Method for acquiring compound having molecular recognition ability, method for identifying sample, and trained model for predicting compound having molecular recognition ability
The method of using a substrate array and machine learning to predict compounds with molecular recognition ability addresses inefficiencies in existing sequence evaluation, enabling efficient identification and detection of target molecules and samples.
Patent Information
- Application Number
- PCT/JP2024/015564
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-16
- Filing Date
- 2024-04-19
- Publication Date
- 2024-11-21
AI Technical Summary
Current methods for evaluating molecular recognition ability of compounds with diverse sequences are inefficient, particularly due to biases in SELEX methods and limitations in testing large numbers of sequences, leading to potential missed high-binding strength sequences.
A method involving a substrate array contacted with candidate compounds and a target molecule to detect interaction signals, followed by machine learning prediction of compounds with molecular recognition ability, allowing for iterative selection and evaluation.
This approach enables efficient identification and selection of compounds with high molecular recognition ability, facilitating the detection of target molecules and sample identification with improved accuracy and throughput.
Smart Images

Figure JP2024015564_21112024_PF_FP_ABST
Abstract
Description
Method for obtaining compounds with molecular recognition ability, method for identifying samples, and trained model for predicting compounds with molecular recognition ability
[0001] The present invention relates to a method for obtaining a compound having molecular recognition ability, a method for identifying a sample, and a trained model for predicting a compound having molecular recognition ability.
[0002] By using a compound that has the function of recognizing a target molecule (molecular recognition ability), it is possible to inhibit the function of the target molecule or specifically detect the target molecule. Examples of compounds that have molecular recognition ability that target biomolecules include antibodies, nucleic acid aptamers, and peptide aptamers. Nucleic acid aptamers and peptide aptamers are attracting attention as next-generation target molecule recognition materials from the viewpoints of ease of production, storage stability, etc. These target molecule recognition materials can recognize different target molecules depending on the base or amino acid sequence. Therefore, a highly diverse library can be easily created by varying the base or amino acid sequence. The diversity of sequences can be increased to 10 12 ~10 18Because the number of compounds with different sequences is so large, it is difficult to evaluate in detail the molecular recognition ability of all compounds with different sequences. Therefore, it is necessary to select candidate compounds that bind to target molecules from the vast variety of compounds with different sequences by some method for detailed evaluation. For example, when evaluating the molecular recognition ability of a nucleic acid aptamer, nucleic acid aptamer candidate sequences can be selected from a nucleic acid library by SELEX (systematic evolution of ligands by exponential enrichment). However, in view of bias caused by PCR (polymerase chain reaction) performed in the SELEX method and the long period required for the SELEX method, in recent years, selection methods that do not use the SELEX method have also been developed, as disclosed in Patent Document 1 and Prior Art Document 1. In addition, there is a technology in which aptamers with high binding strength are obtained by the SELEX method by introducing modified nucleic acids into a nucleic acid library (see Patent Document 2). However, these methods make it practically difficult to select only candidate compounds that bind. Therefore, these methods require binding tests to determine whether or not a compound exhibits binding to a target molecule. Evaluations using enzyme-linked immunosorbent assay (ELISA) and surface plasmon resonance are the mainstream tests for examining binding strength. These methods require the synthesis of individual sequences. Furthermore, these methods are limited in the number of sequences that can be evaluated at one time, at approximately 100 to 1,500. On the other hand, next-generation sequencers are used to obtain candidate sequences, allowing the acquisition of a vast number of candidate sequences, ranging from tens of thousands to hundreds of thousands. Therefore, with conventional methods, only a portion of the sequences can be tested for binding, potentially resulting in the overlooking of sequences with high binding strength.
[0003] Japanese Patent Application Laid-Open No. 2018-029601 Japanese Patent Application Laid-Open No. 2019-527043
[0004] Saito et al. , “Single-Round DNA Aptamer Selection by Combined Use of Capillary Electrophoresis and Next Generation Sequencing: An Aptaomics Approach for Identifying Unique Functional Protein-Binding DNA Aptamers”, Chemistry-A European Journal, volume 27, Issue 39, 2021, 10058-10067
[0005] The present invention has been made in view of the above circumstances. The problem to be solved by the present invention is to provide a method for efficiently obtaining compounds with high molecular recognition ability, and a method for identifying samples using the compounds obtained thereby. Another problem to be solved by the present invention is to provide a trained model that enables efficient obtaining of compounds with high molecular recognition ability.
