide sequence mutations like these are called "silent mutations," and this is because they are "conservatively modified." It can be considered a species of "mutation." Therefore, it encodes a fluorescent protein variant. Each polynucleotide sequence disclosed herein as such is a possible cy It will also be recognized that this explains the Rent mutation. Furthermore, it is the only codon of normal methionine. AUG is the only codon of tryptophan, and UUG is the only codon of tryptophan, but polynucleates Each codon in the rheotide is modified using standard techniques to obtain a functionally identical molecule. It will also be recognized that this is possible. Therefore, the sequence of the coded polypeptide may change. Each silent mutation in a polynucleotide that does not cause this mutation is implicitly described herein. This refers to a single amino acid or a small percentage of amino acids (typically 5%) in the encoded sequence. Each substitution, deletion, or addition alters, adds, or deletes (generally less than 1%) This can be considered a conservatively modified mutation, provided that the modification is in a chemically similar amino acid. It will be recognized that this is conditional on the resulting amino acid substitution. Functionally similar Conservative amino acid substitutions that provide amino acids may include the following six groups, and these Each contains an amino acid that is considered a conservative substitution for the other: 1) Alanine (Ala, A), Serine (Ser, S), Threonine (Thr, T), 2) Aspartic acid (Asp, D), glutamic acid (Glu, E), 3) Asparagine (Asn, N), glutamine (Gln, Q), 4) Arginine (Arg, R), Lysine (Lys, K), 5) Isoleucine (Ile, I), leucine (Leu, L), methionine (Met, M) ), valine (Val, V), and 6) Phenylalanine (Phe, F), tyrosine (Tyr, Y), tryptophan (T rp, W).

[0034] Amino acid sequences or nucleotide sequences can be compared with each other or across a given comparison window. If a reference sequence shares at least 80% sequence identity, then two or more amino acid sequences Or two or more nucleotide sequences are "substantially identical" or "substantially similar". Therefore, substantially similar sequences are, for example, sequences that are at least 85% similar. Identity, at least 90% sequence identity, at least 95% sequence identity, or less Both contain sequences with 99% sequence identity.

[0035] If the complement of the target nucleotide sequence is substantially identical to that of the reference nucleotide sequence, The elephant nucleotide sequence is considered to be "substantially complementary" to the reference nucleotide sequence. It will be done.

[0036] Fluorescent molecules exhibit fluorescence resonance energy transfer (FR) with donor and acceptor molecules. Useful in ET. Efficiency of FRET between donor molecules and acceptor molecules and To optimize detectability, several factors need to be balanced. Donor The emission spectrum can be combined with the acceptor's excitation spectrum to maximize the overlap integral of the emission spectrum. It should overlap as much as possible. Also, the quantum yield of the donor part and the absorbance of the acceptor part The number represents the distance over which the energy transfer efficiency is 50%, and in order to maximize RO, as much as possible. It should be higher. However, fluorescence resulting from direct excitation of the acceptor is FR Because it can be difficult to distinguish from fluorescence produced by ET, donors and acceptors The excitation spectrum of the acceptor shows that the donor efficiently excited the acceptor without directly exciting it. To find possible wavelength ranges, there should be as little overlap as possible. Similarly, the emission spectra of the donor and acceptor may show two distinct emission levels. Therefore, it should be as redundant as possible. The light emitted from the acceptor is the only reading. If it is to be measured either as an outtake or as part of the emission ratio A high fluorescence quantum yield in the acceptor region is desirable. Select a donor and acceptor pair. One factor to consider when doing so is the efficiency of fluorescence resonance energy transfer between them. Preferably, the FRET efficiency between the donor and acceptor is at least 10%. More preferably, it is at least 50%, and even more preferably, at least 80%.

[0037] The term "fluorescence properties" refers to the molar extinction coefficient at the appropriate excitation wavelength, fluorescence quantum efficiency, and excitation Shape of the excitation spectrum or emission spectrum, maximum excitation wavelength and maximum emission wavelength, 2 Ratio of excitation amplitude at different wavelengths, ratio of emission amplitude at two different wavelengths, excitation This refers to the lifetime of the activated state, or fluorescence anisotropy. It also refers to spectral variation between the wild-type and parent fluorescent protein. A measurable difference in any one of these characteristics between the ant or its variants is useful. Yes, there is a measurable difference in the amount of any quantitative fluorescence characteristic, for example, fluorescence at a particular wavelength. This can be determined by determining the quantity or the integral of fluorescence across the emission spectrum. Determining the ratio of excitation amplitude or emission amplitude at different wavelengths (respectively, "excitation"). The "amplitude ratio calculation" and "emission amplitude ratio calculation" are particularly advantageous, as this is a ratio calculation process. The system provides an internal reference, determining the absolute brightness of the excitation source, the sensitivity of the detector, and the light scattering by the sample. This is to offset fluctuations in disturbance or quenching. When used herein, The term "fluorescent protein" refers to a protein whose fluorescence is due to chemical tagging. Proteins that have been exposed to ultraviolet light, and their emission peaks at ultraviolet wavelengths (i.e., less than approximately 400 nm) However, for the purpose of implementing the systems and methods discussed herein, it is considered a fluorescent protein. It fluoresces only in the presence of certain amino acids, such as tryptophan or tyrosine. Except for polypeptides, these polypeptides can emit fluorescence when excited by appropriate electromagnetic radiation. It refers to the protein of choice. Generally, it is used to prepare the composition of the implementation of the system discussed herein. Fluorescent proteins useful for, or for use in, the implementation of the methods discussed herein. Fluorescent proteins are proteins that obtain fluorescence by autocatalytically forming chromophores. Proteins are either naturally occurring or manipulated (i.e., variants or (Mutant) May contain amino acid sequences. When used in relation to fluorescent proteins, "mutant" The terms "variant" or "body" refer to a protein that is different from the reference protein.

[0038] The term "blue fluorescent protein" originally refers to a protein that emits blue fluorescence. It is widely used in detailed documents. The term "blue fluorescent protein" or "BFP" is used in Used in the broadest sense, especially from mTagBFP, secBFP2, and any species. The blue fluorescent protein, as well as its variants (their ability to emit blue fluorescence) (As long as you keep it)

[0039] The terms "mutant" or "variant" refer to the corresponding wild-type or parent fluorescent protein. This is used herein in relation to fluorescent proteins containing mutations in quality. This shows a mutant fluorescent protein that has different fluorescence properties compared to the corresponding wild-type fluorescent protein. To do this, we use "spectral variants" or "spectral mutants" of fluorescent proteins. And, as referred to herein.

[0040] Throughout this disclosure, various aspects of the implementation of the systems and methods discussed herein are described. It may be presented in range format. The range format is used simply for convenience and conciseness. Therefore, it should not be interpreted as an inflexible limitation on the scope of the present invention. It should be interpreted as: Therefore, the description of the scope should include all possible sub-scopes, Furthermore, it should be considered that the individual numerical values ​​within that range have been specifically disclosed. For example, 1~ The ranges indicated, such as 6, are partial ranges such as 1-3, 1-4, 1-5, 2-4, 2-6, 3-6, etc. The range, and the individual numbers within that range, for example, 1, 2, 2.7, 3, 4, 5, 5.3, 6 , and any whole and partial increments between them shall be deemed to have been specifically disclosed. This applies regardless of the scope of the matter.

[0041] In some aspects of the systems and methods discussed herein, the following are provided herein: The software that executes the instructions may be stored on a non-temporary computer-readable medium. The software, when executed on the processor, implements the methods considered herein. Perform some or all of the steps.

[0042] The systems and methods described herein are performed using computer software. Regarding the algorithms used. Specific embodiments are written in a specific programming language. things, or specific operating systems or computing platforms It may be described as being executed on a form, but the system considered herein The implementation of this method can be any particular computing language, platform, or other It is understood that the algorithms described herein are not limited to these combinations. The software that runs on it is C, C++, C#, Objective-C, Java, J AvaScript, Python, PHP, Perl, Ruby, or Visual Basic Write and compile in any programming language, including but not limited to these. , or may be interpreted as such. The elements of the systems and methods considered herein are servers Cloud instances, workstations, thin clients, mobile devices, Embedded microcontrollers, televisions, or any other suitable computing Any acceptable computing platform, including but not limited to devices. This provides a better understanding of what can be done on the form.

[0043] Some of the system implementations discussed herein are executed on computing devices. The software described herein is described as such. The software described herein is one particular On a fixed computing device (e.g., a dedicated server or workstation) It may be disclosed as working software, but the software may also be inherently portable. Furthermore, software that runs on a dedicated server can also be used on desktop or mobile devices. Chairs, laptops, tablets, smartphones, watches, and wearable electronic devices are also included. or other wireless digital / mobile phones, televisions, cloud instances, embedded systems Microcontroller, thin client device, or any other suitable computer The system discussed herein includes any of the wide range of devices, including tinkling devices. It may be performed for the purpose of implementing the calling method.

[0044] Similarly, some implementations of the systems considered herein involve various wireless or wired connections. It is described as communicating over a computer network. The systems considered in this specification For the purpose of implementing the method and the network, the terms "network," "networking," and "net" are used. The term "working" refers to wired Ethernet, fiber optic connections, and various 802.11 standards. Wireless connectivity including any of the following: 3G, 4G / LTE, or 5G networks. Ruler WAN infrastructure, Bluetooth®, Bluetooth oth(registered trademark) Low Energy (BLE), or Zigbee(registered trademark) ) A communication link, or the ability of one electronic device to communicate with another electronic device. It is understood to encompass other methods of intent. In some embodiments, these are considered herein. The networking component of the system implementation is a virtual private network (VPN). It may be implemented on )

[0045] The implementations of the systems and methods discussed herein include machine learning algorithms, and This relates to machine learning engines, or neural networks. Neural networks are... Based on various attributes of proteins, such as the atomic environment of amino acids within known proteins It may be trained, and based on its attributes, it proposes one or more amino acids in a protein. The output may include the changes. In some embodiments, the attributes are atomic type, electrostatic, and base. This may include factors, solvent contactability, secondary structure, aromaticity, or polarity. Then, the obtained Mino acids may be judged according to one or more quality indicators, and the weight of the attributes shall be such that the quality indicator is most important. It may be optimized for scaling. In this way, the neural network is experimental. It can be trained to predict and optimize any quality metric that can be measured objectively. Examples of quality metrics that a network can be trained on include the accuracy of wild-type amino acids and known safety standards. Statistical / destabilization position, accuracy of amino acid groups, and any other suitable types that can be measured. Quality metrics can be cited. In some embodiments, the neural network is multi-ta It has a scaling function and can enable simultaneous prediction and optimization of multiple quality indicators.

[0046] In embodiments that implement such a neural network, queries are expressed in various ways. This can be done. The query can be performed using desired parameters, for example, thermally through the melting curve, or Enhance protein stability that can be chemically embodied using guanidine or urea denaturation. Therefore, we ask the neural network to identify the amino acids within a given protein. It is possible. The neural networks of the implementations of the systems and methods considered herein are That predicted identity (evaluated by a neural network) is the same as its natural identity. This identifies one or more amino acid residues of different proteins, thereby improving the protein Proteins can be produced by mutating native amino acid residues to predicted amino acid residues. This may indicate that the predicted amino acid residues are any natural or Non-natural (e.g., artificial or synthetic) amino acids are also acceptable.

[0047] In some embodiments, the neural network receives an input amino acid sequence or The neural network is trained using the desired parameter values ​​associated with the residues. It may also be updated by this. Updating the neural network in this way will result in the optimal The ability of neural networks to propose amino acid residues can be improved. Several implementations Morphologically, training a neural network involves predicting mutations in amino acid residues. This may include using values ​​for desired parameters related to the protein. For example, how many In that embodiment, training the neural network is done at the proposed amino acid location. Predicting the value of a desired parameter and comparing the predicted value to a parameter related to a known amino acid. This involves comparing the value to a corresponding value and training the neural network based on the results of the comparison. This may include the fact that if the predicted value is the same as or substantially similar to a known value, then The overall network may be updated minimally, or not updated at all. If the predicted value differs from the known amino acid value, the neural network will... It can be substantially updated to correct it better. How neural networks are retrained Regardless of whether it is refined or not, using a retrained neural network, additional amino acids An acid could be suggested.

[0048] The technology described in this application relates to enhancing protein stability, but this is not related to other technologies. Protein parameters or attributes, such as half-life, activity, degradation resistance, solubility, and thermal stability. Qualitative analysis, post-translational modification, enhanced pH tolerance, reduced maturation time, nucleic acid binding, protein-protein interactions Because these technologies can be applied to hydrophobic or hydrophobic properties, or combinations thereof, the applications of these technologies are not limited. It should be understood that it is used for training neural networks. Depending on the type of data being processed, the neural network uses different types of proteins. This can be optimized for protein-protein interactions and / or protein attributes. In this way, the neural network is trained to analyze proteins, also known as peptides. This can improve the identification of possible amino acid sequences. Queries can be performed on a neural network. Performing this may involve inputting an initial amino acid sequence to the protein. The network may have been previously trained using different amino acid sequences. Queries to the network are for proteins with higher stability than the initial amino acid sequence. This may also relate to the proposed amino acid sequence. Each residue of the proposed amino acid sequence The proposed amino acid sequence that shows a specific amino acid for [the given amino acid] was obtained from a neural network. It is acceptable to receive it.

[0049] Inputting an array with discrete representations, or a neural network with continuous representations. Receiving the output from and continuously inputting it into the neural network By discretizing the output before providing it, the neural network can iteratively process questions. The techniques described herein related to performing the reprocessing may be applied to other machine learning applications. Such techniques may be particularly useful in applications where a final output having a discrete representation is desirable. Such techniques use data to associate discrete attributes with the characteristics of a set of discrete attributes. By applying a model generated by a trained neural network, It can be generalized to identify a set of discrete attributes. In the context of identifying amino acids in a sequence... Furthermore, the discrete properties may include different amino acids.

[0050] In some embodiments, the model includes data derived from molecular simulations. A set of initial data, each having discrete attributes located at a given position, is used as input. It may be received. Each of the discrete attributes in the initial set is one of several discrete attributes. Querying a neural network involves inputting an initial set of discrete attributes. And, a set of discrete outputs having different levels of characteristics than the initial set of characteristics. This may also include generating attributes. It may also involve querying a neural network. In response, regarding the output sequence and the different discrete attributes for each position in the output sequence The sequential values ​​may be received from a neural network. For each position in the sequence, The value of a discrete attribute is a new value related to the level of the characteristic when the discrete attribute is selected for position. It can correspond to the prediction of a linear network and form a dataset of continuous values. The values ​​are position It may extend across discrete attributes of the output, identifying a set of discrete versions. It may be used when doing so. In some embodiments, a series of discrete versions are output. Identifying each position in the series is done by looking at the different discrete attribute values ​​for the position. This may include selecting the discrete attribute with the highest value. The gender may be received as output that identifies the discrete version.

[0051] In some embodiments, the iterative process is performed on the output sequence of neural networks. Querying the work, receiving the output sequence, and the output sequence It is formed by identifying scattered versions. Additional iterations of the iterative process are the previous ones. This may include inputting a series of discrete versions output from an iteration. S occurs when the current sequence of output matches the previous sequence of output from the previous iteration. It may be stopped at that point.

[0052] In some embodiments, to identify amino acid sequences having multiple quality indicators, A desired value for a single quality metric, including one for training a multi-network. Rather, it has desired values ​​for multiple quality indicators (for example, values ​​higher than the values ​​in another array). The proposed amino acid sequence is identified. Such techniques can produce different types of amino acids. Identifying the proposed amino acid sequence of a protein can be particularly useful for its desired applications. In the implementation of such technologies, training data is used to train the neural network. The data may include information relating to different characteristics of each of the amino acid sequences being analyzed. The models generated by training a multi-network have different combinations of characteristics. It may have one or more parameters corresponding to it. In some embodiments, the parameters Ta may represent the weight between the first and second features, which is the proposed amino The acid sequence is used to balance the possibility of having the first characteristic compared to the second characteristic. It may be done in some embodiments. In some embodiments, training a neural network is different This includes assigning a score to a particular characteristic, and the score is determined by the proposed amino acid sequence. It can be used to estimate values ​​for the parameters of the model used for prediction. In some such embodiments, the training data is related to the atoms in the amino acid sequence. It may include a microenvironment, which is used to train neural networks. In this case, a model is generated that is used to predict the proposed amino acid sequence. Training a network can involve assigning scores and parameters. The value can be estimated using the score.

