Viral vector inspection method and machine learning system

Abstract

Provided are a preparation method for removing impurities in a viral vector dispersion, and an inspection method for identifying the internal state of a viral vector in a dispersion column from which impurities have been removed, the methods comprising the steps of: forming a reaction solution by mixing a test sample and a reaction reagent containing attached magnetic beads obtained by attaching to the surface of magnetic beads an attachment substance that specifically binds to a viral vector; forming a conjugate of the viral vector and the attached magnetic beads in the reaction solution; applying a magnetic field from the outside of a container containing the reaction solution; substituting the reaction solution with a cleaning solution in a state in which the conjugate has been immobilized in the container, and subsequently removing the cleaning solution; and furthermore introducing an elution reagent into the container, thereby liberating the viral vector from the conjugate so as to produce a viral vector dispersion eluate in a state in which the viral vector is dispersed in the elution reagent.

Description

Viral vector testing method and machine learning system

[0001] The present invention relates to a method for testing viral vectors and a machine learning system. In particular, the present invention relates to a method for testing viral vectors and a machine learning system for evaluating the quality of viral vectors useful in gene therapy.

[0002] Gene therapy is useful for the treatment and prevention of various diseases, and is expected to be applied to a variety of purposes, from gene suppression, replacement, and editing to vaccine development. Many cell and gene therapy drugs are already in clinical trials, and some have been approved by regulatory authorities.

[0003] Generally, negatively charged nucleic acid molecules such as messenger RNA, microRNA, and plasmid DNA do not efficiently cross cell membranes. Furthermore, they can be rapidly metabolized by enzymatic degradation. To solve this problem, these delivery substances must be housed within vectors and delivered to cells. Among these, viral vectors are expected to be efficient vectors, as they have the ability to transfer genetic information into host cells.

[0004] On the other hand, from the viewpoint of safety and efficacy, it is necessary to eliminate empty vectors that do not contain the delivered gene during the process of packaging the delivered gene into the vector. It is also necessary to ensure that the vector does not contain immunogenic or carcinogenic substances or nucleic acid fragments derived from the plasmid. Therefore, in the viral vector manufacturing process and quality evaluation after production, it is important to have a technology that can accurately identify the internal state of the viral vector, in addition to the number and aggregation degree of the viral vector.

[0005] Currently, methods proposed for evaluating the quality of viral vector production include transmission electron microscopy (Patent Document 1), ultracentrifugation analysis (Patent Document 2), capillary electrophoresis (Patent Document 3), flow cytometry (Patent Document 4), hydrophilic interaction chromatography (Patent Document 5), and mass photometry (Patent Documents 6 and 7), and efforts to ensure the safety and performance of drugs are ongoing.

[0006] In addition, even after viral vector production, chemical instability such as amidation, phosphorylation, acetylation, and isomerization of capsid proteins, as well as physical instability such as aggregation, unfolding, surface adsorption, and genome release of viral vectors, remain issues, and techniques for easily evaluating these are becoming increasingly important (Non-Patent Document 1).

[0007] International Publication No. 2023 / 238756 Japanese Patent Application Laid-Open No. 2022-095863 Special Publication No. 2023-510841 International Publication No. 2023 / 190974 Japanese Patent No. 7439077 Japanese Patent Application Laid-Open No. 2023-520316 Japanese Patent Application Laid-Open No. 2023-542931 Japanese Patent No. 6985687 Japanese Patent No. 6807529

[0008] Philip Grossen et al, The ice age - A review on formulation of Adeno-associated virus therapeutics, European Journal of Pharmaceutics and Biopharmaceutics, 190(2023)1-23Buddini Iroshika Karawdeniya et al, Adeno-associated virus characterization for cargo discrimination through nanopore responsiveness, Nanoscale, 2020, Nanoscale, No. 46, p. 23721-23731

[0009] In assessing the quality of viral vectors, the above-mentioned conventional techniques each have their advantages and disadvantages, and in particular, no method has been established that can accurately identify the internal state of a viral vector using a small amount of sample.

[0010] One of the most important quality evaluation items in the viral vector manufacturing process is to detect the internal state of each viral vector particle, such as whether the viral vector is packaged with the correct gene to be delivered (hereinafter referred to as "full particle"), whether it is empty (hereinafter referred to as "empty particle"), whether it is contaminated with an incomplete gene (hereinafter referred to as "partial particle"), or whether it is packaged with other harmful substances, and to estimate the ratio of the number of each particle at the final process or during an intermediate manufacturing process.

[0011] For example, genome sequencing can quantitatively examine the genetic information contained in a test sample to infer the state of genes packaged in a viral vector, but this method does not determine whether the genes detected in the test sample are inside or outside the viral vector.

[0012] Currently, transmission electron microscopy (TEM) is used as a method for estimating the internal state of viral vectors. Observing a viral vector test sample negatively stained with uranyl acetate using a transmission electron microscope makes it possible to distinguish between full and empty particles, and count them one by one within the field of view. However, the number of particles within the TEM field of view is only a small fraction of the entire test sample, making quantitative analysis a major challenge. Furthermore, ultracentrifugation analysis estimates the abundance ratio of unknown viral vector internal states, such as full and empty particles, by utilizing differences in sedimentation coefficients due to differences in the internal state of the viral vector. However, ultracentrifugation requires a large amount of test sample (500 μL or more), resulting in high testing costs. Furthermore, due to the large sample volume required, it is not practical to repeat testing during intermediate purification steps.

[0013] To address these issues, an attempt has been made to identify and evaluate the internal state of a viral vector using the signal emitted when the viral vector passes through a pore using an electrical detection zone method (see Non-Patent Document 2, cited above). This method allows for the identification of each viral vector as a full particle, an empty particle, etc., enabling highly accurate quantification. However, in conventional electrical detection zone methods, if a sample contains contaminants, there is no way to selectively allow only the viral vector particles among the contaminants to pass through the pore. As a result, the signal obtained when the contaminants pass through the pore becomes noise, making it impossible to properly identify the internal state of the viral vector.