[0006] The present inventors have investigated the causes of the above-mentioned problems in order to solve them. As a result, the present inventors have found that the above-mentioned problems can be solved by a step A of contacting a target molecule with a substrate on which candidate compounds are arranged in an array and detecting an interaction signal between the candidate compound and the target molecule, and a step B of predicting a compound having molecular recognition ability by machine learning based on the interaction signal, and have arrived at the present invention. That is, the above-mentioned problems are solved by the following means.
[0007] 1. A method for acquiring a compound with molecular recognition ability, comprising: Step A: bringing a target molecule into contact with a substrate on which candidate compounds are arranged in an array, and detecting an interaction signal between the candidate compounds and the target molecule; and Step B: predicting a compound with molecular recognition ability by machine learning based on the interaction signal.
[0008] 2. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability predicted in step B is used as the candidate compound in step A, and step A and step B are repeated.
[0009] 3. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability is a nucleic acid or a peptide.
[0010] 4. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability is a nucleic acid having a base sequence of 10 to 100 bases.
[0011] 5. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability is an artificial nucleic acid or a modified nucleic acid.
[0012] 6. The method for obtaining a compound having molecular recognition ability according to Item 1, wherein the compound having molecular recognition ability is a compound that exhibits a change in luminescence upon interaction with the target molecule.
[0013] 7. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability is a compound having a luminescent site and a spacer that does not affect the luminescence of the luminescent site.
[0014] 8. A method for identifying a sample, comprising using a compound obtained by the method for obtaining a compound having molecular recognition ability described in Item 1 to identify the sample.
[0015] 9. The method for identifying a sample according to claim 8, wherein the sample is identified by machine learning based on the pattern of the luminescent spots.
[0016] 10. The method for identifying a sample according to Item 8, wherein the sample is a biological sample.
[0017] 11. A trained model for predicting compounds with molecular recognition ability, which is machine-learned using the specific characteristics of candidate compounds as explanatory variables and the interaction signal between the candidate compounds and target molecules obtained by using a substrate on which candidate compounds are arranged in an array as the objective variable.
[0018] The above-described means of the present invention can provide a method for efficiently obtaining compounds with high molecular recognition ability, and a method for identifying samples using the compounds obtained thereby. Furthermore, the above-described means of the present invention can provide a trained model that enables efficient obtaining of compounds with high molecular recognition ability.
[0019] Flowchart showing an example of obtaining a compound with molecular recognition ability from a nucleic acid aptamer Clustering results in an example of screening for IgE aptamers Histogram of binding signals in an example of screening for IgE aptamers Fluorescence microscope image in an example of sample identification (reaction pattern when reacting with fetal bovine serum) Fluorescence microscope image in an example of sample identification (reaction pattern when reacting with mouse serum) Fluorescence microscope image in an example of designing a fluorescent dye-introduced aptamer
[0020] The method for acquiring a compound having molecular recognition ability of the present invention is characterized by comprising: a step A of contacting a target molecule with a substrate on which candidate compounds are arranged in an array and detecting an interaction signal between the candidate compound and the target molecule, and a step B of predicting a compound having molecular recognition ability by machine learning based on the interaction signal. This feature is a technical feature common to or corresponding to the following embodiments.
[0021] In an embodiment of the method for obtaining a compound having molecular recognition ability of the present invention, it is preferable to repeat steps A and B using the compound having molecular recognition ability predicted in step B as the candidate compound in step A. This makes it easier to obtain a compound having higher molecular recognition ability.
[0022] In an embodiment of the method for obtaining a compound having molecular recognition ability of the present invention, the compound having molecular recognition ability is preferably a nucleic acid or a peptide. This is preferable from the viewpoints of ease of production of the compound, storage stability, etc. Nucleic acids and peptides are also preferable in that they can be synthesized on a substrate by inkjet technology.