[0053] Biological applications of convolutional neural networks are relatively rare. Proteins are not analyzed as amino acid sequences, but rather their three-dimensional structure is determined by... It is being evaluated in its crystallization form. One embodiment of the implementation of the method discussed herein is 20 A three-dimensional convolutional neural network characterizes the unique chemical environment of each amino acid. This involves training the neural network to best fit a given environment. It is possible to predict the amino acids that will be produced. The neural network described herein is , 1.6 million amino acid environments across 19,000 phylogenetically distant protein structures It is trained with [a specific model]. After training, the within-sample accuracy of this network is 80.0%, and [the network is not specified in the original text]. The simple external accuracy is 72.5%, which is an improvement of approximately 20-30% (approximately 40%) compared to the current level of technology. (Out-of-sample accuracy)

[0054] Sites where there is a large discrepancy between the expected amino acids and the observed amino acids indicate stability. This also shows targets for manipulating protein characteristics such as folding and maturation. The systems and methods described in the document are three biological examples, beta-lactama - Antibiotic markers, coral-derived blue fluorescent protein, and yeast Candida We experimentally characterized phosphomannose isomerase derived from Albicans, and neural Predictions from the network demonstrate improved protein function and stability in vivo. These results predict new biological tools at the intersection of AI and molecular biology. .

[0055] In one embodiment, the implementation of the method discussed herein is a neural network, for example However, the neural network published by Torng and Altman, which is referenced above, Network implementations are utilized. Implementations of the systems and methods discussed herein are as follows: As the experimental results discussed below show, the publicly available neural network design is essentially This improves upon the original Torng and Altman set, which had 32,760 training models. 3696 training protein families yielding structures and 1601 test structures. It contains 194 test protein families.

[0056] The implementation of the systems and methods discussed herein addresses the problem of protein stabilization. To achieve this, the Torng and Altman framework is used as the basis. In the example, the crystalline structure of a protein is treated as a three-dimensional image. There are many observations about individual amino acids and their atomic environments. Some methods suggest that We focus a consistent reference frame on one of those amino acids. From this advantageous position... Oxygen, nitrogen, sulfur, and carbon atoms in a 20x20x20 angstrom box. The molecules are separated, and all atoms associated with the central amino acid are removed. Then the environment and This set of environmentally compatible amino acids is for three-dimensional convolutional neural networks. This may be used as an annotated training set for this trained neural network. Therefore, experimentally introduced destabilizing mutations can be detected.

[0057] The implementation of the systems and methods discussed herein is intended to identify novel stabilizing mutations. The underlying model is improved. The improvements described herein improve the quality of predictions, addressing known instabilities. In addition to justifying mutations, it identifies unknown destabilizing residues and suggests stabilizing mutations. To make it into a sufficient state.

[0058] In some implementations, the systems and methods discussed herein involve input tampers This enables the identification of wild-type amino acids located in a favorable environment on the protein. Such implementations This can narrow the sequence space of residues with a very low wild-type probability. Compared to the current level of technology The improvements provided by the implementation of the systems and methods discussed herein are combined When this is done, it is a significant step in identifying candidate protein residues for improving overall utility. These can be described as several individual improvements that form an improved model.

[0059] Figure 1A shows a computer-implemented neural network for improving synthetic protein properties. This is a diagram of the work implementation. Some of the properties of the protein that the engineer wants to modify are: Maturation kinetics, thermal stability, K m、 K cat , cation or for proper folding These are anion dependence and pH tolerance. In 101, the protein is a protein Each residue in the protein may be translated into the microenvironment, and the three-dimensional model of the protein and A microenvironment is generated. Other methods for generating a three-dimensional model include... There are other methods as well, but unknown protein models are candidate fragments taken from known protein structures. When a fragment set is constructed from a pool of known protein segments, the amino acid sequence is determined. Segment matching if there is a match, or a known protein model is selected ("Template" (Alignment) The residues in the amino acid sequence are mapped to residues in the template sequence (alignment (The constraints on various distances, angles, and dihedral angles within the array are applied to the template structure.) A comparison based on the achievement of spatial constraints when constraint violations are minimized, derived from the elements. Protein modeling is one example. When a three-dimensional model of the protein crystal structure is generated... This generates a corresponding microenvironment related to the structure.

[0060] In some embodiments, the three-dimensional model simply illustrates the protein without a microenvironment. It may be displayed. In some implementations, the three-dimensional model is mapped to a three-dimensional array. This may also be done. For example, the coordinates of a three-dimensional model are stored in a three-dimensional array. In this embodiment, the three-dimensional image may be generated from a three-dimensional model, and the three-dimensional image is a cubic The original array may be mapped. The image data in the array is a voxelized matrix. It can be called that. A pixel can represent an addressable element of an image in two-dimensional space. In other words, a voxel represents an addressable element in three-dimensional space.

[0061] In some implementations, image features are processed through a three-dimensional convolutional layer and a maximum pooling layer. It may be extracted in this way. The three-dimensional filter in the three-dimensional convolutional layer contains 20 amino acid microparticles. To isolate a subenvironment, we examine the repeating spatial patterns that best capture local biochemical features. Search. The maximum pooling layer performs downsampling to the input and the network parallel To improve progressive invariance. Regarding convolutional neural network architectures, see below. Further considerations are made.

[0062] The first convolutional layer 121 detects low-level features via a filter. A neural network uses convolution to enhance the features of a dataset. In the convolutional layer of a neural network, a filter is applied to a three-dimensional array. , generate a feature map. In the convolutional layer, the filter is the input and the filter element. Each dot is slid across the stack, and the input is stored as a feature map. Several implementations In this case, a 3x3x3 filter may be applied to a three-dimensional image.

[0063] The convolutional filter and feature maps from the image are shown by 102. In this embodiment, the reference frame may be created around the central amino acid in the image. Features may be extracted around the central amino acid. Image and filter convolution The feature map created from this summarizes the presence of filter-specific features in the image. Increasing the number of filters used increases the number of features that can be tracked. Then, 100 filters were applied to create an 18x18x18 feature map. In the setup, other numbers of filters may be used. The resulting feature map is then used to analyze the features. To explain nonlinear patterns, an activation function may be passed through the data.

[0064] In some implementations, a normalized linear function with the equation f(x) = max(0,x) is activated. It may also be applied to the feature map as a transformation function. The normalized linear activation function is for positive values It behaves linearly, making this function easy to optimize, and then the neural network This enables the achievement of high prediction accuracy. Furthermore, the normalized linear activation function can handle any negative case. It outputs zero for the input, meaning it is not a true linear function. Therefore, The output of a convolutional layer in a convolutional neural network is a feature map, and the feature map The values ​​within the function can pass through a normalized linear activation function.

[0065] A second convolutional layer is exemplified in 122. Increasing the number of convolutional layers can lead to tracking The complexity of the features can increase. The convolutional layer in 122 is used to track the features. Incorporate 100 filters. In some embodiments, the filters track the features To ensure accuracy, it is the same as in the first convolutional layer. Alternative implementation In this configuration, different filters may be incorporated into the second convolutional layer. In this state, atoms related to the central amino acid may be removed via a filter.

[0066] In some implementations, a smaller dataset of dimensions 16x16x16 is used by 10³ (Other implementations may use other dimensions, or more or fewer dimensions) (A numerical filter is applied). The dot product of the convolutions in the second convolutional layer is the data Reduce the size of the set. Dataset 103 is a composite of the original protein image 101. Includes a feature map that tracks rough characteristics.

[0067] In some implementations, the first pooling layer with dimensions 2x2x2 is implemented as 123. Good. A pooling layer may be implemented to downsample the data. A ring window may be applied to the feature map. In some embodiments, a pulley The ng layer outputs the maximum value of the data within the window, and the data within the window is downscaled. Pooling is performed. Maximum pooling highlights the most prominent features in the pooling window. In another embodiment, the pooling layer outputs the average value of the data within the window.

[0068] The downsampled data in 104 consists of 200 independent 8x8x8 arrays. Represents a ray. By downsampling the data, the neural network is involved. It becomes possible to store linked information. Having a large amount of data will be discussed further below. This is advantageous in that it allows the network to fine-tune the accuracy of its weights. It's possible, but with large amounts of data, neural networks would require a considerable amount of processing time. There is a possibility that downsampling the data may be necessary for the network. This can be important in neural networks to reduce computer computation. Dimension 2 2x2 pooling layer 123 and downsampled data with dimensions 8x8x8 As shown with the, other implementations have pooling windows of other sizes and down Sampled data may be used.

[0069] In some implementations, the subsequent convolutional layer 124 consists of 200 independent 2x2x2 particles. Using Luta, the downsampled data is reprocessed, and the features in the new feature map are... Emphasize the features. In contrast to 3x3x3, the smaller filter 2x2x2 is used by Dow To explain the unsampled data, a convolutional layer is implemented at 124. The depth of the convolution filter depends on the depth of the data in order to perform the dot product matrix multiplication well. It should be the same as above. Other implementations may have different sizes or dimensions as discussed above. Legal filters may be used.

[0070] The feature maps from the convolutional layer 124 and the image are shown in 105. Downsample The feature map created from the convolution of the filtered data and filters shows the features in the image. This summarizes the existence of Ruta-specific features. In the implementation exemplified in 105, there are 200 independent 7 We have a 7x7 array. The dot product from convolution further reduces the data size.

[0071] The convolutional layer 125, as shown in the figure, processes 400 images from the low-resolution dataset 105. By using an independent 2x2x2 filter, for example, additional filters can be used. More complex features can be extracted. Increasing the number of filters applied to the image allows tracking to be performed. The number of features obtained increases. This data is downsampled from pool layer 123. Because the size is effectively reduced, more filters can be applied to this convolutional layer. Therefore, without requiring massive processing or memory requirements, image features of protein 101 can be extracted. It can be used to emphasize or call something.

[0072] The feature map from convolutional layer 125 is shown in 106. Downsampled The feature map created from the convolution of the data and filters is specific to the filter in the image. Summarize the existence of the feature. In the implementation exemplified in 106, 400 independent 6x6x6 elements While there are rays, in various implementations, arrays of other numbers or sizes may be used. Convolution The dot product from the combined data further reduces the data size.

[0073] Some implementations have dimensions of 2x2x2 (or any other suitable size). A second pooling layer is implemented at 126 to further downsample the data. In some embodiments, the same type of pooling layer is implemented in the first pooling layer. It may be implemented in a second pooling layer as shown. Depending on the type of pooling layer The pooling window used to downsample the data is determined. For example, the maximum pooling layer may be implemented in 123 and 126. Other implementations In this case, different pooling layers may be implemented in the convolutional neural network. For example, the maximum pooling layer may be implemented in 123, and the average pooling layer may be... It may be implemented in 126. The maximum pooling layer is the most prominent in the pooling window. While highlighting distinctive features, the mean pooling layer outputs the average value of the data within the window. ru.

[0074] In the example implementation, the downsampled data at 107 is 400 This represents an independent 3x3x3 array, but arrays of other numbers or dimensions may be used. Having a large amount of data means that the network's weights While this may be advantageous in that it allows for fine-tuning of accuracy, with large amounts of data, A multi-network can consume a considerable amount of processing time. Downsample the data. Ringing reduces the amount of computer computation required in a network, may be useful in neural networks.

[0075] When the data size is reduced, the data may be further flattened in some implementations , meaning that the data can be arranged into a one-dimensional vector. The data is flattened for the purpose of matrix multiplication that occurs in fully connected layers. Therefore, the fully connected layer 127 may receive a flattened one-dimensional vector of length 10800 (for example , from the 400×3×3×3 array of step 107, but the vector may have a different length in other implementations ). In the fully connected layer of a convolutional neural network , each number in the one-dimensional vector is applied to a neuron. The neuron sums the inputs and applies an activation function. In some embodiments, the activation function is a rectified linear function . In alternative embodiments, the activation function may be a hyperbolic tangent or a sigmoid function .

[0076] In the illustrated implementation, the first fully connected layer 127 outputs a one-dimensional vector of length 10800 at 108 (although other lengths may be utilized as discussed above ). The vector output by the fully connected layer represents a vector of real numbers . In some embodiments, the real numbers may be output and classified. In other embodiments, the real numbers may be further input to subsequent fully connected layers to improve the accuracy of the convolutional neural network .

[0077] In the present embodiment, the output of the first fully connected layer 108 is input to the second fully connected layer shown at 128 . The output of the first fully connected layer 108 is, is already a one-dimensional vector, so there is no need to flatten it before it is input to the subsequent fully connected layer . In some embodiments, to improve the accuracy of the neural network , additional fully connected layers are implemented. The number of additional fully connected layers may be limited by the processing capability of the computer running the neural network . Alternatively, the addition of fully connected layers may be limited by a slight increase in accuracy compared to the increase in computer computation time for processing the additional fully connected layers .

[0078] In the illustrated implementation, the second fully connected layer 128 outputs a 1-dimensional vector of length 1000 at 109 (although other lengths may be utilized) . The vector output by the fully connected layer represents a vector of real numbers. In some embodiments , the real numbers may be output and classified. In other embodiments, the real numbers may be further input to subsequent fully connected layers to improve the accuracy of the convolutional neural network . .

[0079] At 129, in some implementations, the output of the fully connected layer 109 is input to a softmax classifier . The softmax classifier uses a softmax function or a normalized exponential function to convert real number inputs into a normalized probability distribution over predicted output classes . In alternative embodiments, a sigmoid function may be used to classify the output of the convolutional neural network . The sigmoid function may be used when there is one class. The softmax function is a multi-class sigmoid function .

[0080] ​​​In step 110, the output of the softmax layer is such that each of the 20 identified amino acids is targeted This is the probability of improving the properties of the target protein (however, a larger or smaller number of amino acids) (However, this may be used in other implementations.) This output is an additional convolutional neural network. The system can perform different queries given the predicted amino acid sequence. It may be input to the convolutional neural network, or output 110 may be the target tan It may be used directly as a predictive amino acid to improve the properties of the protein.

[0081] Figure 1B shows the implementation flow of a method for determining amino acid residues at the center of the microenvironment. This is a chart. A neural network classifies outputs given a specific input. To learn the law, neural networks are trained on known input / output pairs. It may be done. Once the neural network classifies known input / output pairs As it learns the rules, the neural network predicts what the classified output should be. Therefore, it can operate with unknown inputs. In this embodiment, a neural network It is trained to predict amino acids at the center of the microenvironment. During testing, the neural The network is provided with an amino acid sequence and analyzes the microenvironment surrounding the amino acids, natural It is possible to predict amino acid residues different from amino acid residues. Neural network prediction amino Acids cause improved proteins to mutate native amino acid residues into predicted amino acid residues. This shows that it can be generated by [this method].

[0082] In step 130, some implementations train the neural network Various sets of protein samples used for the purpose may be compiled or constructed. The more diverse the sample set is, the more the neural network will perform in its classification can be more robust. For example, the neural network attempts to classify input / out put pairs during the first iteration of learning. During the next learning iteration, if the input / output pairs are similar to the learned input / output pairs of the first iteration, the neural network not because the neural network is robust, but merely because the data are similar, can perform artificially higher than it should A diverse set of input / output pairs is then used for the third iteration When input to the network, the classification error is much more likely to be larger than if the first two input / output pairs were diverse. The similarity of the first two input / output pairs The similarity of the first two input / output pairs may cause the neural network to fine-tune itself to learn the similar input / output pairs of the first two iterations. This may be referred to as "overtraining the network". The similarity of the first two input / output pairs may cause the neural network to fine-tune itself to learn the similar input / output pairs of the first two iterations. This may be referred to as "overtraining the network". "the network".

[0083] Alternatively, if the second iteration of training uses distinct input / output pairs compared to the input / output pairs of the first iteration, the neural network is forced to be able to classify a broader range of input / output pairs. Since the output is not known during testing, it is ideal that the network is capable of classifying a wide range of input / output pairs. the neural network is forced to be able to classify a broader range of input / output pairs. Since the output is not known during testing, it is ideal that the network is capable of classifying a wide range of input / output pairs. it is ideal that the network is capable of classifying a wide range of input / output pairs.

[0084] Accordingly, in some implementations of step 130, the training dataset for the neural network is composed of proteins that all diverge phylogenetically over a certain threshold the training dataset for the neural network is composed of proteins that all diverge phylogenetically over a certain threshold It is constructed from. In various embodiments, the dataset is at least 20%, 30%, 40%. It is constructed from % or 50% phylogenetically branched proteins. Retaring removes highly similar / duplicate proteins that may occur repeatedly in the training set. This improves efficiency. Such improvements are beneficial for oversampled proteins. This can reduce the biases that exist in the current state of technology.