[0014] Thus, conventional techniques, including the proposed electrical detection zone method, cannot identify the internal state of viral vectors present in samples containing contaminants. This means that the internal state of viral vectors cannot be identified in samples obtained during intermediate purification steps. For this reason, currently, after the virus production process, purification is performed without knowing the internal state of the viral vector until the completion of the purification process, and identification using the conventional techniques described above is required once purification is complete. Although the internal state of the viral vector is already determined at the completion of the virus production process, it is not possible to determine whether the produced viral vector is normal or not until after the enormous cost of completing purification has been incurred. If it is found to be defective at this stage, all purification costs will be wasted. For this reason, there is a strong demand for a technology that can identify the internal state of viral vectors in samples obtained during the purification process and surrounded by contaminants.

[0015] The present invention has been made in consideration of these circumstances, and aims to provide, in one embodiment, a method for testing viral vectors that can correctly identify the internal state of viral vectors even when samples containing impurities are used. In another embodiment, the present invention aims to provide a machine learning system that can efficiently perform such a method for testing viral vectors.

[0016] As a result of extensive research, the present inventors have found that the above-mentioned problems can be solved by binding a viral vector using magnetic beads with specific attachment substances attached, and then performing treatments such as a magnetic field and elution reagents, thereby minimizing the effects of contaminants. The present invention was completed based on the above findings, and is exemplified below.

[0017] [1] A method for testing a viral vector, comprising: identifying the internal state of the viral vector in a test sample using a pore sensor having one pore, two chambers connected to both sides of the pore, and a pair of electrodes installed in each of the two chambers; and using a reaction reagent containing attached magnetic beads having an attachment that specifically binds to the viral vector attached to the magnetic bead surface; when a reaction solution is prepared by mixing the test sample and the reaction reagent, a bond between the viral vector and the attached magnetic beads is generated in the reaction solution; applying a magnetic field from outside a container containing the reaction solution, and in a state where the bond is fixed in the container, replacing the reaction solution with a washing solution and then removing the washing solution; further, introducing an elution reagent into the container to liberate the viral vector from the bond, and generating a viral vector dispersion eluate in a state where the viral vector is dispersed in the elution reagent; introducing the viral vector dispersion eluate into one of the chambers of the pore sensor and an electrolyte into the other chamber of the pore sensor, and then applying a voltage to the pair of electrodes; A method for testing viral vectors, comprising: identifying the internal states of multiple viral vectors that have passed through the pores based on the characteristics of the shape of a pulse signal generated by a transient current change in the electrode pair when the viral vectors in the viral vector dispersion eluate pass through the pores. [2] The method for testing viral vectors described in [1], wherein the internal states include multiple predetermined types, and the method further comprises classifying the identified viral vectors into one of the multiple types of internal states. [3] The method for testing viral vectors described in [2], further comprising estimating the abundance ratio of the viral vectors in each internal state in the test sample by counting the number of viral vector particles for each classified internal state. [4] The method for testing viral vectors described in [1] or [2], wherein the attached substance is an antibody that specifically binds to the viral vector. [5] The method for testing viral vectors described in [1] or [2], wherein the attached substance is a lectin that specifically binds to a sugar chain expressed on the surface of the viral vector.[6] The method for testing a viral vector according to [2], wherein the internal state includes at least two types: a state in which a gene to be delivered is packaged in the viral vector, and a state in which a gene is not packaged in the viral vector. [7] A machine learning system for testing viral vectors, the system being configured to identify the internal state of a viral vector by using a pore sensor having one pore, two chambers connected to both sides of the pore, and a pair of electrodes installed in each of the two chambers, and analyzing a pulse signal measured each time a viral vector contained in a test sample passes through the pore; the system further comprises: a teacher sample and a reaction reagent including magnetic beads having an attachment attached to the magnetic bead surface that specifically binds to a teacher viral vector having a known internal state, to prepare a teacher reaction solution, whereby a teacher conjugate between the teacher viral vector and the magnetic beads is generated in the teacher reaction solution; a magnetic field is applied from the outside of a teacher sample container containing the teacher reaction solution, and the teacher reaction solution is replaced with a washing solution while the teacher conjugate is fixed in the container, and the washing solution is then removed; and an elution reagent is introduced into the teacher sample container to release the teacher viral vector from the teacher conjugate, thereby producing a teacher viral vector dispersion eluate in which the teacher viral vector is dispersed in the elution reagent; A machine learning system characterized by being configured to introduce the teacher virus vector dispersion eluate into one of the chambers of the pore sensor and an electrolyte into the other of the chambers of the pore sensor, apply a voltage to the electrode pair, and use as teacher data the characteristics of the shape of the teacher pulse signal due to the current transient change that occurs in the electrode pair when the teacher virus vector in the teacher virus vector dispersion eluate passes through the pore, and train using the known internal state as a teacher label to create a trained machine learning model.[8] The machine learning system further comprises: mixing a test sample containing a test viral vector whose internal state is unknown with the reaction reagent to prepare a test reaction solution, whereby a test binding entity between the test viral vector and the attached magnetic beads is generated in the test reaction solution; applying a magnetic field from the outside of a test sample container containing the test reaction solution, so that the test binding entity is fixed in the container, and then replacing the test reaction solution with the washing solution, and removing the washing solution; further introducing the elution reagent into the test sample container to liberate the test viral vector from the test binding entity, and generating a test viral vector dispersion eluate in which the test viral vector is dispersed in the elution reagent; introducing the test viral vector dispersion eluate into one of the chambers of the pore sensor and an electrolyte into the other of the chambers of the pore sensor, and then applying a voltage to the electrode pair; The machine learning system described in [7] is configured to estimate the internal state of the test virus vector contained in the test sample by inputting into the trained machine learning model the characteristics of the shape of an unknown pulse signal due to a current transient change that occurs in the electrode pair when the test virus vector in the test virus vector dispersion eluate passes through the pore. [9] A method for preparing a test viral vector dispersion eluate, comprising: preparing a test reaction solution by mixing a test sample containing a test viral vector whose internal state is unknown with a reaction reagent containing magnetic beads having a substance attached to the surface of the magnetic beads that specifically binds to the viral vector, thereby producing a test binding complex between the test viral vector and the magnetic beads; applying a magnetic field from the outside of a test sample container that contains the test reaction solution, so that the test binding complex is fixed in the container, replacing the test reaction solution with a washing solution, and then removing the washing solution; and further introducing an elution reagent into the test sample container, thereby liberating the test viral vector from the test binding complex, thereby producing a test viral vector dispersion eluate in which the test viral vector is dispersed in the elution reagent.