[0023] In an embodiment of the method for obtaining a compound having molecular recognition ability of the present invention, the compound having molecular recognition ability is preferably a nucleic acid consisting of a base sequence in the range of 10 to 100. This is preferable from the viewpoint of molecular recognition ability.
[0024] In an embodiment of the method for obtaining a compound having molecular recognition ability of the present invention, the compound having molecular recognition ability is preferably an artificial nucleic acid or a modified nucleic acid, which is preferable from the viewpoint of molecular recognition ability.
[0025] In an embodiment of the method for obtaining a compound having molecular recognition ability of the present invention, the compound having molecular recognition ability is preferably a compound that exhibits a change in luminescence upon interaction with the target molecule, thereby facilitating the simultaneous detection of multiple target molecules.
[0026] In an embodiment of the method for obtaining a compound having molecular recognition ability of the present invention, the compound having molecular recognition ability is preferably a compound having a luminescent moiety and a spacer that does not affect the luminescence of the luminescent moiety, which is preferable from the viewpoint of ease of detection of an interaction signal.
[0027] The method for identifying a sample of the present invention is characterized in that the sample is identified using a compound obtained by the method for obtaining a compound having molecular recognition ability of the present invention.
[0028] In an embodiment of the sample identification method of the present invention, it is preferable to identify the sample by machine learning based on the pattern of the luminescent spots, which is preferable from the viewpoint of the efficiency of sample identification.
[0029] In an embodiment of the sample identification method of the present invention, the sample is preferably a biological sample. Biological samples are generally difficult to identify because they contain a large number of molecules, but by using the identification method of the present invention, more accurate identification is possible.
[0030] The trained model for predicting compounds with molecular recognition ability of the present invention is characterized by being machine-learned using the specific characteristics of the candidate compounds as explanatory variables and the interaction signals between the candidate compounds and target molecules obtained using a substrate on which the candidate compounds are arranged in an array as objective variables.
[0031] The present invention will be described in detail below. In this application, the symbol "to" is used to mean that the numerical values before and after it are included as the lower limit and upper limit.
[0032] [Method for obtaining a compound with molecular recognition ability] The method for obtaining a compound with molecular recognition ability of the present invention is characterized by comprising the following steps A and B. Step A: A target molecule is contacted with a substrate on which candidate compounds are arranged in an array, and an interaction signal between the candidate compound and the target molecule is detected. Step B: A compound with molecular recognition ability is predicted by machine learning based on the interaction signal.
[0033] [Target Molecule] The type of target molecule is not particularly limited, and may be, for example, a target molecule used in the medical field or a target molecule used in the industrial field.
[0034] Examples of target molecules include the following: - Molecules that function as biomarkers, whose detection or quantification allows for the diagnosis of disease or the detection of pre-disease - Biomolecules whose function can be inhibited or enhanced to treat disease or alleviate symptoms - Molecules that need to be purified or removed in the production of products
[0035] The target molecule according to the present invention is, for example, a molecule contained in a biological sample. Examples of biological samples include whole blood, leukocytes, peripheral blood mononuclear cells, plasma, serum, sputum, breath, urine, semen, saliva, cerebrospinal fluid, amniotic fluid, glandular fluid, lymph, nipple aspirate, bronchial aspirate, synovial fluid, joint aspirate, cells, cell culture medium, cell extract, feces, tissue extract, cerebrospinal fluid, etc. The target molecule does not have to be a biomolecule such as a protein, but may also be a low molecular weight compound, polymer, etc.
[0036] [Compound with molecular recognition ability] The term "compound with molecular recognition ability" refers to a compound that has the ability to recognize a target molecule. The compound can inhibit the function of the target molecule or specifically detect the target molecule.
[0037] The term "candidate compound" refers to a compound that is a candidate for the compound having high molecular recognition ability that will ultimately be obtained.
[0038] The compound having molecular recognition ability according to the present invention may be, for example, an antibody, a nucleic acid, or a peptide. From the viewpoints of ease of production, storage stability, etc., the compound having molecular recognition ability is preferably a nucleic acid or a peptide. Nucleic acids and peptides are also preferred because they can be synthesized on a substrate by inkjet technology.