[0085] In some embodiments, individual proteins in the training dataset are unannotated. These protein database (PDB) structures are modified by adding hydrogen atoms. In one embodiment, the addition of hydrogen atoms is performed by a software converter, for example, pdb2pqr This is achieved using [a specific method]. In another embodiment, atoms are determined by the bonding ability of each atom, and the DNA bone. Further separation occurs due to the presence of other atoms, such as phosphorus, within the compound.

[0086] In some embodiments, individual proteins in the training set are partially charged, and beta factors are used. Additional characteristics of proteins, including but not limited to secondary structure, aromaticity, and polarity. Taking gender into consideration, the repair can be performed by adding biophysical channels to the protein model. I decorated it.

[0087] In some embodiments, high-resolution and low-resolution models of the same protein are provided. Training data may be removed if it can coexist within the protein database. According to some implementations of the method discussed in the book, the related structure has a resolution below a threshold. All genes are grouped together into a group that has sequence similarity exceeding a certain percentage threshold. It may be divided into sections. Where used herein, “resolution” typically refers to angs. This refers to the resolution of the electron density map of a molecule measured by a Trom (A) device. The electron density map is This means that the image can be resolved at a lower distance between points, allowing more features of the molecular structure to be seen. Therefore, molecular models with "lower" resolution are "higher" molecular models It is of higher quality than Dell. For example, the relevant structures, as well as resolutions of less than 2.5 Å and All genes with at least 50% sequence similarity are grouped together, and A low-resolution available structure is selected for use in the training model, and a higher resolution structure is selected. Lower-resolution (lower-quality) molecular models are removed.

[0088] In some embodiments, amino acid sampling is performed to ensure that all 20 amino acids are equal. In contrast to the expression, cysteine ​​was normalized by its abundance in the PDB. In terms of form, amino acid sampling may be normalized to spontaneous generation. One embodiment Therefore, amino acid sampling may be normalized to spontaneous generation within a given species. Because cysteine ​​can be artificially assigned a high probability at any given position, Amino acids were modified in the data samples. Cysteine ​​is the rarest amino acid observed in the PDB. It is an amino acid, and therefore, a more abundant amino acid may be under-sampled and occupied. The diversity of the possible protein microenvironments may not have been fully represented. Modifying the cysteine ​​amino acids in the sample results in a significant increase in wild-type accuracy. The accuracy for each amino acid ranged from 96.7% to 32.8% (see Figure 2A). stomach).

[0089] In step 131, amino acids in the protein are randomly selected from the amino acid sequence. It may be sampled. In one embodiment, up to 50% of the amino acids in the protein are sampled. Samples are taken unless the protein is large, and in that case, 100% of the individual proteins are taken. A sample of one or fewer amino acids was taken. In another embodiment, the upper limit is per individual protein. It consisted of approximately 200 amino acids. The disclosed sampling method involved the outer layer of the protein. Remove bias from the dataset regarding residues.

[0090] In step 132, the three-dimensional model of the protein crystal structure includes each amino acid in the structure. It can be created along with an acid-related microenvironment. For example, to generate a three-dimensional model. There are several other methods, but an unknown protein model is a known protein Fragment set when constructed from a pool of candidate fragments taken from the qualitative structure, known protein Segment matching occurs when a segment matches the amino acid sequence, or when a known protein sequence matches. Dell is selected ("template"), and the residues of the amino acid sequence are the residues in the template sequence. It is mapped (aligned) and constrained to various distances, angles, and dihedral angles in the array. The empty space is derived from the alignment with the template structure, and constraint violations are minimized. Comparative protein modeling based on the achievement of inter-constraints is one example. When a three-dimensional model is generated, a microenvironment related to each amino acid, including its structure, is also created. One of the shortcomings of existing protein structure databases is that when new proteins are added... The difference lies in the fact that different methods are used to create the crystal structure. Different methods may have different biases or archetypes that could affect the accuracy of the model. You can add a function. You can rebuild the structure using the latest version of the same method. Therefore, the training structure does not contain artifacts or errors present in older versions. This confirms that the chemical composition will change.

[0091] In step 133, the three-dimensional model generated from step 132 is used to create a three-dimensional array. It may be mapped to I. In one example, the coordinates of the three-dimensional model are stored in a three-dimensional array. In some embodiments, the three-dimensional image may be generated from a three-dimensional model. The 3D image may be mapped to a three-dimensional array. The image data in the array is boxed. It can be called a matrix. Pixels are addressable in the image in two-dimensional space. Just as a voxel represents an element, it represents an addressable element in three-dimensional space.

[0092] In step 134, the image is processed into a convolutional layer within the convolutional neural network. The input is received. The convolutional layer detects image features through a filter. The filter is applied to the image. It is designed to detect the presence of specific features within. In a simplified example, high-pass filters are used. The filter detects the presence of a high-frequency signal. The output of the high-pass filter is the signal that has a high frequency. This is the part with the number. Similarly, image filters are designed to track specific features within an image. It is possible. The more filters applied to an image, the more features can be tracked.

[0093] In step 135, the image is convolved with a filter in the convolutional layer, and within the image Extract the filter-specific features. In the convolutional layer, the filter is the input and filter. The dot stack for each element of the input is slid, and the input is stored as a feature map.

[0094] The decision in 136 depends on whether there are more filters. As such, more filters implemented aim to track more features within the image. It can be perceived as such. Each filter is independently convolved with the image to create an independent feature map. If more filters are convolved with the image, steps 134 and 135 are repeated. It is possible. If all filters are convolved with the image, the process is as follows: Step 13 Proceed to 7. In some embodiments, feature maps are concatenated together and applied to the image. It is possible to create a feature map as deep as the number of filters. In other embodiments, the feature map They may be processed one at a time.

[0095] In step 137, the activation function is the convolution of the convolutional neural network. This is applied to the feature map of the layer. The activation function is the feature extracted by the neural network. This enables the detection of nonlinear patterns within the map. The formula is f(x) = max(0,x). A normalized linear function having the following characteristics may be applied to the feature map. The normalized linear activation function is It behaves linearly for positive values, making this function easy to optimize, and then neural This enables the network to achieve higher accuracy. Also, the normalized linear activation function... This means that it outputs zero for any negative input, and that it is not a true linear function. Therefore, the output of the convolutional layer in a convolutional neural network is in the feature map. Yes, the values ​​within the feature map have passed through a normalized linear activation function.

[0096] The decision at 138 depends on whether there are further convolutional layers. Increasing the number can increase the complexity of the features that can be tracked. In addition, a new filter may be applied to the image, and the process repeats steps 134-138. It is possible. In some embodiments, the filter ensures accuracy of the tracked features. In an alternative embodiment, different The filter may be incorporated into the second convolutional layer. If there are no further convolutional layers, The process proceeds to step 139.

[0097] In step 139, the pooling layer downsamples the data. A pooling window may be applied to the feature map. In some embodiments, pooling The layer outputs the maximum value of the data within the window, and then downsamples the data within the window. Pooling. Maximum pooling highlights the most prominent features in the pooling window. Other In this embodiment, the pooling layer outputs the average value of the data within the window.

[0098] The decision at 140 depends on whether there are further convolutional layers. Increasing the number can increase the complexity of the features that can be tracked. In addition, a new filter may be applied to the image, and the process repeats steps 134-140. It is possible. In some embodiments, the filter ensures accuracy of the tracked features. This is the same as in the first convolutional layer. In an alternative embodiment, a different filter is used. However, it may be incorporated into the second convolutional layer. Therefore, steps 134-136, 1 The repeated iterations from 34 to 138 and 134 to 140 demonstrate the flexibility of the tracked features. This provides increased complexity. Without further convolutional layers, the process is step Proceed to 141.

[0099] In step 141, some implementations flatten the downsampled data. It is simplified. This means that the data is arranged in a one-dimensional vector. The matrix is ​​flattened for the purpose of matrix multiplication that occurs in fully connected layers.

[0100] In step 142, in some implementations, the flattened one-dimensional vector is nu It is input into the fully connected layers of a convolutional neural network. In a fully connected layer of the matrix, each number of one-dimensional vectors is suitable as input to the neuron. It is used. Neurons sum up their inputs and apply an activation function. In some embodiments In this embodiment, the activation function is a normalized linear function. In an alternative embodiment, the activation function is a hyperbola. It may also be a tangent or sigmoid function.

[0101] In some embodiments, the output of a first set of neurons in a fully connected layer is , may be input to another set of neurons via weights. Each subsequent set of neurons The hidden layer can be called the "hidden layer" of a neuron. The number of hidden layers in a fully connected neuron is... They may be removed. In other words, the number of hidden layers in a neural network is equal to the number of neurons. The network can adaptively change as it learns how to classify the output.

[0102] In step 143, some implementations include a fully connected network. Neurons are connected to other neurons by weights. Weights are the weights of some neurons It is adjusted to enhance the effect and weaken the effect of other neurons. Through adjustment, the neural network becomes capable of classifying outputs more effectively. The weights connecting the neural networks are used to classify inputs or "training" values. The law is adjusted while it is being learned. In some embodiments, within the neural network The number of neurons can be removed. In other words, within a neural network The number of active neurons determines how the neural network learns to classify its output. It changes adaptively as you learn.

[0103] The decision at 144 depends on whether there are additional fully connected layers. How many? In one embodiment, the output of one fully connected layer is sent to a second fully connected layer. It can be used as input. In some embodiments, it can be used to improve the accuracy of a neural network. Therefore, an additional fully connected layer is implemented. The number of additional fully connected layers is n It may be limited by the processing power of the computer running the network. Adding fully connected layers requires computing power to process the additional fully connected layers. The increase in computation time may be limited by a slight increase in precision. Alternative implementation In this state, the output of one fully connected layer may be sufficient to classify an image. If there is a fully connected layer, steps 142 and 143 are performed when the input vector is heavy It is repeated so that it is supplied to neurons that are connected to each other via the membrane. If there are no connected layers, the process proceeds to step 145.

[0104] In step 145, in some implementations, the fully connected layers are real vectors Outputs a value. In some embodiments, real numbers may be output and classified. Alternative In this embodiment, the output of the fully connected layer is input to a softmax classifier. The Tomax classifier uses either the softmax function or the normalized exponential function to classify real numbers Transforms the input into a normalized probability distribution for the predicted output class. Other implementations In this approach, the sigmoid function is used to classify the output of the convolutional neural network. That's also good. The sigmoid function can be used when there is one class. Softmax function The number is a multiclass sigmoid function. In some embodiments, a neural network The output of the workpiece represents the predicted amino acid residue at the center of the chemical microenvironment.

[0105] For example, a neural network outputs a vector of length 20 containing 20 real numbers. It is also possible. The vector has a length of 20, and this is because there are 20 possible amino acids. This is because it can exist at the center of the microenvironment. The value within the vector is at the center of the microenvironment. To represent the possibilities of amino acids, the real numbers in the vector pass through a softmax classifier. .

[0106] In step 146, in some implementations, the predicted amino acid residue is central to the chemical environment. It is compared to natural amino acids in [the context of amino acids]. For example, the true amino acid vector is a vector of length 20. It may also be a toru, and a single "1" indicates a natural amino acid at the center of the chemical environment, Other values ​​within the ctor will hold "0".

[0107] In neural networks, learning involves comparing known input / output pairs during training. This type of learning is called supervised learning. The difference between the predicted value and the known value is determined. The information is then backpropagated through the neural network. The weights are then used to determine the error. This method of training neural networks is called backpropagation. It is called [name].

[0108] In step 147, in some implementations, the weights are updated via the steepest descent method. The following equation 1 shows how the weights are adjusted in each iteration n.

number

[0109] The steepest descent method is an optimization technique that minimizes the objective function. In other words, the steepest descent method is It is possible to adjust unknown parameters in the direction of the steepest descent. During training, the neural The weight values ​​that optimize the network's classification accuracy are unknown. Therefore, the weights are... This is an unknown parameter that adjusts in the direction of a steep descent.

[0110] In some embodiments, the objective function may be the cross-entropy error function. Minimizing the differential entropy error function involves the probability distribution of the predicted amino acid vector and the natural amino acid This represents minimizing the difference between the probability distribution of the amino acid vector and the given distribution. In some embodiments, this means minimizing the difference between the probability distribution and the given distribution. The objective function may also be the squared error function. Minimizing the squared error objective function is the goal of each This refers to minimizing the instantaneous error of neurons.

[0111] During each training iteration, the weights are adjusted to approach their optimal values. Depending on the position of the eurons, different formulas are used to determine how the weights are adjusted to the objective function. This determines whether it will happen. Equation 2 below shows that the weights between neuron i and neuron j are This shows how it is adjusted for the cross-entropy error function.

number

[0112] In some embodiments, the weights may be adjusted each time a weight modification is determined. This type of training can be called online or incremental training. One advantage of Tahl training is that it allows neural networks to track small changes in input. One of its capabilities is that the weights are input / The output pairs may be corrected after receiving a batch. This type of training is batch training and It can be called. One advantage of batch training is that the neural network to the optimized weight values One advantage is faster convergence of the network. In this embodiment, the neural network is used We trained on 1.6 million pairs of amino acids and microenvironments. In this embodiment, 20 A different size was used. In step 148, the counter is incremented. The neural network completes one round of batch training when the counter reaches 20. To finish. In other words, the neural network is based on 20 input / output pairs. Evaluating this itself completes one round of training.

[0113] The decision in 149 depends on whether the current batch of training samples has completed. If the number of training samples required to fill one batch is achieved, the network will Then proceed to step 150. As discussed above, one training batch consists of 20 inputs / outputs. A force pair is needed. The number of samples required to fill one batch has not been achieved. If not, the neural network repeats steps 134-149.

[0114] In step 150, the weights temporarily stored in step 147 are corrected. The weights are calculated. The weight values ​​are calculated by adding the newly modified weights to a new batch of 20 input / output pairs. The value is used for evaluation, and the corrections are adjusted according to the total corrections.

[0115] The decision in 151 depends on whether the maximum number of training iterations has been reached. Batch training Once one round is completed, one training iteration is finished. In some situations, the weights Because the weights keep fluctuating around that optimal value, they never actually reach that optimal value. There is a possibility that... Therefore, in some embodiments, the neural network is A maximum number of repetitions can be set to prevent the twerk from being trained indefinitely.

[0116] If the maximum number of iterations has not been reached, the neural network created in step 130 Train the network again using another input / output pair from the data sample. It may be permitted. The iteration counter is used when the neural network completes one batch of training. After completion, it increases in step 153.

[0117] When the maximum number of iterations is reached, the neural network can store the weight values. P152 indicates that the weight values ​​are stored. These weights are trained by the network. These are refined weights, which are then used when testing the neural network. Therefore, it is stored in memory.

[0118] If the number of iterations is not reached, the error between the predicted amino acid residue and the known native amino acid residue. This may be evaluated. This evaluation is carried out in step 154. Several situations Therefore, the error between the predicted value and the known natural value is very small, and thus the error is acceptable. It can be considered that, once a neural network is established, it does not need to continue training. Then, the weight values ​​that resulted in such a small error rate are stored and subsequently used in the test. It may be done. In some embodiments, the neural network is a neural network The tool either predicts one output very well or predicts one output very well incorrectly. To ensure that it did not learn how to make predictions, a small error over several iterations. The rate must be maintained. The network must maintain a small error over several iterations. When you request this from the network, it becomes more likely that the network is properly classifying a diverse range of inputs. If the error between the predicted value and the known value is still too large, the neural network will... Continue training itself and repeat steps 131-154. In many implementations, During the iterations of steps 131-154, the neural network receives a new dataset. Use this to train the neural network.

[0119] Figure 1C is a flowchart of the implementation of a method to enhance the properties of synthetic proteins during testing. Yes. In step 160, the weights memorized from the training scenario are used in step 172. These weights are set as the weights of a fully connected layer. It should be highly likely that accurate classification is possible, and the weights should be broad and Because it is trained through a diverse set of inputs, when an unknown input needs to be classified... It is used.

[0120] In step 161, in some implementations, an unknown protein is randomly sun It is pulled. In one embodiment, up to 50% of the amino acids in the protein are pulled. Samples are taken unless the sample size is large, in which case fewer than 100 samples are taken from each individual protein. The amino acids were sampled. In another embodiment, the upper limit was 2 per individual protein. These are 00 amino acids. The disclosed sampling method involves sampling the residues on the outside of the protein. Remove bias from the dataset.