[10] A method for testing viral vectors, comprising identifying the internal state of a test viral vector in a test sample containing the test viral vector, the internal state of which is unknown, using a reaction reagent containing magnetic beads having a magnetic bead surface to which an attachment that specifically binds to the viral vector is attached, and an elution reagent that detaches the viral vector bound to the magnetic beads; the method comprises carrying out the following steps (1) and (2): (1) mixing a teacher sample containing a teacher viral vector whose internal state is known with the reaction reagent to prepare a teacher reaction solution, generating a teacher conjugate between the teacher viral vector and the magnetic beads in the teacher reaction solution; applying a magnetic field from the outside of a teacher container containing the teacher reaction solution, and replacing the teacher reaction solution with a washing solution while the teacher conjugate is fixed in the teacher container, and then removing the washing solution; and further introducing the elution reagent into the teacher container to release the teacher viral vector from the teacher conjugate in the elution reagent. (2) When the test sample and the reaction reagent are mixed to prepare a test reaction liquid, a test binding entity between the test viral vector and the attached magnetic beads is generated in the test reaction liquid, a magnetic field is applied from the outside of a test vessel containing the test reaction liquid, and while the test binding entity is fixed in the test vessel, the test reaction liquid is replaced with a washing liquid and then removed, and the elution reagent is further introduced into the test vessel, thereby liberating the test viral vector from the test binding entity in the elution reagent; the method then comprises: (3) measuring the teacher viral vector and the test viral vector dispersed in matrices having the same composition, and comparing the characteristics of the measurement results to estimate the internal state of the test viral vector in the test sample.

[0018] According to one embodiment of the present invention, there is provided a method for testing viral vectors that can correctly identify the internal state of viral vectors even when samples containing impurities are used. According to another embodiment of the present invention, there is provided a machine learning system that can efficiently perform such a method for testing viral vectors.

[0019] According to a preferred embodiment of the present invention, it is possible to identify the internal state of viral vectors contained in a small amount of sample, such as several to several tens of μL, and to quantify the number of viral vectors in each internal state, not only for samples in which purified viral vector particles are dominant, but also for samples in intermediate steps of purification in which impurities are contained.

[0020] FIG. 1(a) is a diagram showing an example of the structure of a pore sensor that can be used in one embodiment of the viral vector testing method of the present invention. FIG. 1(b) is a diagram showing an electron microscope image of a pore sensor pore 140 used to measure a pulse waveform. FIG. 1(b) is a diagram showing pulse-like transient changes in the ionic current flowing between the electrodes of the pore sensor shown in FIG. 1. FIG. 1(c) is a diagram showing the relationship between a viral vector and impurities from viral vector production to the completion of purification. FIG. 6(a) shows the distribution of pulse widths and pulse heights obtained by measuring sample 1. FIG. 6(b) is a typical pulse waveform. FIG. 6(c) is the distribution of pulse widths and pulse heights obtained by measuring sample 2. FIG. 6(d) is a typical pulse waveform. This is a flowchart illustrating an example of training a machine learning model using samples F1 and E1 in FIG. 5 as training samples, and quantifying full particles and empty particles in a test sample T1 using the trained model. FIG. 6 is a schematic diagram for explaining an example in which a machine learning model is trained using samples F1 and E1 in FIG. 5 as teacher samples, and the trained model is used to quantify full particles and empty particles in test sample T1. FIG. 7 is a diagram for explaining the relationship between teacher data and test samples in the first to third embodiments of the present invention. FIG. 8 is a diagram for explaining a fourth embodiment of the present invention that uses an unsupervised machine learning algorithm. FIG. 9 is a diagram for explaining the relationship between teacher data and test samples in the fifth embodiment of the present invention. FIG. 10 is a diagram for explaining a verification experiment regarding the effects of the present invention. FIG. 11 is a diagram for explaining a verification experiment regarding the effects of the present invention.

[0021] Next, embodiments of the present invention will be described in detail with reference to the drawings. It should be understood that the present invention is not limited to the following embodiments, and that appropriate design changes and improvements may be made based on the ordinary knowledge of those skilled in the art without departing from the spirit of the present invention.

[0022] (1. Method for Testing Viral Vectors (First Embodiment)) FIG. 1( a) shows an example of the structure of a pore sensor that can be used in one embodiment of the method for testing viral vectors of the present invention. The pore sensor 100 has a cross-sectional structure in which two chambers 110 and 120 are separated by a partition wall 141 and connected via a pore 140 provided in the partition wall 141. Electrodes 112 and 122 are installed in the two chambers, respectively. A sample containing particles to be measured dispersed in an electrolyte is introduced into chamber 110 via inlet 111, and the electrolyte is introduced into chamber 120 via inlet 121, and a voltage is applied to the two electrodes by a voltage source 152. These electrodes are connected to an amplifier 150, an ammeter 151, and a voltage source 152. For example, when a voltage is applied between electrode 112 and electrode 122, an ionic current flows through the pore.