[0039] Examples of nucleic acids include modified nucleic acids and artificial nucleic acids. Examples of modified nucleic acids or artificial nucleic acids include the following: LNA (locked nucleic acid) BNA (bridged nucleic acid) Phosphorothioate 2'-OMe, AmNA (amido-bridged nucleic acid)
[0040] From the viewpoint of molecular recognition ability, the compound having molecular recognition ability is preferably a nucleic acid having a base sequence of 10 to 100 bases.
[0041] The compound having molecular recognition ability may partially contain a low molecular weight compound or a low molecular weight functional group, or an artificially created compound. Examples of the low molecular weight compound or low molecular weight functional group include biotin and an amino group. Examples of the artificially created compound include modified nucleic acids, artificial nucleic acids, artificial amino acids, and luminescent dyes. Among these, it is preferable that the compound having molecular recognition ability is an artificial nucleic acid or a modified nucleic acid from the viewpoint of molecular recognition ability.
[0042] The compound having molecular recognition ability is preferably a compound that exhibits a change in luminescence upon interaction with a target molecule. This facilitates simultaneous detection of multiple target molecules. In the present invention, "luminescence" includes fluorescence and phosphorescence. The compound that exhibits a change in luminescence upon interaction with a target molecule has a luminescent site derived from, for example, a luminescent dye.
[0043] The luminescent dye may be pyrene, a luminescent dye having the luminescent properties of a monomer or excimer, or a combination of a luminescent dye and a quencher used in a molecular beacon. Examples of fluorescent dyes include 6-FAM, JOE, TET, HEX, Cy3, ROX, Texas Red, Cy5, and Cy5.5. Examples of quenchers include BHQ-1, BHQ-2, BHQ-3, and DABCYL. These fluorescent dyes and quenchers may be substituted, for example, at the base site of a nucleic acid.
[0044] The compound having molecular recognition ability is preferably a compound having a light-emitting site and a spacer that does not affect the light emission of the light-emitting site. This makes it easier to detect the interaction signal. Examples of spacers include alkyl chains, sugar backbones of nucleic acids, polyethylene glycol, and peptides.
[0045] In the present invention, when a compound having molecular recognition ability or a candidate compound is a nucleic acid, it is also referred to as a "nucleic acid aptamer." Similarly, when a compound having molecular recognition ability or a candidate compound is a peptide, it is also referred to as a "peptide aptamer."
[0046] A method for obtaining a compound having molecular recognition ability will be described below using the flowchart shown in Figure 1. Figure 1 is a flowchart illustrating an example of obtaining a compound having molecular recognition ability from a nucleic acid aptamer. The present invention may include steps A and B. The other steps shown in the flowchart in Figure 1 do not limit the present invention.
[0047] In S1, a binding experiment between the target molecule and the nucleic acid aptamer is conducted, and the number of nucleic acid aptamers that have been confirmed to bind to the target molecule is narrowed down to approximately several hundred thousand.
[0048] In S2, the sequence of the nucleic acid aptamer that has been confirmed to bind to the target molecule in the binding experiment in S1 is analyzed using a sequencer.
[0049] An embodiment of S1 and S2 will be described using an example in which a compound with molecular recognition ability is obtained from a nucleic acid aptamer. Candidate sequences for nucleic acid aptamers can be selected using the SELEX method to identify sequences capable of binding to target molecules. A general method for obtaining candidate aptamer sequences, such as the SELEX method, can be used to select nucleic acid sequences capable of binding to target molecules. For example, a nucleic acid library containing a primer region and a random region is mixed with magnetic beads on which target molecules are immobilized, and nucleic acid sequences capable of binding to the target molecule are isolated. Only nucleic acids are isolated from the complex of nucleic acids and target molecules on the magnetic beads by using a solution containing sodium hydroxide or heat treatment. The isolated nucleic acids are used as templates for PCR amplification, and then single-stranded by treating with a solution containing sodium hydroxide or heat to create a library for the next round. By repeating this process, nucleic acid sequences capable of binding to the target molecule are enriched. The enriched sequences are sequence-analyzed using a next-generation sequencer to obtain nucleic acid sequences capable of binding to the target molecule. The method for immobilizing the target molecule is not limited to magnetic beads, and general carriers such as membranes, gels, and resins can be used. Nucleic acids capable of binding to target molecules may be selected without immobilizing the target molecules. For example, a nucleic acid library and a target molecule may be mixed and subjected to an ultrafiltration filter to separate nucleic acid sequences capable of binding to the target molecule. Alternatively, a sample containing a mixture of a nucleic acid library and a target molecule may be subjected to capillary electrophoresis to separate the nucleic acid sequences. The method for selecting these nucleic acids is not limited to the SELEX method. Nucleic acid sequences capable of binding to target molecules may also be selected without undergoing a library enrichment step by PCR amplification. Furthermore, sequence analysis is not limited to next-generation sequencers, and a nanopore sequencer may also be used.