[0121] In step 162, the three-dimensional model of the protein crystal structure includes each amino acid in the structure. It can be created along with a microenvironment related to acid. Several for generating a three-dimensional model There are other methods, but one is to determine whether an unknown protein model is a known protein structure. Fragment set when constructed from a pool of candidate fragments, known protein segment Segment matching occurs when the amino acid sequence matches, or when a known protein model is selected. Selected ("template"), residues in the amino acid sequence map to residues in the template sequence. Alignment is performed, and constraints on various distances, angles, and dihedral angles in the arrangement are applied. Spatial constraints derived from alignment with rate structures, where constraint violations are minimized. Comparative protein modeling based on the achievement of the three-dimensional model of protein crystal structure is one example. When a molecule is generated, a microenvironment related to each amino acid, including its structure, is also created. One of the flaws in protein structure databases is that when a new protein is added, the crystalline structure changes. The difference lies in the use of different methods to create the structure. The law considers different biases or artifacts that may affect the accuracy of the model. It can be added by rebuilding the structure using the latest and same version of the same method. The training structure is chemical, rather than an artifact or error present in older versions. It becomes certain that the composition will change.

[0122] In step 163, the three-dimensional model generated from step 162 is used to create a three-dimensional array. It may be mapped to I. In one example, the coordinates of the three-dimensional model are stored in a three-dimensional array. In some embodiments, the three-dimensional image may be generated from a three-dimensional model. The 3D image may be mapped to a three-dimensional array. The image data in the array is boxed. It can be called a matrix. Pixels are addressable in the image in two-dimensional space. Just as a voxel represents an element, it represents an addressable element in three-dimensional space.

[0123] In step 164, the image is processed into a convolutional layer within the convolutional neural network. It may also be input. The convolutional layer detects image features through filters. It is designed to detect the presence of specific features in an image. In a simplified example, high A pass filter detects the presence of a high-frequency signal. The output of a high-pass filter contains high frequencies. This is the part of the signal that does this. Similarly, an image filter tracks specific features in an image. It can be designed that the more filters applied to an image, the more features can be tracked. .

[0124] In step 165, the image is convolved with a filter in the convolutional layer, and within the image Extract the filter-specific features. In the convolutional layer, the filter is the input and filter. The dot stack for each element of the input is slid, and the input is stored as a feature map.

[0125] The decision in 166 depends on whether there are more filters. As such, more filters implemented aim to track more features within the image. Each filter is independently convolved with the image to create an independent feature map. If more filters are convolved with the image, steps 164 and 165 are repeated. It is possible. If all filters are convolved with the image, the process is as follows: step 167 Proceed to the next step. In some embodiments, feature maps are concatenated together and applied to the image. It is possible to create feature maps as deep as the number of filters. In other embodiments, the feature map is Alternatively, they may be processed one at a time.

[0126] In step 167, some implementations use an activation function that is a convolutional neural network. It is applied to the feature map of the convolutional layer of the network. The activation function is the neural network This allows for the detection of nonlinear patterns within the extracted feature map. Equation f(x A normalized linear function with ) = max(0,x) is applied to the feature map as an activation function. It is also acceptable. A normalized linear activation function behaves linearly for positive values, and this function can be optimized. This makes it easier, and then enables neural networks to achieve high prediction accuracy. To enable it. Also, a normalized linear activation function outputs zero for any negative input, and This means that it is not a true linear function. Therefore, convolutional neural network The output of the convolutional layer is a feature map, and the values ​​within the feature map are normalized linear activation relations. It can pass through the numbers.

[0127] The decision in 168 depends on whether there are further convolutional layers. Increasing the number can increase the complexity of the features that can be tracked. Then, a new filter is applied to the image, and steps 164-168 may be repeated. In some embodiments, the filter performs a first tatami mat to ensure the accuracy of the tracked features. It is the same as in the composting layer. In an alternative embodiment, a different filter is used in the second composting layer. It may be incorporated into the convolutional layer. If there are no further convolutional layers, the process is step Proceed to 169.

[0128] In step 169, the pooling layer downsamples the data. A pooling window may be applied to the feature map. In some embodiments, pooling The layer outputs the maximum value of the data within the window, and then downsamples the data within the window. Pooling. Maximum pooling highlights the most prominent features in the pooling window. Other In this embodiment, the pooling layer outputs the average value of the data within the window.

[0129] The decision at 170 depends on whether there are further convolutional layers. Increasing the number can increase the complexity of the features that can be tracked. Then, a new filter is applied to the image, and steps 164-170 may be repeated. In some embodiments, the filter performs a first tatami mat to ensure the accuracy of the tracked features. It is the same as in the composting layer. In an alternative embodiment, a different filter is used in the second composting layer. It may be incorporated into the convolutional layer. If there are no further convolutional layers, the process is step Proceed to 171.

[0130] In step 171, some implementations flatten the downsampled data. It is simplified. This means that the data is arranged in a one-dimensional vector. The matrix is ​​flattened for the purpose of matrix multiplication that occurs in fully connected layers.

[0131] In step 172, in some implementations, the flattened one-dimensional vector is nu It is input into the fully connected layers of a convolutional neural network. In a fully connected layer, each number of one-dimensional vectors is applied to a neuron. The euron sums the inputs and applies an activation function. In some embodiments, the activation function The number is a normalized linear function. In an alternative embodiment, the activation function is the hyperbolic tangent or cyane. A gmoid function would also be acceptable.

[0132] In step 173, some implementations include a fully connected network. Neurons are weighted. The weights in a fully connected network are stepwise. These are the weights initialized at 160. These weights are such that the weights accurately divide the unknown input. The weights should be broad and diverse, so that it is highly likely that similarity is possible. Because it is trained through a set, it is used when unknown inputs are being evaluated.

[0133] The decision in 174 depends on whether there are additional fully connected layers. How many? In one embodiment, the output of one fully connected layer is sent to a second fully connected layer. It can be used as input. In some embodiments, it can be used to improve the accuracy of a neural network. Therefore, an additional fully connected layer is implemented. The number of additional fully connected layers is n It may be limited by the processing power of the computer running the network. Adding fully connected layers requires computing power to process the additional fully connected layers. The increase in computation time may be limited by a slight increase in precision. Alternative implementation In this state, the output of one fully connected layer may be sufficient to classify an image. If there is a fully connected layer, steps 172 and 173 are performed when the input vector is heavy It is repeated so that it is supplied to neurons that are connected to each other via the membrane. If there are no connected layers, the process proceeds to step 175.

[0134] In step 175, the fully connected layers output a vector of real numbers. In that embodiment, real numbers may be output and classified. In an alternative embodiment, completely The output of the connected layer is input to the softmax classifier. The softmax classifier is Using the softmax function or a normalized exponential function, a real number input is predicted. Convert to a normalized probability distribution for the output class. In other embodiments, a sigmoid function The output of the convolutional neural network may be classified using numbers. The number can be used when there is one class. The softmax function is used when there are multiple classes. This is a gmoid function. In some embodiments, the output of the neural network is a tangent function. This represents candidate residues and amino acid residues predicted to improve protein quality indicators.

[0135] In step 176, the synthetic protein is produced according to the output of the neural network. Synthetic proteins can be generated. Synthetic proteins are used in computing to run neural networks. The computing device that runs the neural network is a computing device that runs the neural network. Another computing device communicating with a third-party manufacturer or researcher Candidate amino acid residues identified by location or by neural networks and It can be generated by another entity that performs substitutions according to the predicted amino acid residues. For example, several In the embodiment, the synthetic protein is generated by a neural network and / or Neural networks or computing that runs neural networks One or more substitutions are made according to the predicted amino acid residues and candidate residues identified by the device orientation. This may be obtained by an entity that performs this action. In some embodiments, a neural network This can predict amino acid residues that are the same as natural amino acid residues. In other embodiments, The neural network can predict amino acid residues that differ from those of natural amino acids. The Ral network predicts amino acids, and improved proteins predict natural amino acid residues. This demonstrates that it can be produced by mutating amino acid residues. Therefore, synthetic tan The proteins can be generated according to the output of the neural network.

[0136] Figure 1D shows block diagrams of neural networks under training, based on several implementations. The input is supplied to the neural network at 180. Furthermore, neural networks are capable of accepting various inputs. In the embodiment, the neural network accepts an amino acid sequence or residue. In this embodiment, the neural network has discrete attributes located at each of a series of positions. You may receive a chain of amino acids.

[0137] In the block diagram, 181 represents a neural network that changes over time. As described above, during training, the neural network adaptively processes each new input / output iteration. The weights are calculated based on the error signal, which is determined by the difference between the predicted output and the known output. Because it is constantly being updated, neural networks are adaptively updated.

[0138] In the block diagram, 182 represents the output predicted by the neural network that satisfies the query. This indicates that... For example, a neural network can identify specific amino acid residues that can be modified... Queries may be made and trained to determine the behavior. In these situations, a neural network The output of the workpiece may be an amino acid residue, and the amino acid residue has improved properties. It may be used to synthesize new proteins. In other embodiments, neural The network output may also be an amino acid residue that can be used as a substitution, and the substitution is It may be used to synthesize new proteins with improved properties. Other implementations Morphologically, neural networks are proteins with parameters different from the initial amino acid sequence. Queries may be made regarding the proposed amino acid sequence for the substance. So, the output of the neural network is a specific amino acid for each residue in the amino acid sequence. It may also be an amino acid sequence that exhibits acidity.

[0139] In the block diagram, 186 represents the desired value. This type of training is performed on a neural network. In order to train a robot, the input corresponding to the output must be known, therefore supervised training is necessary. This is called training. During training, the neural network produces results that are as close as possible to the desired value. They are asked to exert their strength.

[0140] The desired value 186 and the output value from the neural network 182 are at 185. The difference between the output value and the desired value is determined and processed through a neural network. Since this error signal 183 is propagated again, the neural network learns from this error. It can be learned. As shown in equations 1 and 2 above, the weights are based on the error signal. It will be updated.

[0141] Figure 1E is a block diagram of a convolutional neural network in several implementations. In the block diagram, 190 represents a convolutional layer. The convolutional layer filters the image through a filter. It can detect the features of. The filter is designed to detect the presence of specific features in the image. In a simplified example, a high-pass filter detects the presence of a high-frequency signal. The output of an image filter is the portion of the signal that has high frequencies. Similarly, an image filter is used to filter an image. It can be designed to track specific features within the image. The more filters applied to the image, the better. However, this also means there will be more features that can be tracked.

[0142] In some implementations, the image is convolved with a filter in a convolutional layer, and the frames in the image are processed. Extract filter-specific features. In the convolutional layer, the filter extracts the input and filter characteristics. The input is stored as a feature map by sliding along the dot product of each element. The activation function is This is applied to the feature maps of the convolutional layers of a convolutional neural network. Activation function This involves the neural network detecting nonlinear patterns within the extracted feature map. This makes it possible. A normalized linear function with the equation f(x)=max(0,x) can be used in a feature map. It may be applied. A normalized linear activation function behaves linearly for positive values, and this function To make optimization easier, and then enable the neural network to achieve high prediction accuracy. This makes it possible. Furthermore, the normalized linear activation function outputs zero for any negative input. This means that it is not a true linear function. Therefore, convolutional neural network The output of the convolutional layer in the workpiece is a feature map, and the values ​​within the feature map are normalized linear activity. It passes through the transformation function. In other embodiments, the sigmoid function or the hyperbolic tangent function is used. It can be applied as an activation function.

[0143] The extracted feature map, which is acted upon by an activation function, is then processed by 191. As shown, the data may be fed into a pooling layer. The pooling layer downsamples the data. Pooling. A pooling window may be applied to the feature map. In this embodiment, the pooling layer outputs the maximum value of the data within the window, and within the window The data is downsampled. Maximum pooling is most pronounced in the pooling window. To emphasize a distinctive feature.

[0144] The downsampled pooled data is then, in some implementations, convolved. It can be flattened before being input to the fully connected layer 192 of the neural network. .

[0145] In some embodiments, a fully connected layer has only one set of neurons. This may be done. In an alternative embodiment, the fully connected layer is the neuron in the first layer 193 It may have a set of neurons, and a set of neurons in the subsequent hidden layer 194. The number of hidden layers in the connected network may be removed. In other words, the neural network The number of hidden layers in a network increases as the neural network learns how to classify its outputs. It can change adaptively.

[0146] In fully connected layers, each neuron in layers 193 and 194 is connected to each other. They are connected by weights. During training, the weights are connected to several neurons. It is adjusted to enhance the effect of Ron and weaken the effect of other neurons. By adjusting the intensity, the neural network can better classify the output. In some embodiments, the number of neurons in the neural network is reduced. It is also acceptable to do so. In other words, the neurons that are active within the neural network The number changes adaptively as the neural network learns how to classify the output. ru.

[0147] After training, the error between the predicted value and the known value may be very small, therefore the error is... If deemed acceptable, the neural network does not need to be trained further. In those situations, the weight values ​​that resulted in such a small error rate are stored, and then, It may be used in experiments. In some embodiments, the neural network is neural The network either predicts one output very well or predicts one output very well To ensure that it did not learn how to predict incorrectly, for several iterations A small error rate must be maintained. Maintain a small error over several iterations. When a request is made to the network, it is possible that the network is appropriately classifying a diverse range of inputs. The likelihood increases.

[0148] In the block diagram, 195 represents the output of the neural network. Fully connected layers. The output is a vector of real numbers. In some embodiments, the real numbers are used for output and classification. This may be done. In an alternative embodiment, the output of a fully connected layer is a softmax classifier. It will be entered into.

[0149] In the block diagram, 196 represents the softmax classifier layer. The softmax classifier is, Using the softmax function or a normalized exponential function, a real number input is predicted. Convert to a normalized probability distribution for the output class. In other embodiments, a sigmoid function The output of the convolutional neural network may be classified using numbers. The number can be used when there is one class. The softmax function is used when there are multiple classes. This is a gmoid function. In some embodiments, the output of the neural network is a tangent function. This represents candidate residues and amino acid residues predicted to improve protein quality indicators. In this embodiment, the output of the neural network identifies each residue in the amino acid sequence. It may also be an amino acid sequence showing the amino acids.

[0150] In some embodiments, the problematic residue is identified, and multiple independently trained neurons are used. By combining predictions from a multi-network, a new type of residue has been proposed. By identifying residues based on independently trained neural networks. , a neural network appears during training, and any individual neural network Biases caused by inherent singularities can be removed. Many independent neural networks Averaging the work eliminates the biases associated with any given individual neural network.

[0151] Accurate accuracy was improved through various improvements to the existing algorithm. Figure 2B shows... As shown, the various improvements, when taken together, in one embodiment, lead to the prediction of wild-type amino acids. Dell accuracy was increased from approximately 40% to over 70% across all amino acids.

[0152] Manipulated proteins The implementation of the systems and methods discussed herein may involve natural proteins or parental proteins. One or more modifications that modify the desired trait or characteristic of the protein compared to the trait or characteristic of the quality. Further providing or identifying compositions comprising engineered proteins containing mutations. In this context, the systems and methods described herein are generated or identified by the implementation of these systems and methods. Modified proteins are used in the three-dimensional convolution of the systems and methods discussed herein. One or more predictions made by a neural network (3DCNN) prediction pipeline Conferring a desired trait or characteristic to a protein by including one or more mutations in the amino acid residue. This includes mutations in residues predicted from analysis using a 3DCNN prediction pipeline. The systems and methods described herein are generated by the implementation of the systems and methods described herein. The identified manipulated proteins are, in this specification, proteins manipulated by 3DCNN. It is called a "ku" quality.

[0153] 3DC generated or identified by the implementation of the systems and methods discussed herein. Examples of traits or properties that can be modified in proteins manipulated with NN include stability and parental rights. These include compatibility, activity, half-life, fluorescence properties, and sensitivity to photobleaching, but these are... Not limited.

[0154] 3DC generated or identified by the implementation of the systems and methods discussed herein. Proteins manipulated with NN can be prepared using chemical methods, for example, 3DCN. Proteins manipulated with N are produced using solid-phase technology (Roberge JY et al (199) 5) Synthesized by Science 269:202-204), cut from resin, It can be purified by preparative high-performance liquid chromatography. Automated synthesis is used, for example, in manufacturing. Follow the instructions provided by the manufacturer to use the ABI 431 A peptide synthesizer (Perki This can be achieved using (n Elmer).