[0023] FIG. 1(b) shows an electron microscope image of the pore 140 used to measure the pulse waveform in the experiment described below. The pore diameter is selected to be larger than the diameter of the particle to be measured. Note that FIG. 1 is merely one example of a pore sensor used for measurements in the present invention. The pore sensor used in the present invention is not particularly limited in terms of the shape or material of the chamber, pore, and inlet, as long as it has one pore, two chambers connected to both sides of the pore, and an electrode pair installed in each of the two chambers, and these may be selected appropriately as needed.

[0024] As shown in FIG. 1A , when particles of the test object 101 present in the chamber 110 pass through the aperture 140, the ionic current is temporarily interrupted, and then returns to normal after the test object 101 passes through the chamber 120. Therefore, each time one test object 101 passes through the aperture 140, the ionic current flowing between the electrodes in FIG. 1 exhibits a pulse-like transient change as illustrated in FIG. 2 . In the example of FIG. 1 , this is measured by an ammeter 151. In FIG. 2 , the pulse signal 210 represents a current value 202 at each time 201. The vertical axis 202 may also represent a voltage value. The baseline 200 represents the current value when there is no pulse (or the average over a certain period if there is noise), and the pulse height 208 represents the difference between the baseline 200 and the current value at which the largest current phenomenon occurs in the pulse waveform (or the average over a certain period if there is noise).

[0025] In viral vector production, a plasmid containing a delivery gene, a plasmid encoding the capsid, and a helper plasmid that aids in viral vector replication and assembly are introduced into host cells by transfection. After the viral vector is assembled in the host cell, viral vector particles are released. As conceptually shown in Figure 3 , viral vectors with various internal states can be produced, including full particles 310 in which the delivery gene 311 is properly packaged within the viral vector particle, empty particles 320 in which the delivery gene is not packaged within the viral vector particle, partial particles 330 in which a defective delivery gene 331 is packaged, and contaminated viral vectors 340 containing contaminants 341, such as nucleic acid fragments or proteins derived from the host cell or plasmid.

[0026] After viral vector production, the culture medium is collected, debris is removed, and non-viral components are removed. The purified viral vector undergoes high-purity purification and concentration, allowing for quality confirmation. One of the most important aspects of quality confirmation is identifying the internal state of the purified viral vector. It is important to evaluate the internal state of the viral vector, such as full particles, empty particles, and partial particles, in purified viral vector formulations. Essentially, purified viral vector formulations must contain only full particles, and must not contain empty or partial particles that adversely affect efficacy and safety. Therefore, it is necessary to identify the internal state of the viral vector, such as full particles or empty particles, and then quantify the number or concentration of viral vector particles in the formulation for each internal state.

[0027] Figure 4 illustrates the state of a viral vector. After purification is complete, i.e., at the time of quality check, the viral vector should ideally consist only of viral vector 411 in which delivery gene 412 is properly packaged, as in sample 410. However, as in sample 420, full particles 421 and empty particles 422 may be mixed, and unpackaged delivery gene 423 may exist outside the viral vector. If no delivery gene is packaged into the viral vector at all, only empty viral vector 431 remains, as in sample 430. Furthermore, during the purification process, in addition to viral vector 441, as in sample 440, empty viral vector 442 and contaminants 445 may also be present. In reality, as in sample 450, in addition to full and empty particles, contaminants 455, partial particles 453 in which a portion of the delivery gene is missing, or contaminating viral vector 454 containing nucleic acid fragments or proteins other than the delivery gene may also be produced. Here, samples 410, 420 and 430 are schematic illustrations of purified samples, and samples 440 and 450 are schematic examples of the state of samples in the intermediate purification process.

[0028] In quality confirmation, it is desirable to evaluate the number, concentration, or ratio of viral vectors for each internal state as accurately as possible. For example, in sample 420, 5 / 8 of the total are full particles 421 and 3 / 8 are empty particles 422. Using the electrical detection zone method, the shape of the pulse signal generated when each viral vector particle passes through the pore changes depending on the internal state of the viral vector. By utilizing this property, the internal state of each viral vector particle passing through the pore can be identified (Non-Patent Document 2). Furthermore, by aggregating the results, the number and abundance ratio of each vector internal state can be quantified, as described above for sample 420. Note that each diagram in Figure 4 is a schematic representation of only the dominant or most abundant particles in each sample; viral vector particles and other particles in internal states not shown may also be included. For example, sample 410 may contain small amounts of empty particles, partial particles, or impurities.

[0029] In a first embodiment of the viral vector testing method of the present invention, a test sample is subjected to a specific process, and then the internal state of the viral vector in the test sample is identified. The following description will be given using an example of a teacher sample containing a teacher viral vector whose internal state is known, but a similar procedure can also be used for a test sample containing a test viral vector whose internal state is unknown.

[0030] In this embodiment, when training data is required, a training sample containing a large number of training viral vectors with known internal states is measured using a pore sensor such as that shown in FIG. 1 as training data, and the internal states are used as training labels to train a machine learning model. Then, a test sample containing an unknown viral vector with an unknown internal state is measured using a pore sensor such as that shown in FIG. 1 , and the results are input into the trained machine learning model, allowing the internal states of the unknown viral vectors to be identified by estimating them one by one. If there are multiple predetermined types of internal states, the identified viral vectors can be classified into one of the multiple internal states. Furthermore, by aggregating these results, the viral vectors contained in the test sample can be quantified for each internal state and their abundance ratios can be calculated. The multiple types of internal states can typically include two types: a state in which a delivery gene is packaged in a viral vector, and a state in which a gene is not packaged in a viral vector. If training data already exists or if training data is not required, testing for unknown viral vectors can be performed directly.