[0050] In S3, the sequence of the candidate compound to be placed on the substrate is determined. The sequence of the candidate compound to be placed on the substrate may be selected from sequences identified by a sequencer or the like. Alternatively, the sequence of the candidate compound to be placed on the substrate may be selected using a machine learning method such as unsupervised learning, semi-supervised learning, or supervised learning. An example of selection using a machine learning method is described below.
[0051] (Example of Selection of Candidate Compound Sequences Using an Unsupervised Machine Learning Model) By training any unsupervised machine learning model using the identified sequence information, etc. as explanatory variables, compounds can be grouped by similar sequences. Examples of unsupervised machine learning models include k-means clustering, principal component analysis, and the deep learning model VAE (variational autoencoder). By training the identified sequence in a VAE, the sequence information can be encoded (converted) into a numerical value (latent space) of a specified number of dimensions. See Nature Computational Science, volume 2, pages 378-386, 2022. In this case, the numerical values in the latent space are expected to be similar between similar sequences, particularly between sequences having characteristic partial sequences (characteristic sequences). For this reason, it is also possible to select a group of sequences having a specific partial sequence by selecting sequences having similar numerical values in the latent space. Conversely, by selecting sequences so that the values in the latent space are as varied as possible, it is also possible to increase the variety of sequences to be selected. D-optimal programming and the like are examples of methods for varying the values in the latent space of the sequences to be selected. A trained unsupervised machine learning model can predict the clustering patterns and latent spaces of not only the trained sequences but also any untrained sequences. This allows sequences that have not been trained by the unsupervised machine learning model to also be included in the selection of candidate compounds to be placed on the substrate.
[0052] (Example of Selection of Candidate Compound Sequence Using Supervised Machine Learning Model) The sequence of a candidate compound may be selected with reference to the predicted value of any supervised machine learning model. Examples of supervised machine learning models include a multiple regression model, a PLS regression model, and a deep learning model. For example, a supervised machine learning model is trained using the interaction signal of each sequence with the target molecule as the objective variable and sequence information as the explanatory variable. This makes it possible to predict the interaction signal with the target molecule from the sequence information. This supervised machine learning model can predict the interaction signal with the target molecule not only for the trained sequence but also for any untrained sequence. This supervised machine learning model is made to predict the interaction signal with the target molecule of any sequence, and a sequence with a high predicted value is selected. This makes it possible to efficiently derive sequences that are likely to have high binding strength.
[0053] When selecting the sequence of a candidate compound to be placed on a substrate, any sequence can be used as a candidate for selection, not limited to sequences identified by a sequencer. The number of candidate sequences for selection may be increased by any method, such as generating sequences using a generative model that adds mutations to sequences using a genetic algorithm or the like. Examples of such generative models include VAE, GAN (generative adversarial networks), FLOW, and a diffusion model. Previously reported sequences may be used as candidates for selection.
[0054] Explanatory variables in machine learning are not limited, but examples include sequence information, three-dimensional structure, and physical property values of any compound. Objective variables are also not limited, but examples include interaction signal values with target molecules, dissociation constants, etc. Both explanatory variables and objective variables may include information about previously published sequences.
[0055] In S4, candidate compounds are arranged in an array on a substrate. At this time, the candidate compounds may be synthesized on the substrate. Printing techniques such as inkjet printing or photolithography techniques can be used to synthesize the candidate compounds on the substrate.