[0155] Proteins manipulated with 3DCNNs are, alternatively, combined by translation of the coding nucleic acid sequence. They may be produced by alternative means or by cleavage from longer protein sequences. The composition of proteins manipulated by 3DCNN can be determined by amino acid analysis or sequencing. It may be approved.

[0156] 3DC generated or identified by the implementation of the systems and methods discussed herein. Protein variants manipulated with NN are characterized by (i) having one or more amino acid residues, Substituted with a conserved or non-conserved amino acid residue (preferably a conserved amino acid residue), and The substituted amino acid residues may or may not be encoded by the gene code. (ii) One or more modified amino acid residues, for example, residues modified by the attachment of substituents (iii) a protein fragment manipulated with 3DCNN, and / or (iv) )A protein manipulated by 3DCNN is fused with another protein or polypeptide. It may be the result of manipulating the protein sequence in the original 3DCNN. It contains polypeptides produced via quantifiable cleavage (including multi-site protein degradation). The riant may be modified post-translation or chemically. Such variants are original Based on the detailed instructions, it is considered to be within the scope of those skilled in the art.

[0157] As is known in the art, the "similarity" between two polypeptides is that one polypeptide The amino acid sequence of Tide and its conserved amino acid substitutions are used in the sequence of the second polypeptide. Determined by comparison. A variant is a different segment from the original array. Less than 40% of the residues per segment differ from the original sequence. Less than 25% of the sequence differs from the original, and less than 10% of the residues per target segment differ. or, by only a few residues per target segment, the original protein sequence is different. And at the same time, the functionality of the original sequence and / or ubiquitin or ubiquitinated tannins. To retain the ability to bind to the protein, the polypeptide is sufficiently homologous to the original sequence. Defined to include a cyd sequence. Using the implementation of the systems and methods discussed herein Using this method, at least 60%, 65%, 70%, 72%, and 74% of the original amino acid sequence are obtained. 76%, 78%, 80%, 90%, 91%, 92%, 93%, 94%, 95%, 96% It produces amino acid sequences that are 97%, 98%, or 99% similar or identical. The identity between the two amino acid sequences can be identified, preferably by the BLASTP algorithm. [BLAST Manual, Altschul, S., et al., NCBI NLM NIH Bethesda,Md.20894,Altschul,S.,et Use al., J.Mol.Biol.215:403-410(1990) It is determined by and

[0158] 3DC generated or identified by the implementation of the systems and methods discussed herein. Proteins manipulated with NN can be modified for translation. For example, the sigma discussed herein Post-translational modifications included within the scope of stem and method implementation include signal peptide cleavage. Glycosylation, acetylation, isoprenylation, proteolysis, myristoylation, tan Examples include protein folding and protein degradation. Alternatively, the processing event may require the introduction of additional biological machinery, such as signal peptides. Processing events such as cleavage and core glycosylation occur in the canine microsomal membrane or Xenop By adding US egg extract (US Patent No. 6,103,489) to the standard translation reaction... It will be inspected.

[0159] 3DC generated or identified by the implementation of the systems and methods discussed herein. Proteins manipulated with NN can acquire non-natural amino acids through post-translational modification or during translation. It may contain unnatural amino acids formed by their introduction. Various approaches are available for introducing amino acids. One example is suppressor - Special tRNAs, such as suppressor tRNAs, which possess specific characteristics, are site-specific. It is used in the process of natural non-natural amino acid substitution (SNAAR). In SNAAR, mRNA acts to target non-natural amino acids to specific sites during protein synthesis. A specific codon on the suppressor tRNA is required (as described in WO90 / 05785). (It is being done). However, suppressor tRNA is present in the protein translation system. It must not be recognizable by minoacyl-tRNA syngase. In certain cases, it is natural. It specifically modifies amino acids without significantly altering the functional activity of aminoacylated tRNA. Using a chemical reaction, after the tRNA molecule is aminoacylated, non-natural amino acids are formed. These reactions are called post-aminoacylation modifications. For example, the same tRNA ( tRNA LYSThe epsilon-amino group of lysine linked to ) has amine-specific photoaffinity It can be modified with a sign.

[0160] 3DC generated or identified by the implementation of the systems and methods discussed herein. Proteins manipulated with NN are used to prepare fusion proteins, along with other proteins. It may be conjugated with the molecule, for example, an N-terminal or C-terminal fusion protein. This may also be achieved by synthesis of protein, however, the resulting fusion protein may be 3DCNN The condition is that the functionality of the manipulated protein is preserved.

[0161] Protein mimetic manipulated with 3DCNN In some embodiments, the composition of interest is a protein peptide manipulated by 3DCNN. It is a peptide mimetic. Peptide mimetic is based on peptides and proteins, or similar substances. These are compounds derived from [these]. By implementing the systems and methods discussed herein, The peptide mimes that are formed or identified are typically non-natural amino acids, conformationally restricted, etc. Structural modifications of protein sequences manipulated by known 3D CNNs using electron distribution substitution, etc. This can be obtained. The target peptide mimetic has a structure between the peptide and the non-peptide synthetic structure. It constitutes a continuum in space, and therefore the peptide mimetic depicts the pharmacophore, and the parent Translating peptides into non-peptide compounds with protein activity manipulated by 3DCNN. It may be useful in helping to do so.

[0162] Furthermore, as is evident from this disclosure, the proteins manipulated by the target 3DCNN Mimetopes may be provided. Such peptide mimetic compounds must be non-hydrolyzable (e.g.) For example, the stability of the corresponding peptide against proteases or other physiological conditions. (Improvement of), improved specificity and / or potency, and intracellular localization of peptide mimetic bodies It may have attributes such as improved cell permeability. For illustrative purposes, the cis considered herein Peptide analogs produced or identified by the implementation of the system and method include, for example, benzo Diazepines (e.g., Freidinger et al. in Peptides:C) hemistry and biology, GR Marshall ed., ES See COM Publisher: Leiden, Netherlands, 1988. (to be done), substituted gama lactam ring ring)(Garvey et al.in Peptides:Chemistry and Biology,GRMarshall ed.,ESCOM Publ isher:Leiden,Netherlands,1988,p123), C-7 model Mimics (Huffman et al. in Peptides: Chemistry a nd Biology,GRMarshall ed.,ESCOM Publicis her: Leiden, Netherlands, 1988, p.105), Keto-Methi Len pseudopeptide (Ewenson et al. (1986) J Med Chem 2) 9:295, and Ewenson et al. in Peptides:Struc ture and Function(Proceedings of the 9th American Peptide Symposium)Pierce Chemi cal Co. Rockland, Ill., 1985), β-turned dipeptide core ( Nagai et al. (1985) Tetrahedron Lett 26:64 7, and Sato et al. (1986) J Chem Soc Perkin Trans 1:1231), β-amino alcohol (Gordon et al. (1 985)Biochem Biophys Res Commun 126:419, and Dann et al. (1986) Biochem Biophys Res C Ommun 134:71), diaminoketone (Natarajan et al. (1 984)Biochem Biophys Res Commun 124:141), Furthermore, methylene amino modification (Roark et al. in Peptides:Ch emistry and Biology, GR Marshall ed., ESC OM Publisher:Leiden,Netherlands,1988,p13 4) can be generated using. Also, generally, Session III: Analyt ic and synthetic methods,in Peptides: Chemistry and Biology,GR Marshall ed.,E SCOM Publisher: Leiden, Netherlands, 1988) Please refer to this.

[0163] Various methods can be used to generate protein peptide mimetic models manipulated with 3D CNNs. In addition to side chain substitution, the implementation of the systems and methods discussed herein involves peptide secondary The aim is to use conformationally constrained mimetic structures for the amide bond of peptides. Numerous substitutes have been developed. Substitutes frequently used for amide bonds include: The following groups: (i) trans-olefins, (ii) fluoroalkenes, (iii) methylene compounds Examples include (iv) phosphonamides and (v) sulfonamides.

[0164] nucleic acid In one embodiment, an implementation of the system and method discussed herein is used in 3DC Generate isolated nucleic acids containing nucleotide sequences encoding proteins manipulated with NN or It can be identified.

[0165] The nucleotide sequences encoding proteins manipulated by 3DCNN are, alternatively, obtained The polynucleotides are polynucleotides according to the implementation of the systems and methods discussed herein. Under the condition of encoding a peptide, a sequence mutation relative to the original nucleotide sequence, for example, It may include substitutions, insertions, and / or deletions of one or more nucleotides. This applies to implementations of the systems and methods discussed herein, as listed herein. The nucleotide sequence is virtually identical to that of the protein manipulated by 3DCNN. It may be possible to generate or identify a nucleotide sequence.

[0166] In the sense used herein, a nucleotide sequence is defined as a nucleotide sequence that is At least 60%, at least 70%, at least 85%, at least 95%, and at least 96%, at least 97%, at least 98%, or at least 99% of nucleoci If there is a degree of identity with respect to the nucleotide sequence, any of the nucleotide sequences described herein It is "substantially identical" to the other. Nuclei encoding proteins manipulated by 3DCNN Nucleotide sequences that are substantially homologous to the ocidal sequence are typically, for example, conserved substitutions. Alternatively, by introducing non-conservative substitutions, information contained in the nucleotide sequence can be used. Accordingly, the systems and methods discussed herein are generated or identified by their implementation. Polypeptide-producing organisms can be isolated from them. Other examples of possible modifications include in the sequence. Insertion of one or more nucleotides, one or more nucleotides at any of the ends of the sequence Addition of a nucleotide, or deletion of one or more nucleotides at any end or inside a sequence. Examples include the following. The identity between the two nucleotide sequences is preferably determined by the BLASTN algorithm. Rhythm [BLAST Manual, Altschul, S., et al., NCBI] NLM NIH Bethesda,Md.20894,Altschul,S.,e Use [t al., J.Mol.Biol.215:403-410(1990)] It is determined by [something].

[0167] In another embodiment, an implementation of the systems and methods discussed herein is used for 3DCN A construct comprising a nucleotide sequence encoding a protein or a derivative thereof manipulated with N. It may generate or identify. In certain embodiments, the construct is transcribed and optionally translated. The structure is operably coupled to the control element. The structure is the system and the structure considered herein. The expression of nucleotide sequences generated or identified by the implementation of the method is operably bound. A regulatory sequence can be incorporated, thus forming an expression cassette.

[0168] Proteins manipulated with 3DCNN or proteins manipulated with chimeric 3DCNN are They may be prepared using recombinant DNA methods. Therefore, the 3DCNN manipulated Nucleic acid molecules encoding proteins or proteins manipulated with a 3D CNN are 3D Good results for proteins manipulated with CNN or chimeric 3DCNN It may be incorporated into an appropriate expression vector to ensure reliable expression.

[0169] Therefore, in another embodiment, using an implementation of the systems and methods discussed herein The nuclei generated or identified by the implementation of the systems and methods discussed herein Vectors containing rheotide sequences or constructs can be generated or identified. Vector selection is The subsequent introduction of host cells depends on the specific embodiments described herein. The vectors generated or identified by the implementation of the system and method are expression vectors. Suitable host cells include a wide variety of prokaryotic and eukaryotic host cells. In certain embodiments, Expression vectors are derived from viral vectors, bacterial vectors, and mammalian cell vectors. Selected from the group. Prokaryotic and / or eukaryotic vector-based systems are specified herein. Use in conjunction with the implementation of the systems and methods discussed therein, polynucleotides or It can be used to produce homologous polypeptides. Many such systems are commercially available. It is also widely available.

[0170] Furthermore, the expression vector may be delivered to the cells in the form of a viral vector. Viruses that are useful as a result include retroviruses, adenoviruses, and adeno-related viruses. Examples include, but are not limited to, viruses, herpesviruses, and lentiviruses. Generally, a suitable vector is one that has a functional origin of replication in at least one organism, and a progenitor. Motor array, convenient restrictive endonuclease site, and one or more selectable markers - Contains (e.g., WO 01 / 96584, WO 01 / 29058, and US (See Japanese Patent No. 6,326,193).

[0171] Vectors suitable for polynucleotide insertion include expression vectors in prokaryotes, for example, p UC18, pUC19, Bluescript and its derivatives, mp18, mp19, pBR322, pMB9, ColE1, pCR1, RP4, phage, and "Shuttle "Vectors, e.g., pSA3 and pAT28; expression vectors in yeast, e.g., 2m Cron plasmid type vectors, embedded vectors, YEP vectors, centrome Aplasmids, etc.; expression vectors in insect cells, e.g., pAC series and pVL Vectors; expression vectors in plants, e.g., pIBI, pEarleyGate, pAV A, pCAMBIA, pGSA, pGWB, pMDC, pMY, pORE series, etc. Furthermore, expression vectors in eukaryotic cells based on viral vectors (adenovirus, adeno Virus-related viruses, such as retroviruses, and especially lentiviruses); In addition, non-viral vectors, such as pSilencer 4.1-CMV (Ambion ), pcDNA3, pcDNA3.1 / hyg, pHMCV / Zeo, pCR3.1, p EFI / His, pIND / GS, pRc / HCMV2, pSV40 / Zeo2, pTR ACER-HCMV, pUB6 / V5-His, pVAX1, pZeoSV2, pCI, Derived from pSVL, PKSV-10, pBPV-1, pML2d, and pTDT1 It is a vector.

[0172] As an example, a vector into which a nucleic acid sequence is introduced into a host cell is introduced into the host cell This may be a plasmid that is integrated into or not integrated into the genome. Nucleotide sequences generated or identified by the implementation of the systems and methods discussed. Alternatively, an exemplary and non-limiting example of a gene construct that may be inserted is in eukaryotic cells. Examples of tet-on inducible vectors for this purpose include...

[0173] In certain embodiments, the vector is a vector useful for transforming animal cells. .

[0174] Recombinant expression vectors can also be proteins manipulated with 3DCNNs or chimeric 3DCs. Increased expression of proteins manipulated by NN, or proteins manipulated by 3DCNN or The part that results in improved protein solubility when manipulated with a chimeric 3D CNN is encoded, By acting as a ligand for / or affinity purification, manipulated in 3DCNNs Nucleic acid molecules that assist in the purification of proteins or proteins manipulated by chimeric 3D CNNs. It may contain, for example, the protein degradation cleavage site is a protein manipulated by 3DCNN After insertion into the protein and purification of the fusion protein, 3D CNN is used to analyze the fusion protein region. Enables the isolation of manipulated proteins or proteins manipulated with chimeric 3D CNNs. To obtain. An example of a fusion expression vector is glutathione S-transferase (GST). The recombinant protein is fused with either maltose E-binding protein or protein A. pGEX (Amrad Corp., Melbourne, Australia) , pMAL (New England Biolabs, Beverly, Mass.) , and pRIT5 (Pharmacia, Piscataway, NJ) are examples. It can be done.

[0175] Additional promoter elements, or enhancers, regulate the frequency of transcription initiation. Typologically, these are located in the 30-110 bp region upstream of the initiation site, but many Recent studies have shown that motors also contain functional elements downstream of the initiation site. The spacing between motor elements is often flexible, and as a result, the promoter function is such that the elements It is retained when it reverses or moves relative to each other. Thymidine kinase (TK) promol In this case, the spacing between promoter elements increases until it reaches 50 bp before activity begins to decrease. It may be added. Depending on the promoter, the individual elements may be added either collaboratively or independently. It appears to function and be able to activate transcription.

[0176] The promoter is located upstream of the code segment and / or exon, and is a 5 noncode By isolating the sequence, a gene or polynucleotide sequence can be obtained. It may also be a promoter related to nature. Such promoters are "endogenous" and It can be called. Similarly, an enhancer is located either downstream or upstream of its sequence. It may be a naturally occurring enhancer of polynucleotide sequences, or it may have specific advantages. The coding polynucleotide segment is controlled by a recombinant promoter or a heterologous promoter. (This refers to promoters that are not typically associated with polynucleotide sequences in their natural environment.) Obtained by placing under the control of a recombinant enhancer or heterogeneous enhancer. Sir also has enhancers that are not normally associated with polynucleotide sequences in their natural environment. This refers to such promoters or enhancers of other genes. It is an enhancer, as well as an enhancer, that can be isolated from any other prokaryotic cell, viral cell, or eukaryotic cell. A promoter or enhancer that is not "naturally present," i.e., Promo containing mutations that alter the expression of different elements in different transcriptional regulatory regions and / or other elements. It may include a promoter or enhancer. Nucleic acid sequences of the promoter and enhancer. In addition to synthetically producing the sequence, the sequence is related to the compositions disclosed herein, Produced using nucleic acid amplification techniques including replacement cloning and / or PCR®. This is also acceptable (U.S. Patent Nos. 4,683,202, 5,928,906). This involves the transcription and / or expression of sequences within non-nuclear organelles such as mitochondria and chloroplasts. It is intended that a directional control array may also be used.