[0031] The detailed procedure of the first embodiment will be described. First, a teacher sample in which full particles are dominant or most abundant, such as sample 410, is measured using a pore sensor as shown in FIG. 1. For each pulse obtained, multiple feature quantities representing the characteristics of the pulse waveform are calculated, and these feature quantities are labeled as "full particles" to train a machine learning model. Similarly, for a teacher sample in which empty particles are dominant, such as sample 430, the same feature quantities as those for sample 410 are calculated for each pulse measured using a pore sensor, and these feature quantities are labeled as "empty particles" to train the machine learning model. The machine learning model trained in this manner becomes a trained machine learning model that has acquired the pulse waveform characteristics for both full particles and empty particles.

[0032] Then, for example, a test sample containing a mixture of full and partial particles, such as sample 420, is measured using the same pore sensor shown in Figure 1, and the same feature values ​​are calculated for each pulse waveform. These values ​​are then input into this trained machine learning model, which makes it possible to distinguish between full and empty particles for each pulse waveform, i.e., for each viral vector that has passed through the pore. By aggregating these results, the number and abundance ratio of full and empty particles contained in unknown sample 420 can be quantified.

[0033] Here, a purified viral vector sample is assumed as the training sample, and the quantification of the number of full particles and empty particles has been described. Similarly, if a training sample containing only partial particles or a training sample containing predominantly or abundantly viral vector particles packaged with a specific substance can be prepared, the number and abundance ratio of these particles can be quantified by similarly training a machine learning model. The above is an embodiment in which a purified sample with no or few impurities is used as training data to train a machine learning model, and similarly quantification is performed on a test sample with no or few impurities.

[0034] Furthermore, the present invention makes it possible to identify the internal state of viral vectors in samples containing contaminants, something that was not possible with conventional technology. The conceptual diagram in Figure 5 explains the relationship between viral vectors and contaminants from viral vector production to the completion of purification. In Figure 5, marks 591, 592, and 593 represent empty particles (E), full particles (F), and contaminants (G), respectively. E, F, and T (501 to 503) represent the state of the sample during virus production, E, F, and T (521 to 523) represent the state of the purified sample, and E, F, and T (511 to 513) represent the state of the intermediate sample during the purification process, each of which schematically represents the ratio of empty particles, full particles, and contaminants. Samples during virus production contain a high proportion of contaminants, while viral vectors become dominant in the purified sample after the purification process. In the first embodiment described above, in FIG. 5, the number of full particles and the number of empty particles in the test sample T2 are estimated using a trained machine learning model trained using E2 and F2 as teacher samples.

[0035] In Figure 5, for example, the training data acquisition sample E0 represents a state in which the produced empty particles are present among impurities. From there, purified sample E2 is produced through purification steps 551 and 561. Through these purification steps, the ratio of impurities decreases from E0 to E1 and then to E2, and the concentration of viral vectors per unit volume increases through concentration. The same applies to the training data samples F0, F1, and F2, or the test samples T0, T1, and T2.

[0036] For viral vector production, there is a need to identify the internal state of viral vectors not only in purified samples but also in intermediate samples during the purification process. With conventional techniques, it has been difficult to quantify the internal state of viral vectors contained in such samples. Furthermore, the first embodiment of the present invention also trains a machine learning model using purified samples E2 and F2 as training samples and tests a purified test sample T2, but does not identify the internal state of viral vectors in samples containing impurities.

[0037] Intermediate samples from production processes or samples from virus production contain contaminants, and contaminants are often the dominant component. Quantifying the internal state of viral vectors in such samples requires solving the following two problems: the first problem is the removal of pulse signals derived from the contaminants, and the second problem is the matching of matrix conditions.

[0038] The first problem is the removal of pulse signals derived from impurities. When a sample containing impurities, such as sample 440, is measured using a pore sensor such as that shown in FIG. 1, pulses derived from the impurities are generated in addition to pulses derived from the viral vector. In the first embodiment, the characteristics of the pulse waveform derived from empty particles can be measured using teacher sample E2, which is predominantly or predominantly empty particles, and the characteristics of the pulse waveform derived from full particles can be measured using teacher sample F2, which is predominantly or predominantly full particles. However, when measuring samples such as E0, E1, F0, and F1, which are predominantly or predominantly impurities, many pulses derived from impurities are measured, making it impossible to teach the characteristics of empty particles and full particles to a machine learning program.

[0039] The second problem is matching the matrices of the training data sample and the test sample. In pulse measurements using a pore sensor like the one shown in Figure 1, the pulse waveform changes depending on the matrix, even for the same particles. Figure 6 shows an example of a comparison of the pulse waveforms of Sample 1, in which standard polystyrene beads with a particle size of 50 nm are suspended in a matrix of phosphate buffer containing 0.125 M NaCl, and Sample 2, in which the same polystyrene beads are suspended in a matrix of phosphate buffer without NaCl, measured using a pore sensor with the same 100 nm pore diameter. These two measurements are the results of measuring the same particles, but differ only in the salt concentration of the matrix. Figure 6(a) shows the distribution of pulse width 614 and pulse height 612 for measurement result 611 of Sample 1, and Figure 6(b) shows a typical pulse waveform 616. Similarly, Figure 6(c) shows the distribution of pulse width 624 and pulse height 622 for measurement result 621 of Sample 2, and Figure 6(d) shows a typical pulse waveform 626. Here, each plot in Figures 6(a) and (c) represents one pulse signal. Also, 613 and 623 are histograms of pulse height, and 615 and 625 are histograms of pulse width. As such, the pulse shape changes even for the same particle if the matrix is ​​different. Training a machine learning model is premised on the assumption that the same pulse waveform characteristics are measured for the same particle. To achieve this, the matrix of the training sample and the matrix of the test sample must have the same composition.