[0056] S5 is step A. In step A, a substrate on which candidate compounds are arranged in an array is contacted with a target molecule to detect an interaction signal between the candidate compound and the target molecule. The interaction signal is a signal generated by an interaction such as binding between the candidate compound and the target molecule. Interactions include chemical bonds, intermolecular interactions, etc. An interaction signal generated by chemical bonds is also called a binding signal. The interaction signal is, for example, luminescence generated by binding between the candidate compound and the target molecule. Luminescence includes fluorescence and phosphorescence.
[0057] In one example of step A, a solution containing a target molecule is dropped onto a substrate on which candidate compounds are arranged in an array. The candidate compound and the target molecule at each spot are reacted, and an interaction signal between the candidate compound and the target molecule is detected.
[0058] Methods for detecting interaction signals include indirect fluorescence methods using antibodies labeled with fluorescent dyes, and methods for designing and detecting candidate compounds that change fluorescence upon binding to target molecules.
[0059] S6 is step B. In step B, a compound having molecular recognition ability is predicted by machine learning based on the interaction signal detected in step A. Specifically, a trained model generated by machine learning predicts a compound having molecular recognition ability based on the input interaction signal and outputs the predicted compound.
[0060] A trained model for predicting compounds with molecular recognition ability can be generated, for example, by machine learning using the specific features of a candidate compound as explanatory variables and the interaction signal between the candidate compound and a target molecule obtained using a substrate on which the candidate compounds are arranged in an array as a response variable. The specific features of the candidate compound can be, for example, the base sequence, three-dimensional structure, physical properties, etc. of the nucleic acid aptamer that is the candidate compound.
[0061] In S7, it is determined whether or not the prediction of the compound having molecular recognition ability in step B is sufficient. If the prediction of the compound having molecular recognition ability is sufficient, the process proceeds to evaluation experiments in S8.
[0062] If the prediction of the compound having molecular recognition ability is not sufficient, the process returns to S3 to repeat steps A and B. At this time, steps A and B may be repeated using the compound having molecular recognition ability predicted in step B as a candidate compound in step A.
[0063] In S8, an evaluation experiment is carried out on the compound having the molecular recognition ability predicted in step B. The evaluation experiment is, for example, a binding experiment with a target molecule.
[0064] In S9, compounds having molecular recognition ability are obtained from the results of the evaluation experiment in S8.
[0065] [Method for Identifying a Sample] A compound obtained by the method for obtaining a compound having molecular recognition ability of the present invention can be used for identifying a sample.
[0066] The sample to be identified may be, for example, a biological sample, including whole blood, white blood cells, peripheral blood mononuclear cells, plasma, serum, sputum, breath, urine, semen, saliva, cerebrospinal fluid, amniotic fluid, glandular fluid, lymphatic fluid, nipple aspirate, bronchial aspirate, synovial fluid, joint aspirate, cells, cell culture medium, cell extract, feces, tissue extract, cerebrospinal fluid, etc.
[0067] A biological sample contains a large number of molecules. If a compound capable of molecular recognition for a large number of molecules can be obtained, it will be possible to more accurately identify the sample in a complex sample containing a large number of molecules. Among the large number of molecules, molecules that can be target molecules may be either already identified or unidentified.
[0068] One embodiment of the method for identifying a sample includes the following steps: (1) selecting a compound having molecular recognition ability capable of identifying the sample, (2) synthesizing the compound on a substrate by inkjet technology, (3) contacting the sample with the substrate, (4) acquiring a fluorescent image, and (5) analyzing the fluorescent image.
[0069] All molecules in a sample may be labeled with biotin, a luminescent dye, etc. In this case, the sample may be identified by machine learning based on the pattern of observed luminescent spots.
[0070] The present invention will be specifically described below with reference to examples, but the present invention is not limited to these. In the following examples, unless otherwise specified, operations were carried out at room temperature (25°C). Furthermore, unless otherwise specified, "%" and "parts" mean "% by mass" and "parts by mass", respectively.