[0177] Naturally, the cell type, organelle, and organism selected for expression are important. Promoters and / or enhancers that effectively direct the expression of DNA segments It is important to use [this]. The promoters used are, for example, recombinant protein and Introduced DNA segments offer advantages in large-scale production of bi / peptides, etc. Under conditions suitable for directing high levels of expression, constitutive, tissue-specific, inducible, and It may be beneficial or useful. The promoter may be heterologous or endogenous.

[0178] The promoter sequences exemplified in the experimental examples presented herein are immediate early cytomegalo This is a virus (CMV) promoter sequence. This promoter sequence is operable with it. A potent ability to promote high levels of expression of any polynucleotide sequence linked to it. It is a constitutive promoter sequence. However, Simian virus 40 (SV40) is the first Promoter stage, mouse mammary tumor virus (MMTV), human immunodeficiency virus (HIV) ) Long terminal repeat (LTR) promoter, Moloney virus promoter, bird leukemia Virus promoter, Epstein-Barr virus immediate early promoter, Rouss meat tumor virus promoters, and human gene promoters, for example, and not limited to these. However, there are actin promoters, myosin promoters, hemoglobin promoters, and Other constitutive promoters include, but are not limited to, muscle creatine promoters. A vector sequence may also be used. Furthermore, the implementation of the systems and methods discussed herein is The use of constitutive promoters is not limited to this specification. Inducible promoters are also considered herein. It may be generated or identified through the implementation of the system and method that is perceived. The use of inducible promoters generated or identified through a method or method is prohibited. Whether to turn on the expression of a synchronously linked polynucleotide sequence when desired or provides a molecular switch that can turn off expression when expression is not desired. Examples of inductive promoters include the metallothione promoter and the glucocorticoid promoter. The progesterone promoter, progesterone promoter, and tetracycline promoter are These are, but are not limited to, the systems and methods discussed herein. The implementation of the law may allow the use of tissue-specific promoters, and such promoters may be It is active only in the desired tissue. As a tissue-specific promoter, HER-2 promoter is... Examples include, but are not limited to, motor and PSA-related promoter sequences.

[0179] In one embodiment, nucleic acid expression is controlled externally. For example, in one embodiment, This involves using the doxycycline Tet-On system or other inducible or repressive expression systems. It is controlled from the outside.

[0180] Recombinant expression vectors also allow selection of transformed or transfected host cells. It may contain selectable marker genes that promote the process. Suitable selectable marker genes G418 and hygromycin, β-galactone, which confuse resistance to specific drugs. Sidase, chloramphenicol acetyltransferase, firefly luciferase, or immunoglobulin or a portion thereof, for example, immunoglobulin, preferably IgG These are genes that code for proteins such as the Fc region. Selectable markers are the target... The nucleic acid may be introduced onto a vector other than the one being introduced.

[0181] The reporter gene identifies potentially transfected cells and the regulatory sequence It is used to evaluate functionality. Generally, the reporter gene is used in the recipient organism. or are not present within the organization, or are expressed by them, and such expression is, for example, Encoding proteins that are characterized by some easily detectable property, such as enzyme activity. It is a gene. Reporter gene expression occurs after the DNA has been introduced into the recipient cell. The assay is performed at an appropriate time.

[0182] Exemplary reporter genes include luciferase, beta-galactosidase, and chloram. Phenicol acetyltransferase, secretory alkaline phosphatase, or green Genes encoding fluorescent proteins, including but not limited to fluorescent protein genes. This may include (for example, Ui-Tei et al., 2000 FEBS Lett.4) See 79:79-82.

[0183] In one embodiment, the system and method described herein are generated or The identified 3DCNN-manipulated protein is a reporter gene and is in favorable expression. It is included in the system. For example, in one embodiment, it is generated by such a system or method. Alternatively, proteins manipulated with 3DCNNs that are identified show increased fluorescent activity in blue fluorescent proteins. It is a protein. In such embodiments, the systems and methods discussed herein Nuclei encoding proteins manipulated by 3DCNNs, which are generated or identified by the device. Otid sequences may be incorporated into expression systems to enable the detection of heterologous protein sequences. stomach.

[0184] Recombinant expression vectors may be introduced into host cells to produce recombinant cells. These may be prokaryotic or eukaryotic cells. Implementation of the systems and methods discussed herein. Using vectors generated or identified by, for example, eukaryotic cells, for example, yeast cells Cells, Saccharomyces cerevisiae, or mammalian cells, for example Epithelial kidney 293 cells or U2OS cells, or prokaryotic cells, such as bacteria, Esch Transform erichia coli or Bacillus subtilis Nucleic acids can be precipitated by coprecipitation of calcium phosphate or calcium chloride, DEAE- Kistran-mediated transfection, lipofectin, electroporation, or microin It can be introduced into cells using conventional techniques such as injection. The host cells are transformed... A preferred method for transfecting is described by Sambrook et al. (Mo Lecular Cloning: A Laboratory Manual, 2nd Edition,Cold Spring Harbor Laboratory pr This can be found in ess (1989) and other experimental texts.

[0185] For example, generated or identified by the implementation of the systems and methods discussed herein Proteins manipulated by 3DCNNs or proteins manipulated by chimeric 3DCNNs are bacterial cells, e.g., E. coli, insect cells (using baculovirus), yeast cells It can be expressed in cells or mammalian cells. Other suitable host cells include Goeddel, Gen. e Expression Technology:Methods in Enzym ology 185, Academic Press, San Diego, Calif. It can be seen in (1991).

[0186] Modified blue fluorescent protein In one embodiment, using an implementation of the system and method discussed herein, sec BFP2 variant proteins may be identified or generated. In certain embodiments, the composition may be stable Regarding secBFP2 variant proteins containing one or more mutations that enhance qualitative analysis. In a certain manner, the secBFP2 variant protein is compared to wild-type secBFP2. It offers enhanced stability, enhanced fluorescence, enhanced half-life, and slower photofading. Shows one or more of the following.

[0187] In some embodiments, the secBFP2 variant protein has one or more mutations. Includes secBFP2. For example, in some embodiments, secBFP2 varian The protein is associated with full-length wild-type secBFP2, with T18, S28, Y96, S One of the following to choose from: 114, V124, T127, D151, N173, and R198 The secBFP2 contains one or more mutations in the above residues. In one embodiment, complete Nagano wild-type secBFP2 is Contains the amino acid sequence of JPEG0007933942000003.jpg40170.

[0188] In certain embodiments, within the secBFP2 variant protein described herein The mutation is notated in relation to sequence number 1. For example, secBFP containing a mutation in T18. The 2 variant protein refers to secBFP2, but the full-length wild-type secBFP2 ( The mutation is present in the residue corresponding to threonine at position 18 of column 1).

[0189] In some embodiments, the secBFP2 variant protein is the full-length wild-type se In relation to cBFP2, T18X, S28X, Y96X, S114X, V124X, T1 27X, D151X, N173X, and R198X (where X is any amino acid) Includes secBFP2 containing one or more mutations selected from (a) So, the secBFP2 variant protein is related to the full-length wild-type secBFP2. , T18W, T18V, T18E, S28A, Y96F, S114V, S114T, V1 24T, V124Y, V124W, T127P, T127L, T127R, T127D, D151G, N173T, N173H, N173R, N173S, R198V, and R Contains secBFP2 with one or more mutations selected from 198L.

[0190] In one embodiment, the secBFP2 variant protein is a T18X mutation (where X is X). The secBFP2 contains any amino acid. In one embodiment, secBF The P2 variant protein, in association with the full-length wild-type secBFP2, is the T18W mutation. Contains secBFP2 with the T18V mutation or the T18E mutation.

[0191] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000004.jpg40170, or its variants or fragments.

[0192] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000005.jpg43170, or its variants or fragments.

[0193] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000006.jpg41170, or its variants or fragments.

[0194] In one embodiment, the secBFP2 variant protein is an S28X mutation (where X is X). The secBFP2 contains any amino acid. In one embodiment, secBF The P2 variant protein, in association with the full-length wild-type secBFP2, exhibits the S28A mutation. Includes secBFP2.

[0195] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000007.jpg40170, or its variants or fragments.

[0196] In one embodiment, the secBFP2 variant protein is a T96X mutation (where X is X). The secBFP2 contains any amino acid. In one embodiment, secBF The P2 variant protein, in association with the full-length wild-type secBFP2, exhibits the Y96F mutation. Includes secBFP2.

[0197] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000008.jpg41170, or its variants or fragments.

[0198] In one embodiment, the secBFP2 variant protein is the S114X mutation (where, X is any amino acid) and contains secBFP2. In one embodiment, secB The FP2 variant protein, in association with the full-length wild-type secBFP2, exhibits the S114V mutation. Contains secBFP2 with different mutations or S114T mutations.

[0199] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000009.jpg41170, or its variants or fragments.

[0200] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000010.jpg41170, or its variants or fragments.

[0201] In one embodiment, the secBFP2 variant protein is a V124X mutation (where, X is any amino acid) and contains secBFP2. In one embodiment, secB The FP2 variant protein, in association with the full-length wild-type secBFP2, exhibits the V124T mutation. Contains secBFP2 with variants, V124Y mutations, or V124W mutations.

[0202] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000011.jpg40170, or its variants or fragments.

[0203] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000012.jpg42170, or its variants or fragments.

[0204] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000013.jpg41170, or its variants or fragments.

[0205] In one embodiment, the secBFP2 variant protein is a T127X mutation (where, X is any amino acid) and contains secBFP2. In one embodiment, secB The FP2 variant protein, in association with the full-length wild-type secBFP2, undergoes the T127P mutation. Contains secBFP2 with variants, T127L mutation, T127R mutation, or T127D mutation. nothing.

[0206] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000014.jpg41170, or its variants or fragments.

[0207] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000015.jpg41170, or its variants or fragments.

[0208] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000016.jpg42170, or its variants or fragments.

[0209] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000017.jpg40170, or its variants or fragments.

[0210] In one embodiment, the secBFP2 variant protein is a D151X mutation (where, X is any amino acid) and contains secBFP2. In one embodiment, secB The FP2 variant protein, in association with the full-length wild-type secBFP2, exhibits a D151G mutation. Includes secBFP2 with variations.

[0211] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000018.jpg41170, or its variants or fragments.

[0212] In one embodiment, the secBFP2 variant protein is the N173X mutation (where, X is any amino acid) and contains secBFP2. In one embodiment, secB The FP2 variant protein, in association with the full-length wild-type secBFP2, exhibits the N173T mutation. Includes secBFP2 containing variants, N173H mutations, N173R mutations, or N173S mutations. nothing.

[0213] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000019.jpg41170, or its variants or fragments.

[0214] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000020.jpg41170, or its variants or fragments.

[0215] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000021.jpg41170, or its variants or fragments.

[0216] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000022.jpg38170, or its variants or fragments.

[0217] In one embodiment, the secBFP2 variant protein has the R198X mutation (where, X is any amino acid) and contains secBFP2. In one embodiment, secB The FP2 variant protein, in association with the full-length wild-type secBFP2, exhibits the R198V mutation. Contains secBFP2 with different or R198L mutations.

[0218] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000023.jpg41170, or its variants or fragments.

[0219] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000024.jpg40170, or its variants or fragments.

[0220] In one embodiment, the secBFP2 variant protein is a full-length wild-type secBFP In relation to 2, T18X, S28X, Y96X, S114X, V124X, T127X, The D151X, N173X, and R198X mutations (where X is any amino acid) One or more of the following: two or more, three or more, four or more, five or more, six or more, seven or more , including secBFP2 which includes eight or more, or all nine. In one embodiment, secB FP2 variant proteins are associated with full-length wild-type secBFP2 as T18W, T 18V, T18E, S28A, Y96F, S114V, S114T, V124T, V12 4Y, V124W, T127P, T127L, T127R, T127D, D151G, N Of the following: 173T, N173H, N173R, N173S, R198V, and R198L One or more, two or more, three or more, four or more, five or more, six or more, seven or more, eight or more or including secBFP2 with 9 or more entries.

[0221] In one embodiment, the secBFP2 variant protein is a full-length wild-type secBFP Related to 2, T18X, S28X, S114X, V124X, T127X, D151X This includes mutations N173X and R198X (where X is any amino acid). Contains secBFP2. In one embodiment, the secBFP2 variant protein is complete In relation to the Nagano wild-type secBFP2, T18W, S28A, S114V, V124T, T Includes secBFP2 with mutations 127P, D151G, N173T, and R198L. nothing.

[0222] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000025.jpg41170, or its variants or fragments.

[0223] In one embodiment, the secBFP2 variant protein is a full-length wild-type secBFP In relation to 2, S28X, S114X, T127X, and N173X (where X is It contains secBFP2 containing a mutation of any amino acid. In one embodiment, secB FP2 variant proteins are associated with full-length wild-type secBFP2 as S28A, S Includes secBFP2 with mutations 114T, T127L, and N173H.

[0224] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000026.jpg40170, or its variants or fragments.

[0225] In one embodiment, the secBFP2 variant protein is a full-length wild-type secBFP In relation to 2, the mutations of S28X and S114X (where X is any amino acid) Contains secBFP2 with variants. In one embodiment, secBFP2 variant protein This includes mutations in S28A and S114T, associated with the full-length wild-type secBFP2. Includes ecBFP2.

[0226] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000027.jpg41170, or its variants or fragments.

[0227] In one embodiment, the secBFP2 variant protein is the full-length wild-type secBFP2 In relation to this, S28X, S114X, and N173X (where X is any amino acid) It includes secBFP2 containing a mutation of (which is). In one embodiment, secBFP2 varian The protein, in association with the full-length wild-type secBFP2, is S28A, S114T, and It contains secBFP2 with the N173H mutation.

[0228] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000028.jpg41170, or its variants or fragments.

[0229] In one embodiment, the secBFP2 variant protein is a full-length wild-type secBFP In relation to 2, S28X, Y96X, S114X, and N173X (where X is r It contains secBFP2 which includes a mutation of the amino acid (which is of interest). In one embodiment, secBF The P2 variant protein is associated with the full-length wild-type secBFP2 as S28A, Y9 Includes secBFP2 with mutations 6F, S114T, and N173H.

[0230] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000029.jpg41170, or its variants or fragments.

[0231] In one embodiment, the secBFP2 variant protein is a full-length wild-type secBFP In relation to 2, S28X, Y96X, S114X, T127X, and N173X (here In one embodiment, secBFP2 includes a mutation in X (where X is any amino acid). The secBFP2 variant protein, in relation to the full-length wild-type secBFP2, is S secBF containing mutations 28A, Y96F, S114T, T127L, and N173H Includes P2.

[0232] For example, in one embodiment, the secBFP2 variant protein is Includes JPEG0007933942000030.jpg40170, or its variants or fragments.

[0233] In one embodiment, the system and method described herein are generated or The identified composition is a nucleotide compound encoding the secBFP2 variant protein. The molecule contains isolated nucleic acid molecules including sequences 2 to 2. A sequence encoding at least one amino acid sequence described in 28, or its variant Includes parts or fragments.

[0234] A fluorescent protein variant operably linked to one or more target polypeptides. A fusion protein containing the peptide bond is also provided. The polypeptide of the fusion protein contains the peptide bond The fluorescent protein variant can be linked via the linker molecule, or it can target via the linker molecule. It can be linked to polypeptides that become one or more pairs. In one embodiment, the fusion protein is one or more pairs One or more polynucleotides operably linked to the polypeptide that constitutes the elephant Recombinant nucleic acid molecules containing polynucleotides encoding fluorescent protein variants It is expressed from there.