[0040] In FIG. 5 , marks M0 (594), M1 (595), and M2 (596) represent matrices 0, 1, and 2 of the virus production sample, the intermediate sample in the purification process, and the purified sample, respectively. For example, matrix M1 of sample E1 is different from matrix M2 of sample E2. On the other hand, for example, purification processes 561 to 563 are the same process, so samples E2, F2, and T2 have the same matrix M2. This is why, as in the first embodiment, a machine learning model trained using samples E2 and F2, which are matrix M2, as training data, was able to estimate the internal state of the viral vector for test sample T2, which has the same matrix M2. However, for example, a machine learning model trained using samples E2 and F2, which are matrix M2, as training data, cannot estimate the internal state of the viral vector for sample T1, which has matrix M1.

[0041] (2. Method for Inspecting Viral Vectors (Second Embodiment)) In order to quantify the internal state of viral vectors in contaminants, it is necessary to overcome the first and second problems described above. Below, a second embodiment of the present invention for overcoming these problems will be described. FIGS. 7 and 8 illustrate an example in which a machine learning model is trained using samples F1 and E1 in FIG. 5 as teacher samples, and the trained model is used to quantify full particles and empty particles in a test sample T1.

[0042] In one example of the second embodiment of the present invention, three types of reagents are used. The first is a reaction reagent. The reaction reagent comprises magnetic beads, dispersed in a matrix, with an attachment that specifically binds to the viral vector attached or bound to its surface. The substrate of the magnetic beads can be any bead that moves in a liquid when a magnetic field is applied, such as a ferromagnetic material such as iron, nickel, or cobalt, or an alloy of various elements. The matrix composition of the reaction reagent can be any liquid composition that does not inhibit the binding between the viral vector and the magnetic beads. For example, phosphate buffer, Tris buffer, Good's buffer, etc. can be used, and protein components such as BSA or skim milk, or surfactants such as Tween 20, can be added as additives. The attachment that specifically binds to the viral vector can be an antibody, a lectin that binds to a specific sugar chain expressed on the surface of the viral vector, an ion exchange resin, or a hydrophobic interaction resin. Examples of lectins include Griffonia simplicifolia lectin I (GSL I), Lotus tetragolonobus Lectin (LTL), Hippeastrum hybrid Lectin (HHL), Wheat Germ Agglutinin Triticum vulgaris (WGA), Griffonia simplicifolia lectin I (GSL I), Lotus tetragolonobus Lectin (LTL), Hippeastrum hybrid Lectin (HHL), Wheat Germ Agglutinin Triticum Examples of resins that can be used include Bacillus vulgaris (WGA), Maackia amurensis lectin (MAL-I), etc. Examples of ion exchange resins that can be used include Diethylaminoethyl (DEAE), Quaternary ammonium (Q), Diethylaminopropyl (ANX), Carboxymethyl (CM), Sulfopropyl (SP), and Methyl sulfonate (S). Examples of hydrophobic interaction resins that can be used include Phenyl, Butyl, Octyl, Ether, and Isopropyl.These may be used alone or in combination.

[0043] The second is a washing solution. The washing solution may be a liquid of any composition as long as it does not have both the property of cleaving the bond between the viral vector and the attached substance and the property of cleaving the attached magnetic beads from the substrate of the attached magnetic beads. For example, Tween (registered trademark) 20, Span (registered trademark) 20, Brij (registered trademark) 52, Myrj (registered trademark) 45, Triton (registered trademark) X, CHAPS, CHAPSO, etc. can be used.

[0044] The third is an elution reagent that cleaves the bond between the viral vector and the attached material and dissolves it into the matrix. The elution reagent can be any substance, such as an acid, alkali, salt, non-polar solvent, or reducing agent, as long as it has the property of cleaving the bond between the attached magnetic beads and the viral vector.

[0045] An example of the process according to the second embodiment of the present invention will be described with reference to Figures 7 and 8. In the following description, the substance attached to or bound to the magnetic beads will be referred to as an antibody. In addition, in the following description, the magnetic beads may be simply referred to as "magnetic beads."

[0046] First, a purification process intermediate sample F containing mainly full particles and impurities is prepared. １The test sample is mixed with a reaction reagent to generate a reaction solution (step S701). In this state, magnetic beads 813, viral vector 811, and impurities 812 are dispersed in a matrix 814. The matrix 814 is a mixture of the matrix of sample F1 and the matrix of the reaction reagent. An immune reaction then occurs in the reaction solution to form a conjugate 821 of magnetic beads and viral vector particles (step S702). Next, a magnetic field is applied to the reaction solution container, immobilizing the conjugate 821 near the inner wall of the reaction vessel 830 (step S703), forming a magnetically fixed conjugate 832. The source of the applied magnetic field is not particularly limited, and may be a permanent magnet 833 or an electromagnet. Next, with the conjugate immobilized, the matrix 814 is replaced with a cleaning solution (step S704), and after washing, the cleaning solution is removed. 841 denotes a magnetically fixed conjugate. As shown in Figure 8, impurities contained in the reaction solution are removed from the reaction chamber 840. Next, an elution reagent 851 is introduced into the reaction chamber 850 with the conjugates immobilized (step S705). This causes the viral vector 853 particles to separate from the immobilized magnetic beads 852 and disperse in the elution reagent, forming a vector dispersion eluate (step S706). Next, a viral vector dispersion eluate 862 is introduced into one of the chambers of the pore sensor shown in Figure 1, and an electrolyte 861 is introduced into the other chamber of the pore sensor, establishing electrical conductivity via the pore 863 (step S707). In step S707, the electrolyte introduced into the chamber on the opposite side of the pore from the elution solution may be the elution reagent or any other electrically conductive solution.

[0047] In this state, when a voltage is applied between the electrodes of the pore sensor, the current flowing through the pore changes transiently each time a viral vector dispersed in the viral vector dispersion eluate passes through the pore, and a pulse signal 865 such as that shown in Figure 2 is measured by the ammeter 151. Then, a feature quantity expressing the shape characteristics of each pulse signal is calculated. While the pulse height 208 and pulse width 209 shown in Figure 2 are examples of feature quantities of the pulse waveform, the feature quantities used in the present invention are not limited to these, and any quantity that can express the pulse waveform, such as the pulse area, bias, or skewness, may be used.