[0071] [IgE aptamer screening] IgE-immobilized magnetic beads were mixed with the DNA library and stirred for 1 hour. The supernatant was then removed using a magnetic stand. The magnetic beads were then washed with phosphate buffer containing Tween 20. A 0.15N aqueous solution of sodium hydroxide was then added dropwise to dissociate the DNA from the magnetic beads, and the mixture was placed on a magnetic stand to recover the supernatant. The recovered solution was neutralized by adding a 1N aqueous solution of HCl dropwise. This sample was recovered as DNA that is thought to bind to the target molecule. The recovered DNA was subjected to PCR amplification using biotin-labeled primers to prepare single strands, thereby preparing the next round of libraries. This process was repeated three times, and the sequence was analyzed using a next-generation sequencer.
[0072] (Identification of sequences by next-generation sequencer) The fractions after screening were treated with the NEBNext (registered trademark) Multiplex Oligos for Illumina kit. Then, the fractions were subjected to MiSeq to identify the sequences contained in the fractions.
[0073] (Sequence Selection by Unsupervised Learning) The sequences identified by the next-generation sequencer were used as explanatory variables, and a VAE model, a type of unsupervised machine learning model, was trained. This resulted in conversion to a 10-dimensional numerical value. Similar sequences were then converted to similar numerical values. These numerical values were then subjected to GMM clustering, a type of unsupervised machine learning model, to classify similar sequences. Figure 2 shows the results of compressing the 10-dimensional numerical values to two dimensions using the UMAP method and coloring them with the clustering results. Based on these results, the sequences of candidate compounds to be immobilized on the substrate were selected.
[0074] A microarray in which candidate compounds were immobilized on a substrate in an array format was obtained from Agilent Technologies. 2% bovine serum albumin (BSA) was dropped onto the microarray and allowed to stand for 30 minutes for blocking. The microarray was then washed with phosphate buffer containing Tween 20. A 100 nM IgE solution was then applied to the microarray and allowed to stand at 25°C for 1 hour. The microarray was then washed with phosphate buffer containing Tween 20. A fluorescein-labeled anti-IgE antibody was then applied to the microarray and allowed to stand at 25°C for 1 hour. Unreacted fluorescein-labeled anti-IgE antibody was then washed with phosphate buffer containing Tween 20, and the microarray was sealed with phosphate buffer and a cover glass. Fluorescein fluorescence was then captured using a fluorescence microscope. The captured images were then subjected to image analysis using machine learning, and the fluorescence intensity of each spot was quantified as a binding signal. The binding signal value quantified here is referred to as the Round 1 value.
[0075] (Machine learning (regression model)) A deep regression model was trained using each sequence immobilized on the microarray as an explanatory variable and its binding signal value as the objective variable. By providing this deep regression model with arbitrary sequences as explanatory variables, the binding signal value of those sequences was predicted.
[0076] Sequences with high predicted binding signal values were selected. Compounds of the selected sequences were synthesized on a microarray. Binding signals of the synthesized compounds to the target molecules were quantified through binding experiments. The binding signal values quantified here were designated as Round 2 values.
[0077] As shown in the histogram in Figure 3 and Table I, both the average and maximum values of the binding signals improved from Round 1 to Round 2. This confirmed that the present invention can provide compounds with higher molecular recognition ability.
[0078]
[0079] [Sample Identification] Albumin and globulin contained in fetal bovine serum and mouse serum were removed. Albumin and globulin were removed using an Aurum Affi-Gel (Bio Rad) bead purification kit. Each sample was then labeled with biotin, mixed with a DNA library, and subjected to the SELEX method. Based on the sequences analyzed using a next-generation sequencer, candidate compounds with molecular recognition ability were screened from fetal bovine serum and mouse serum, respectively. These candidate compounds were synthesized and arranged in an array on a substrate.
[0080] Biotin-labeled fetal bovine serum or mouse serum was applied to this substrate. Unbound fetal bovine serum or mouse serum was washed away. Then, fluorescein-labeled streptavidin was applied to this substrate. Unbound fluorescein-labeled streptavidin was washed away.
[0081] The fluorescent spots emitted by fluorescein were then photographed using a fluorescence microscope, and the binding signals were detected by performing image analysis on the photographed images.