[0235] The target polypeptides include, for example, peptide tags such as polyhistidine peptides, or cellular polypeptides such as enzymes, G proteins, growth factor receptors, or transcription factors It may include any polypeptide, and can associate to form a complex. It may be one of two or more proteins. In one embodiment, the fusion protein is This is a tandem fluorescent protein variant construct, which is a donor fluorescent protein variant construct. Ant, acceptor fluorescent protein variant, and the donor and the acceptor It contains a peptide linker moiety that binds to the donor, and the cyclized amino acid of the donor has the characteristics of the donor. A certain light is emitted, and the donor and acceptor emit fluorescence resonance energy when the donor is excited. It exhibits gee migration, and the linker portion emits virtually no light to excite the donor. Therefore, the systems and methods discussed herein are generated or identified by their implementation. The fusion protein is operably linked to two or more proteins that can be linked directly or indirectly. It may contain a fluorescent protein variant and further contain one or more target polypeptides. obtain.

[0236] kit In some implementations, the kit is used in the implementation of the systems and methods discussed herein. To promote and / or standardize the use of the compositions provided or identified, and These various methods may be provided to facilitate the methods discussed herein. The materials and reagents required for implementation may be provided in a kit to facilitate the execution of the method. As used herein, the term "kit" refers to a process, assay, analysis, and This term is used in relation to a combination of items that facilitate an operation.

[0237] The kit includes chemical reagents (e.g., polypeptides or polynucleotides), as well as other It may include components. Furthermore, the kits discussed herein may also include, for example, sample collection. Apparatus and reagents for product collection and / or purification, for product collection and / or purification Apparatus and reagents, reagents for bacterial cell transformation, for eukaryotic cell transfection Reagents, already transformed or transfected host cells, sample tubes, holders, Trays, racks, dishes, plates, instructions for kit users, solutions, buffer solutions or other chemicals Includes suitable samples used as reagents, standardization, normalization, and / or control samples. The kit can also, for example, provide convenient storage and safety. For shipping purposes, it may be packaged in a box with a lid.

[0238] In some embodiments, for example, the kits discussed herein are considered herein. Fluorescent proteins generated or identified by the implementation of the system and method described herein The fluorescent proteins produced or identified by implementing the systems and methods discussed are A polynucleotide vector encoding a vector (e.g., plasmid), suitable for vector propagation. We can provide bacterial cell lines and reagents for purifying expressed fusion proteins. In some embodiments, the kits discussed herein have a tendency to oligomerize. Anthozoan fluorescent protein to generate reduced protein variants We can provide the reagents necessary for inducing mutations.

[0239] The kit is generated or identified by the implementation of the systems and methods discussed herein. One or more compositions, for example, one or more fireflies that may be part of a fusion protein One or more polynuclei encoding a photoprotein variant or polypeptide. May contain rheotides. Fluorescent protein variants are oligomerized monomers such as non-oligomerized monomers. It may be a mutant fluorescent protein with a reduced tendency to smear, or a tandem dimer fluorescent protein. It may be a photoprotein, and the kit contains multiple fluorescent protein variants, several of which are multiple mutant fluorescent protein variants, or multiple tandem dimer fluorescent proteins , or a combination thereof.

[0240] The kits discussed herein may also contain one or more recombinant nucleic acid molecules. This partially encodes a fluorescent protein variant that may be the same or different. For example, restriction endonuclease recognition sites or recombinase recognition sites, or A second poly, operably linked, containing or encoding the polypeptide of interest. It may also contain nucleotides. Furthermore, the kit contains the components of the kit, in particular the kit Produced or identified by the implementation of the systems and methods discussed herein. The composition may include instructions for use.

[0241] Those skilled in the art can conveniently select one or more proteins having desirable fluorescent properties for a particular application. Therefore, such kits can contain multiple different fluorescent protein variants. This can be particularly useful when providing different fluorescent protein variants. Kits containing multiple polynucleotides offer numerous advantages. For example, polynucleotides Cleotides contain convenient restriction endonuclease or recombinase recognition sites. It is manipulated in such a way that it codes for a modifier or the target polypeptide. Operable linkage of polynucleotides to polynucleotides, or, if desired, fluorescent tan Linking two or more polynucleotides that encode a protein variant into each other in a operable manner. This can facilitate the process.

[0242] Use of fluorescent protein variants The fluorescent light generated or identified by the implementation of the systems and methods discussed herein Protein variants are useful in any method using fluorescent proteins. Therefore, Fluorescent protein variants including monomers, dimers, and tandem-dimer fluorescent proteins For example, in detection assays such as immunoassays or hybridization assays. For use, or to track the movement of proteins in cells, fluorescent protein This involves binding the riant to an antibody, polynucleotide, or other receptor, and using a fluorescent lens. The marker is useful as a fluorescent marker in many methods already in use. For trace studies, the first (or other) polynucleus encoding the fluorescent protein variant The rheotide is attached to the second (or other) polynucleotide that codes for the target protein. The fused construct can, if desired, be inserted into an expression vector. When expressed in cells, The target protein is one in which the localization of the fusion protein is determined by the fluorescent protein composition of the protein. Without worrying that it is an artifact caused by oligomerization of the fluorescence It can be localized based on this. In one embodiment of this method, the two target proteins are It is independently fused with two fluorescent protein variants that have different fluorescence properties.

[0243] The fluorescent light generated or identified by the implementation of the systems and methods discussed herein Protein variants are useful in systems for detecting transcriptional induction. For example, non-oligonucleotides. Nucleotides encoding merized monomers, dimers, or tandem-dimer fluorescent proteins The sequence may be contained in the current vector in relation to the target promoter or other expression regulatory sequence. The constructs can be fused, transfected into cells, and used as promoters (or other conditioned elements). The induction of the nodal element is measured by detecting the presence or amount of fluorescence, thereby This method makes it possible to observe the responsiveness of the signaling pathway from receptor to promoter. It is possible.

[0244] The fluorescent light generated or identified by the implementation of the systems and methods discussed herein Protein variants are also useful in applications involving FRET, which is a fluorescent donor. The event is detected as a function of the mutual or receding motion of the acceptors. It is possible. One or both of the donor / acceptor pair can be fluorescent protein markers. It could be an ant. Such a donor / acceptor pair would have an excitation peak and emission of the donor. It provides broad separation between peaks, and the donor emission spectrum and acceptor excitation spectrum. It provides good overlap with the Ru.

[0245] Using FRET, the donor and acceptor were bound to the substrate on both sides of the cleavage site. It can detect substrate cleavage. When substrate cleavage occurs, the donor / acceptor pair is formed. Theoretically separates and eliminates FRET. Such assays, for example, involve separating the substrate from the sample. This is carried out by bringing the samples into contact and determining the qualitative or quantitative changes in FRET. (For example, U.S. Patent No. 5,741,657 incorporated herein by reference) (See reference). Fluorescent protein variant donor / acceptor pairs are proteins. It may be part of a fusion protein bound by a peptide that has a cleavage site (for example) See, for reference, U.S. Patent No. 5,981,200, incorporated herein by reference. FRET can also be used to detect changes in potential across membranes. For example, donor - and acceptors are positioned on both sides of the membrane so as to move across the membrane in response to voltage changes. It is placed there, thereby producing a measurable FRET (for example, see reference) See U.S. Patent No. 5,661,035 incorporated herein.

[0246] In other embodiments, the systems and methods discussed herein are generated by the implementation of the systems and methods described herein. The fluorescent proteins identified are those that exhibit fluorescent activity of protein kinases and phosphatases. Nsa, or Ca 2+ Zn 2+ , cyclic 3′,5′-adenosine monophosphate, and cyclic 3 To create indicators for small ions and molecules such as ',5'-guanosine monophosphate. It is useful for [something].

[0247] The fluorescence in a sample is generally measured using a fluorophotometer, and excitation with a first wavelength is used. Excitation radiation from the source passes through the excitation optical system, thereby exciting the sample. In response, the fluorescent protein variant in the sample emits light at a wavelength different from the excitation wavelength. It emits radiation having [a certain characteristic]. Then, the collection optical system collects the light emitted from the sample. The chair is a temperature controller to maintain the sample at a specific temperature while it is being scanned. Multiple samples are held to position different wells that may contain and be exposed to the sample. It may have a multi-axis moving stage for moving the microtiter plate. Multi-axis moving stage, temperature controller, autofocus function, and electric power related to data collection. The child device may be managed by a properly programmed digital computer, digital The computer also separates the data collected during the assay for presentation purposes. It can be converted to the following format. This process has been miniaturized and automated, and is high-throughput. This formula could enable screening thousands of compounds. One way to implement the principle is Lakowicz, "Principles of Fluorescence Spectroscopy”(Plenum Pre ss 1983), Herman, “Resonance energy transf. er microscopy”In“Fluorescence Microscopy of Living Cells in Culture”Part B,Meth. Cell Biol.30:219-243(ed.Taylor and Wang; Academic Press 1989), Turro, “Modern Molec ular Photochemistry”(Benjamin / Cummings P ubl.Co.,Inc.1978), pp.296-361 are cited, and each of these This is incorporated herein by reference.

[0248] Therefore, this disclosure also proposes the implementation of a method for identifying the presence of molecules in a sample. Such methods are provided, for example, in the implementation of the systems and methods considered herein. Therefore, the fluorescent protein variants that are generated or identified are linked to molecules, To detect fluorescence from fluorescent protein variants in samples suspected of containing [unclear]. This can be carried out by and . The molecules to be detected are polypeptides, polynucleotides, and This could be any other molecule, including, for example, an antibody, an enzyme, or a receptor, or a fluorescent protein. The variant may be a tandem dimer fluorescent protein.

[0249] The samples to be tested are biological samples, environmental samples, or samples containing specific molecules. Any sample, including any other sample whose existence is to be determined. This is possible. Preferably, the sample contains cells or an extract thereof. The cells may be human, etc. It can be obtained from vertebrates, including mammals, or from invertebrates, plants or animals. It can be a cell. The cell can be obtained from such a cell, for example, from a cell line culture, or They can be isolated from living organisms. Therefore, cells can be contained in tissue samples, and this , by any means commonly used to obtain tissue samples, for example, human life It can be obtained from living organisms by examination. The method may involve intact living cells or newly isolated tissue. Alternatively, if performed using organ samples, the presence of the target molecule in living cells can be identified. Therefore, for example, it is possible to provide a means for determining the intracellular compartmentalization of molecules. This is possible. Implementation of the systems and methods discussed herein for such purposes. The use of fluorescent protein variants generated or identified by fluorescent protein oligomers In that the possibility of abnormal identification or localization due to chromatography is significantly minimized, it is substantial. It offers advantages.

[0250] Fluorescent protein variants are stable under conditions in which the protein-molecular complex is exposed. It can be linked to the molecule directly or indirectly using any linkage. Therefore, fluorescence Proteins and molecules react through chemical reactions between reactive groups present on proteins and molecules. They can be linked, or the linkage contains reactive groups specific to the fluorescent protein and molecule. It can be mediated by the linker moiety. It links fluorescent protein variants and molecules. The appropriate conditions for this are selected, for example, depending on the chemical properties of the molecule and the desired type of linkage. It will be recognized that this will be done. If the target molecule is a polypeptide, fluorescent tan A convenient means for linking protein variants and molecules is to link them, for example, by polyp A tandem dimer fluorescent fluorite operably linked to a polynucleotide encoding a butyl molecule. Fusion proteins from recombinant nucleic acid molecules, including polynucleotides that encode proteins. This is due to the expression of the phenomenon.

[0251] Methods are also provided for identifying drugs or conditions that modulate the activity of expression regulatory sequences. One such method involves, for example, using a fluorescent protein variant operably linked to an expression regulatory sequence. Recombinant nucleic acid molecules containing polynucleotides that control expression are used to extract polynucleotides from the expression control sequence. Exposure to drugs or conditions suspected of being able to regulate the expression of the drug, and This can be carried out by detecting the fluorescence of fluorescent protein variants due to such exposure. Such methods include, for example, cellular factors involved in tissue-specific expression from regulatory elements. Furthermore, it includes cellular proteins that can regulate expression from expression regulatory sequences, chemically and This is useful for identifying biological drugs. Therefore, expression regulatory sequences are promoters. enhancer, silencer, intron splicing recognition site, polyadenylated part These could be transcriptional regulatory elements such as ribosome junctions, or translational regulatory elements such as ribosome binding sites.

[0252] The fluorescent light generated or identified by the implementation of the systems and methods discussed herein Protein variants also identify specific interactions between the first and second molecules. It is also useful in law. Such methods include, for example, the specific phases of the first and second molecules. Under conditions that enable interaction, the first fluorescent protein variant of the donor is linked to the second fluorescent protein variant. Molecule 1 is linked to a second molecule which is attached to a second fluorescent protein variant of the acceptor. Contact, excitation of the donor, and fluorescence or emission from the donor to the acceptor. Resonance energy transfer is detected, thereby enabling the specific interaction between the first and second molecules. This can be done by identifying the action. The conditions for such interactions are specific to the molecule. It may be possible under any conditions under which interaction is predicted or suspected. In particular, when the molecule being tested is a cellular molecule, the conditions are generally physiological. Therefore, the method uses conditions such as buffer, pH, and ionic strength that mimic physiological conditions. The method may be performed in vitro using cells or cell extracts. It can be implemented using [this method].

[0253] Luminescence resonance energy transfer is chemiluminescence, bioluminescence, lanthanides, or transition metal donors. This involves energy transfer from the red fluorescent protein moiety to the longer portion of the red fluorescent protein. The excitation wavelength is available from a wider variety of donors than is possible with green fluorescent protein variants, and It enables energy transfer over larger distances. Also, longer emission wavelengths... It is detected more efficiently by solid-state photodetectors, and red light is detected much better at shorter wavelengths. It is particularly beneficial for in vivo applications that penetrate tissue. As a chemiluminescent donor, lumin This includes ol derivatives and peroxyoxalate systems. As bioluminescent donors, Quorin, Oberin, Firefly Luciferase, Sea Mushroom Luciferase, Bacterial Lucifer This includes, but is not limited to, lanthanides, lases, and their variants. As a donor, it is linked to multiple ligand groups to protect the metal ion from the solvent water. Examples include, but are not limited to, terbium chelates containing a UV-absorbing sensitizer chromophore. It is not possible. As transition metal donors, the oligopyridine ligands ruthenium and osmium Muchilates are examples, but are not limited to these. Chemiluminescent and bioluminescent donors are, While it does not require excitation light, it is excited by the addition of a substrate, whereas metal-based systems are excited It requires light activation but offers a longer excited state lifetime and eliminates the undesirable background noise of fireflies. This facilitates time-gated detection for distinguishing between light and scattering.

[0254] The first and second molecules are used to determine whether the proteins interact specifically. It could be a cellular protein being investigated to confirm such an interaction. Such first and second cellular proteins, for example, possess the ability to oligomerize. It could be the same as when they are being tested, or they could be proteins, for example, This may differ from cases where it is being tested as a specific binding partner involved in intracellular pathways. Molecules 1 and 2 are also polynucleotides and polypeptides, for example, known ones. Polynucleotides that have been tested for transcriptional regulatory element activity, and known, or It may be a polypeptide being tested for transcription factor activity. For example, the first molecule is Transcriptional regulatory element activity may be tested, and may be random or have known sequence variations. It may contain multiple nucleotide sequences that could be ants, and the second molecule may be a transcription factor. Such methods are useful for identifying novel transcriptional regulators with desirable activity. .

[0255] This disclosure also provides an implementation of a method for determining whether a sample contains an enzyme. Such methods include, for example, using the system and method discussed herein to sample the sample. Contact with tandem fluorescent protein variants generated or identified by the implementation of the law. The tasks are to excite the donor and to determine the fluorescence properties in the sample, The presence of enzymes in the pull determines whether it leads to a change in the degree of fluorescence resonance energy transfer. This can be carried out by [method]. Similarly, this disclosure relates to a method for determining the activity of an enzyme in a cell. This provides an implementation of the following: Tandem fluorescent protein variant construction. The present invention provides cells that express a substance, wherein the peptide linker portion is a donor and acceptor To provide a donor containing a cleavage recognition amino acid sequence specific to the enzyme that binds the donor, and Exciting and determining the degree of fluorescence resonance energy transfer in the cell, The presence of enzyme activity determines whether it leads to a change in the degree of fluorescence resonance energy transfer. It can be implemented by [this method]. [Examples]

[0256] The implementation of the systems and methods discussed herein can be found in the following experimental examples. This is further detailed. These examples are provided for illustrative purposes only and are not specified otherwise. Unless otherwise specified, this is not intended to be limiting. Therefore, the following are considered in this specification. The system and method should not be construed as being limited to the following embodiments. However, all sorts of changes become apparent as a result of the teachings provided herein. It should be interpreted as encompassing form and shape.