[0048] Steps S701 to S708 are common to both training the machine learning model using a teacher sample and identifying the internal state of a test sample viral vector using a trained machine learning model. When training the machine learning model using a teacher sample, after step S708, the calculated feature values ​​are labeled with a teacher label, in this example, "full particle," to train the machine learning model 870. Similarly, the teacher sample E1 is measured using steps S701 to S708, and the results are labeled with a teacher label, "empty particle," to train the machine learning model. The machine learning model generated in this manner becomes a trained learning model that identifies the pulse waveform features of full particles and empty particles. The machine learning model used in the present invention may use any algorithm, such as a support vector machine, a linear discriminant transform, a decision tree, or deep learning. Furthermore, the training method may involve calculating and training the feature values ​​as described above, or the pulse waveform data itself may be provided as teacher data.

[0049] By inputting the feature values ​​obtained as a result of measuring the test sample T1 in the procedure of steps S701 to S708 into such a trained machine learning model, it is possible to identify each full particle and each empty particle contained in the sample T1. Furthermore, by aggregating the identification results, it is possible to estimate the number or abundance ratio of full particles and empty particles for a sample such as T2 shown in Figure 4.

[0050] The second embodiment of the present invention solves two problems necessary for quantifying the internal state of viral vectors contained in a purification process intermediate sample containing impurities. The first problem, impurity removal, can be achieved by removing and washing the matrix from the reaction solution in step S704. The second problem, ensuring uniformity of the matrix, can be achieved by using the elution reagent added in step S705 as the matrix for measurement in step S707 for all of teacher sample E1, teacher sample F2, and test sample T1. Additionally, this method of the present invention can increase the concentration of viral vectors in the viral vector dispersion eluate relative to that in the teacher sample or test sample by reducing the amount of elution reagent introduced in step S705 compared to the amount of test sample introduced in step S701.

[0051] (3. Viral Vector Testing Method (Third Embodiment)) In one example of the third embodiment of the present invention, first, measurement and machine learning model training are performed using the method of steps S701 to S709 on teacher sample E2 and teacher sample F2, i.e., finished samples with few impurities. Using this trained machine learning model, the internal state of the viral vector is analyzed for test sample T1, which is a purification process intermediate sample with many impurities. Even in this manner, the impurities are still removed in step S704, and the matrix used during measurement is the same as the viral vector dispersion eluate, so the two problems required for quantifying the purification process intermediate sample containing the above-mentioned impurities are solved. The essence of the third embodiment of the present invention is to solve the above two problems by removing impurities from a teacher sample, in which the internal state of the contained viral vector is known, and a test sample, in which the internal state of the contained viral vector is unknown, and measuring them using the same matrix. FIG. 7 shows an example of the third embodiment in which measurement by the electrical sensing zone method and training and identification of a machine learning model are performed in steps S708, S709, and S710.

[0052] As mentioned above, the removal of impurities, which is one of the problems of the prior art, can be achieved by removing and washing the matrix from the reaction solution in step S704, so the concentration of impurities in the generated test virus vector dispersion eluate is very low or substantially free of impurities. Therefore, as long as the teacher virus vector and test virus vector are dispersed in a matrix having the same composition, there are no particular limitations on the method for measuring the teacher virus vector and test virus vector. In addition to the pore sensor of the present invention, conventional methods such as transmission electron microscopy (TEM), ultracentrifugation analysis, flow cytometry, and mass spectrometry can also be suitably used.

[0053] Therefore, in another example of the third embodiment of the present invention, steps S701 to S706 are performed on a teacher sample, and then the teacher virus vector is dispersed in an elution reagent, and measurement results of the teacher sample are obtained by measuring the teacher virus vector using a method such as transmission electron microscopy, ultracentrifugation analysis, flow cytometry, or mass spectrometry in step S706. Similarly, steps S701 to S706 are performed on a test sample, and then steps S701 to S706 are performed on the test sample, and then the test virus vector is dispersed in an elution reagent, and the test sample is measured using the same method as the teacher sample, and measurement results of the test sample are obtained. The internal state of the test virus vector in the test sample may be estimated by evaluating the similarity between the measurement results of the teacher sample and the measurement results of the test sample.

[0054] The relationship between the training data and the test sample in the first to third embodiments is shown in Figure 9. In the first embodiment, the trained machine learning model 921 was trained using the training data of measurement results of empty particles and full particles from samples E2 and F2, each of which has few impurities, and then identifies and quantifies the internal state of the viral vector in the test sample T2. In this case, both the training data sample and the test sample have few impurities (the first problem described above) and there is no matrix mismatch (the second problem described above), so the sample measurement results can be used directly for training.

[0055] In the second embodiment, the internal state of viral vectors in a test sample T1 was identified and quantified using a trained machine learning model 912 trained using the measurement results of teacher samples E1 and F1, which contain many impurities, as teacher data. In the third embodiment, the internal state of viral vectors in a test sample T1, which contains many impurities, was identified and quantified using a trained machine learning model 923 trained using the measurement results of teacher samples E2 and F2, which contain few impurities, as teacher data. In the second and third embodiments of the present invention, magnetic collection (step S703), washing (step S704), and measurement using an eluate (steps S705 to S707) enable the identification and quantification of the internal state of viral vectors in a viral vector sample containing many impurities.