[0082] Based on the binding signals, we used machine learning to predict sequences with high binding affinity, and synthesized compounds with the predicted sequences so that they could be arranged in an array on a substrate.
[0083] Biotin-labeled fetal bovine serum or biotin-labeled mouse serum was dropped onto this substrate and allowed to stand at 25°C for 1 hour. After washing, a fluorescein-labeled streptavidin solution was dropped onto the substrate and allowed to stand at 25°C for 1 hour. After washing, the array was photographed using a fluorescence microscope. Analysis of the photographed images revealed different reaction patterns when fetal bovine serum was reacted and when mouse serum was reacted. The reaction pattern when fetal bovine serum was reacted is shown in Figure 4A, and the reaction pattern when mouse serum was reacted is shown in Figure 4B.
[0084] These results confirmed that by obtaining a variety of aptamers for the sample to be identified, it is possible to distinguish between bovine and mouse serum.
[0085] [Design of fluorescent dye-incorporated aptamer] Oligonucleotides with pyrene-substituted bases were inserted into both ends of a thrombin-binding aptamer. A spacer consisting of 10 sugar skeletons with deleted bases was inserted into the 3'-end of this aptamer, and an amino group was added to the 3'-end. This resulted in the design of a fluorescent dye-incorporated aptamer. This was immobilized on a glass substrate by the amine coupling method. Thrombin solution or buffer solution was dropped onto this, and a cover glass was placed on top and allowed to stand at 25°C for 30 minutes.
[0086] The solid phase portion was observed under a fluorescence microscope, and Figure 5 shows a fluorescence microscope image showing the observation results.
[0087] Almost no fluorescent spots were observed in the solid phase where the buffer solution was dropped. Fluorescent spots thought to be excimer emission were observed in the solid phase where the thrombin solution was dropped. These results confirmed that the structure of the aptamer changed upon binding to thrombin.
[0088] Furthermore, when the spacer was not included, a fluorescent spot was observed in the solid phase where the buffer solution was dropped. From these results, it was confirmed that the inclusion of a spacer can reduce background and enable highly sensitive detection of thrombin.
[0089] The present invention can be used in a method for obtaining compounds having molecular recognition ability, a method for identifying samples, and a trained model for predicting compounds having molecular recognition ability.
Claims
1. A method for obtaining a compound having molecular recognition ability, comprising: step A of contacting a target molecule with a substrate on which candidate compounds are arranged in an array, and detecting an interaction signal between the candidate compounds and the target molecule; and step B of predicting a compound having molecular recognition ability by machine learning based on the interaction signal.
2. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability predicted in the step B is used as the candidate compound in the step A, and the steps A and B are repeated.
3. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability is a nucleic acid or a peptide.
4. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability is a nucleic acid having a base sequence in the range of 10 to 100 bases.
5. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability is an artificial nucleic acid or a modified nucleic acid.
6. The method for obtaining a compound with molecular recognition ability according to claim 1, wherein the compound with molecular recognition ability is a compound that exhibits a change in luminescence upon interaction with the target molecule.
7. The method for obtaining a compound having molecular recognition ability according to claim 1, wherein the compound having molecular recognition ability is a compound having a luminescent site and a spacer that does not affect the luminescence of the luminescent site.
8. A method for identifying a sample, comprising using a compound obtained by the method for obtaining a compound having molecular recognition ability as described in claim 1 to identify the sample.
9. The method for identifying a sample according to claim 8, further comprising identifying the sample by machine learning based on the pattern of the luminescence spots.
10. The method for identifying a sample according to claim 8, wherein the sample is a biological sample.
11. A trained model for predicting compounds with molecular recognition ability, using machine learning to determine the specific characteristics of candidate compounds as explanatory variables and the interaction signals between candidate compounds and target molecules obtained by using a substrate on which the candidate compounds are arranged in an array as objective variables.
Citation Information
Patent Citations
Method for producing aptamer
JP2018029601A
Oligonucleotides containing modified nucleosides
JP2019527043A
Method for removing film from substrate and composition used therefor
JP2002053985A
Unique recognition sequences and their use in protein analysis
JP2006511819A
Screening method for ligand
JP2011177092A