[0257] Without further explanation, those skilled in the art will use the foregoing description and the following exemplary examples to describe this It is believed that implementations of the systems and methods discussed in the specification can be created and utilized. Therefore, the following examples illustrate the systems and methods discussed herein. This specifically points out certain embodiments and does not in any way limit the remainder of this disclosure. It should not be interpreted as such.

[0258] Example 1: Protein engineering using neural networks To empirically validate neural networks, we used three different model proteins. Selected, each represents a distinct protein engineering challenge. The first validation model protein is: It is a tem-1 beta-lactamase, and the main reason for this is that 1) its sensitivity to antibiotics is , because it is directly related to the overall stability of that protein, and 2) that protein However, both stabilizing and destabilizing mutations have been well characterized. Therefore, in order to incorporate the non-standard amino acid L-DOPA, metalloprotein Stability was improved by reusing crystalline phosphomannose isomerase as the reporter. However, the enzyme's insufficient stability prevents its use as a reporter. The last protein example is the improvement of the blue fluorescent protein variant, secBFP2. Yes, there are. Blue fluorescent proteins are well characterized, but they exhibit rapid photobleaching, and slow photobleaching. The maturation and folding process, as well as the relatively faint fluorescence, hinders its wider use. .

[0259] First, the wild-type amino acid is experimentally verified as the best residue at that position. By dividing the analysis into these parts, we evaluated the true negative rate of the neural network. The effect of each individual amino acid change was quantified using the fitness of the organism, tem-1 β-lactamers. The test was conducted using Ze's already published mutation scan. 263 were tested with tem-1. Of the locations, 136 sites had relative fitness values ​​of less than zero (i.e., the biological fitness of...) (Areas that could not withstand mutations that were far from the wild-type residue without loss of responsiveness) This collection of 136 sites represents a complete collection of true negatives in tem-1 beta-lactamase. This constructs the true negative sensitivity for each individual change made to the neural network. The benchmark was used. The final version accurately identified 92.6% of 136 true negatives. In fact, this represented an increase of nearly 30% compared to the initial model. Therefore, the developed model This has increased the ability to identify sites within proteins that are not suitable for mutation.

[0260] The experimental results are shown in Figures 3A and 3B. Figure 3A shows the improvement of BFP fluorescence. This bar graph shows the regions predicted by the neural network, and the degree to which they were predicted. The bar 301 on the far right represents the wild-type tactic, each individually suggested by a neural network. Fluorescence observed by performing specific combinations of amino acid substitutions in proteins It shows improvement. A visual representation of the improvement is shown in Figure 3B. Modified blue fluorescent protein 302 is It shines much brighter than wild-type blue fluorescent protein 303.

[0261] Additional results are shown in Figures 4A and 4B. The bar graph in Figure 4B shows phosphomannur This paper demonstrates an improvement to the neural network proposal for suisomerase (PMI). Each stenosing mutation results in a 15% to 50% increase compared to the wild type, but they are used in combination. In this case (bar 401), the improvement is additive and shows a significant improvement in stability of nearly 600%. Rasu.

[0262] Venn diagrams 411 (blue fluorescent protein, pdb:3m24) and 412 (phosphorus) in Figure 4B. Homannose isomerase (pdb:1pmi) is used by neural networks to interact with other cells. Computer-based protein stabilization technology (Foldx PositionScan and Predicts unique residue candidates not identified by Rosetta pmut scan. This indicates that.

[0263] Figure 5 shows the TEM-1 β-lactamasebe identified by the neural network. The riant enables E. coli growth at higher ampicillin concentrations than the ancestral protein. This indicates that the β-lactamase mutants N52K and F60Y were mutagenerated individually. E. coli expressing M182T, E197D, or A249V is each 125ug. It can grow at ampicillin concentrations of 1 / mL or higher, and at that concentration, it is labeled "WT". E. coli expressing the attached ancestral enzyme were unable to grow. E. coli expressing a single enzyme variant containing all five mutations (N52K, F 60Y, M182T, E197D, and A249V (labeled "All") It was able to grow at an ampicillin concentration of 3000 ug / mL. In other words, The neural network is related to the phenotype of the catalyst, in this embodiment, E. coli is antibiotic The phenotype was improved to enable greater resistance to the substance ampicillin. .

[0264] Figure 6 shows that the neural network improved the thermal stability of the blue fluorescent protein. As shown, in one example, after 10 minutes of heat loading, residual fluorescence is observed in the inducible protein Bluebo The amount of SecBFP2.1, the ancestral protein, was less than that of nnet. The extracted blue fluorescent protein was diluted to 0.01 mg / mL in PBS pH 7.4, and 100 µg was added. Aliquots of L are placed on a thermal gradient using a thermal cycler, and then on a PCR strip. The samples were heat-treated for 10 minutes. Fluorescence of the thermally loaded variants and incubation at room temperature were performed. The control was then subjected to excitation and emission wavelengths of 402 nm and 457 nm, respectively. The assay was performed. The fluorescence readings were normalized to the mean of the solution incubated at room temperature. (For example, a measurement of 0.8 means that the heat-treated protein has 80% of the fluorescence of its untreated counterpart.) (This indicates that it held) As shown in Figure 6, Bluebonnet was approximately 84 Higher thermal stability compared to SecBFP2.1 across the entire temperature range from °C to approximately 100°C. It exhibits properties such as fluorescence maintained by the ancestral protein after a 10-minute heat load at 100°C. If not treated, it retained more than 20% of the untreated fluorescence.

[0265] Figure 7 shows that the neural network improved the chemical stability of the blue fluorescent protein. This demonstrates that the fluorescence half-life in guanidine molten material is that of the inducible protein. Bluebonnet has less information on its ancestral protein, SecBFP2.1. The purified blue fluorescent protein was added in 6M guanidine hydrochloride at a concentration of 0.01 mg / m². Diluted to L. Three 100uL aliquots were placed in a 96-well transparent bottom black wall plate. These purified fluorescent proteins were added to a solution and incubated at 25°C for 23 hours. Using excitation wavelengths of 402 nm and 457 nm, and emission wavelengths of 457 nm, respectively, for 30 minutes. The assay was performed at intervals. The plate was agitated before each measurement. The fluorescence values ​​measured at time zero were recorded. We used this to normalize the fluorescence throughout the rest of the assay (for example, a measurement of 0.8 is (This indicates that the protein retained 80% of its initial fluorescence.) As shown in Figure 7, Bluebonnet is for all hours where time is greater than 0 and up to approximately 24 hours. It showed higher chemical stability than SecBFP2.1 across various points.

[0266] Example 2: Bluebonnet, a brighter blue fluorescent protein When scientists look at how and where proteins move throughout a cell, This requires specialized genetic tools. One of these tools is ultraviolet light, In other words, it is a family of proteins that emit fluorescence under fluorescent light. The protein (BFP, pdb:3m24) is a much more commonly used red fluorescent protein. Although it is a high-quality derivative, it suffers from insufficient in vivo activity. Three-dimensional convolutional neural Using a network pipeline, fluorescence is detected when expressed in E. coli cells. We predicted the BFP variants that would cause the increase. Figure 8 shows 17 neural networks. The prediction provides data showing that it was tested for its ability to increase fluorescence (wild type). (Normalized and shown). Figure 9 shows that when beneficial mutations are combined, the result is 8% better than the wild type. This provides data showing that a more than twofold increase in fluorescence was observed. Figure 10 shows S28A, S1 Bluebonnet blue containing combinations of 14T, T127L, and N173H mutations. The increased fluorescence of the color fluorescent protein was noticeable compared to the parent strain and other blue fluorescent proteins. To show that it is visible.

[0267] Computer system diagram Figures 11A and 11B relate to the implementation of the systems and methods discussed herein. This is a block diagram showing a useful computer embodiment. Figures 11A and 11B are examples. A block diagram of a typical computer 1100 is shown. As shown in Figures 11A and 11B. The computer 1100 includes a central processing unit 1102 and a main memory 1104. Computer 1100 also has access to any other elements, such as one or more input / output devices. 130a~130n (generally referred to using reference number 1130), coprocessor 1 106, as well as the central processing unit 1102 and the coprocessor 1106 that are communicating with each other. It may include a 1140-unit memory chip.

[0268] The central processing unit 1102 responds to and fetches from the main memory 1104 An arbitrary logic circuit that processes instructions. In many embodiments, the central processing unit is an Int el Corporation(Mountain View,California) Manufactured by Motorola Corporation (Schaumb Manufactured by (urg, Illinois), International B Business Machines (White Plains, New York) Therefore, manufactured by Advanced Micro Devices (Sun Microprocessors such as those manufactured by Nyvale, California Provided by the device.

[0269] Similarly, the coprocessor 1106 responds to the main memory 1104 and then feeds from it It is an arbitrary logic circuit that processes the given instructions. In some embodiments, the coprocessor SA1106 is on Google (Mountain View, California) Tensor processors, which are integrated circuits manufactured for specific applications of artificial intelligence, are examples of such products. It may include a TPU (Turf Plastic) unit.

[0270] The main memory 1104 stores data, and any storage location is connected to the main processor 1102 Alternatively, it can be directly accessed by the microprocessor of coprocessor 1106. One or more memory chips that can be made to function, for example, static random access SRAM, Burst SRAM, or SynchBurst SRAM (BSRAM), Dynamic Random Access Memory (DRAM), High-Speed ​​Page Mode DRAM (FPM DRAM), Enhanced DRAM (EDRAM), Extended Data Output RAM ( EDO RAM), Extended Data Output DRAM (EDO DRAM), Burst Extended Data Output DRAM (BEDO DRAM), Enhanced DRAM (EDRAM), Synchronous DRAM (S DRAM, JEDEC SRAM, PC100 SDRAM, Double Data Rate SD RAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), SyncLink DRAM (SLDRAM), Direct Rambus DRAM (DRDRAM), or forced induction It could be an electronic RAM (FRAM).

[0271] In the embodiment shown in Figure 11A, the processor 1102 is connected to the system bus 1120 (and It communicates with main memory 1104 via (which will be described in more detail below). Similarly, the coprocessor 1106 communicates with the main memory 1104 via the system bus 1120. Figure 11B is The processor 1102 communicates directly with the main memory 1104 via the memory port. An embodiment of the computer system 1100 is shown. For example, in Figure 11B, the main memory 110 4 may be DRDRAM. In some embodiments, a neural network The engine may be involved in storing the values ​​of the trained weights, therefore the main memory It may reside within the memory device.

[0272] Figures 11A and 11B show the main processor 1102 at times on the "backside" bus. This describes an embodiment in which the cache memory 1140 communicates directly via a secondary bus referred to as [name of secondary bus]. In some embodiments, the coprocessor 1106 caches memory via the secondary bus. In another embodiment, the main processor 1102 communicates directly with the 1140. The system communicates with the cache memory 1140 using the stem bus 1120. In other embodiments, The coprocessor 1106 uses the system bus 1120 to access the cache memory 1140. It may also communicate with. The cache memory 1140 is typically more efficient than the main memory 1104. They also have fast response times, typically through SRAM, BSRAM, or EDRAM. Provided. In some embodiments, the coprocessor is related to the neural network. A tensor processing unit (TPU) or other coprocessor to perform the calculations. For example, it may include an application-specific integrated circuit (ASIC) (primary processor 11 (This may be faster or more efficient than performing such calculations on 02.)

[0273] In the embodiment shown in Figure 11A, the processor 1102 and the coprocessor 1106 are It communicates with various I / O devices 1130 via the local system bus 1120. ESA VL bus, ISA bus, EISA bus, Microchannel architecture (MC) A) Bus, PCI bus, PCI-X bus, PCI-Express bus, or NuBu Using various buses including s, the central processing unit 1102 and coprocessor 1106 are connected to I It can be connected to I / O device 1130. Implementation where the I / O device is a video display. Regarding the state, processor 1102 and / or coprocessor 1106 are advanced. Even when using the AGP Graphics Port to communicate with the display... Good. Figure 11B shows that the main processor 1102 is HyperTransport, Rap Direct communication with I / O device 1130b via id I / O or InfiniBand Figure 11B shows one embodiment of the computer system 1100. This describes an embodiment in which bus and direct communication are combined, where processor 1102 controls I / O devices While communicating directly with S1130b, use the local interconnect bus to use I / O device 11 Communicating with 30a.

[0274] A wide variety of I / O devices 1130 may be present in the computer system 1100. Input devices include keyboards, mice, trackpads, trackballs, and microphones. Examples include drawing tablets and video displays. Ray, speaker, inkjet printer, laser printer, and dye-sublimation printer The I / O devices also include mass storage devices for computer system 1100. For example, hard disk drives, 3.5-inch, 5.25-inch disks or ZI A floppy disk drive for receiving floppy disks such as P-disks, and a CD player. -ROM drives, CD-R / RW drives, DVD-ROM drives, various types of tapes Drive, and Twintech Industry, Inc. (Los Angeles) USB Flash Drive manufactured by Amitos, California VE's device line, and Apple Computer, Inc. (Cupert) The iPod Shuffle device manufactured by ino, California We can provide USB storage devices such as Sline.

[0275] In a further embodiment, the I / O device 1130 communicates with the system bus 1120 and externally. Communication buses, such as USB bus, Apple Desktop bus, and RS-232 serial. Connection, SCSI bus, FireWire bus, FireWire 800 bus, Ethe RNET bus, AppleTalk bus, Gigabit Ethernet bus, non-same Automatic transfer mode bus, HIPPI bus, Super HIPPI bus, SerialPlus S-bus, SCI / LAMP bus, FibreChannel bus, or Serial A bridge between the attached small computer system interface bus. That's fine.

[0276] The general-purpose desktop computers of the type shown in Figures 11A and 11B are typically , operators that control task scheduling and access to system resources It operates under the control of a control system. Typical operating systems include... According to Microsoft Corp. (Redmond, Washington) Microsoft Windows and Apple Computer were manufactured by Microsoft. MacOS, Intel, manufactured by Cupertino, California. rnational Business Machines(Armonk,New Y OS / 2 manufactured by ork, and Caldera Corp. (Salt Freely available operets distributed by Lake City, Utah Linux, a Linux distribution system, is one example.

[0277] Any patents, patent applications, and publications disclosed herein refer to the following: The whole of these is more fully incorporated herein. The present invention is disclosed with reference to specific embodiments. However, other embodiments and modifications of the present invention may deviate from the true spirit and scope of the invention. It is clear that this can be devised by a person skilled in the art without escaping. (See attached claims) This is intended to be interpreted as including all such embodiments and equivalent modifications. ru.

Claims

1. A mutant secBFP2 protein comprising an amino acid sequence in which one or more mutations selected from S28A, Y96F, S114T, and N173H are included in the amino acid sequence of SEQ ID NO: 1, The aforementioned amino acid sequence is a mutant secBFP2 protein having at least 90% identity with the amino acid sequence of SEQ ID NO:

1.

2. The mutant secBFP2 protein according to claim 1, wherein the amino acid sequence includes the S28A mutation.

3. The mutant secBFP2 protein according to claim 2, wherein the amino acid sequence includes mutations at S28A and S114T.

4. The mutant secBFP2 protein according to claim 3, wherein the amino acid sequence comprises the mutations S28A, S114T, and N173H.

5. The mutant secBFP2 protein according to claim 4, wherein the amino acid sequence comprises the mutations S28A, S114T, N173H, and Y96F.

6. The mutant secBFP2 protein according to claim 4, wherein the amino acid sequence includes the mutations S28A, S114T, N173H, and T127L.

7. The mutant secBFP2 protein according to claim 5 or 6, wherein the amino acid sequence comprises the mutations S28A, S114T, N173H, Y96F, and T127L.

8. The mutant secBFP2 protein or fusion protein according to claim 1, wherein the amino acid sequence comprises one of the amino acid sequences of SEQ ID NOs. 5, 6, 8, 18, 24-28.

9. A nucleic acid molecule comprising a nucleotide sequence encoding the mutant secBFP2 protein described in claim 1.

10. A nucleic acid molecule according to claim 9, which is a plasmid.

11. The nucleic acid molecule according to claim 9, which is an expression vector.

12. A nucleic acid molecule according to any one of claims 9-11, further comprising multiple cloning sites for insertion of heterologous protein-coding sequences.

13. A composition comprising the mutant secBFP2 protein described in claim 1.

14. A composition comprising a nucleic acid molecule according to any one of claims 9 or 12.

15. A kit comprising a nucleic acid molecule according to any one of claims 9-12.

Images