[0056] (4. Method for Testing Viral Vectors (Fourth Embodiment)) In the first to third embodiments, teacher labels were attached to the teacher data to train the machine learning model. For this reason, supervised learning was used as the machine learning model. However, due to various constraints in the purification process, in reality, it may not be possible to create a teacher sample containing only a single type of viral vector. In the fourth embodiment of the present invention, an unsupervised machine learning algorithm is used. In this embodiment, as shown in FIG. 10 , after performing the processes of steps S701 to S708 for the test sample T1, the machine learning model 1000 finds differences in pulse shapes caused by differences in the internal state of the viral vector, classifies them into clusters, and then classifies the pulses by internal state without training using teacher data. The method of tallying the number of pulses classified by the internal state of the viral vector and estimating the number of particles contained in the test sample T1 for each internal state is the same as in the first to third embodiments. Examples of unsupervised machine learning algorithms include, but are not limited to, k-means clustering, hierarchical clustering, density-based spatial clustering of applications with noise (DBSCAN), principal component analysis, and self-organizing maps.

[0057] (5. Method for Testing Viral Vectors (Fifth Embodiment)) In a fifth embodiment of the present invention, when there are limitations on the preparation of samples for acquiring training data, a semi-supervised machine learning algorithm may be used. Examples of the semi-supervised machine learning algorithm include, but are not limited to, self-learning, semi-supervised support vector machines, and PUC (Positive and Unlabeled Classification). For example, as shown in Figure 11, in a case where a purified sample F2 with few impurities can be produced for full particles, but only a mixed sample Y2 (1121) of empty and full particles can be obtained for empty particles, the measurement results of the training sample F2 (522) can be used as a positive example in which it is known that almost all of the particles are full particles, and as a negative example in which the unknown particles contain a mixture of full particles and empty particles that are the training target, but the ratio is unknown. By using these as training data 1102 to train the PUC machine learning algorithm 1103, it is possible to identify and quantify the internal state of the viral vector contained in the test sample T1 in the middle of the purification process, which contains impurities. Similarly, for example, if only a purified teacher sample, teacher sample F, and teacher sample Z (1131) containing a mixture of an empty sample and impurities, are available, the measurement results of teacher sample F (522) are similarly used as a positive example, and a negative example in which the training target empty particles are mixed among unknown particles but the ratio is unknown. By using these as teacher data 1104 to train the PUC machine learning algorithm 1105, it is possible to identify and quantify the internal state of the viral vector contained in the test sample T in the intermediate purification process containing impurities. Note that the marks in Figures 9 to 11 are the same as those in the legend 590 in Figure 5. Details of the PUC machine learning algorithm are as described in Patent Documents 6 and 7. However, the present invention is not limited to the algorithms described in these patent documents, and a wide range of semi-supervised learning algorithms may be used in the fifth embodiment.

[0058] Generally, when identifying the internal state of a viral vector using machine learning, it is rare to obtain ideal training samples for machine learning, such as E1, F1, E2, and F2 in Figure 5. Furthermore, even if a sample can be prepared, it is often impossible to assign appropriate training labels using a reference method. This is because potential reference methods, such as transmission electron microscopy (Patent Document 1), ultracentrifugation analysis (Patent Document 2), capillary electrophoresis (Patent Document 3), flow cytometry (Patent Document 4), and hydrophilic interaction chromatography (Patent Document 5), are often unable to measure viral vectors contained in samples containing impurities. In the present invention, the first to fifth embodiments described above can be selectively used or combined depending on the training samples that can be generated depending on the viral vector purification process, thereby enabling the quantification of the internal state of viral vectors contained in samples in various states during the purification process, which was not possible with conventional techniques.

[0059] 7 , a machine learning model was trained, but this was essentially for the purpose of evaluating the degree of similarity between the internal states of a teacher viral vector contained in a teacher sample, whose internal state is known, and a test viral vector contained in a test sample. In the present invention, the internal state of a test viral vector in a test sample may be estimated and quantified by a method other than training a machine learning model, for example, by directly comparing pulse feature quantities obtained from the measurement results of the teacher sample with pulse feature quantities obtained from the measurement results of the test sample.

[0060] (6. Machine Learning System for Testing Viral Vectors) Another aspect of the present invention relates to a machine learning system for carrying out the method for testing viral vectors of the present invention. The machine learning system is configured to identify the internal state of a viral vector contained in a test sample by analyzing a pulse signal measured each time the viral vector contained in the test sample passes through the pore using a pore sensor having one pore, two chambers connected to both sides of the pore, and a pair of electrodes installed in each of the two chambers; the system further comprises: a teacher sample and a reaction reagent including magnetic beads having an attachment attached to the magnetic bead surface that specifically binds to a teacher viral vector having a known internal state, to prepare a teacher reaction solution, whereby a teacher conjugate between the teacher viral vector and the magnetic beads is generated in the teacher reaction solution; a magnetic field is applied from the outside of a teacher sample container containing the teacher reaction solution, and the teacher reaction solution is replaced with a washing solution while the teacher conjugate is fixed in the container, and the washing solution is then removed; and an elution reagent is introduced into the teacher sample container to release the teacher viral vector from the teacher conjugate, thereby producing a teacher viral vector dispersed eluate in which the teacher viral vector is dispersed in the elution reagent; The teacher virus vector dispersion eluate is introduced into one of the chambers of the pore sensor, and an electrolyte is introduced into the other of the chambers of the pore sensor, and then a voltage is applied to the electrode pair; the characteristics of the shape of the teacher pulse signal due to the current transient change that occurs in the electrode pair when the teacher virus vector in the teacher virus vector dispersion eluate passes through the pore are used as teacher data, and the known internal state is used as a teacher label for training, thereby creating a trained machine learning model.

[0061] The method for testing viral vectors that can be implemented in the machine learning system is as described in each embodiment of the method for testing viral vectors of the present invention, and detailed description thereof will be omitted here. In the machine learning system, the program for creating a trained machine learning model can be executed by a computer and can be recorded on any storage medium.

[0062] (7. Method for Preparing Test Virus Vector Dispersion Eluate) Another aspect of the present invention relates to a method for preparing a test virus vector dispersion eluate. The method for preparing a test virus vector dispersion eluate includes the steps of: preparing a test reaction solution by mixing a test sample containing a test virus vector whose internal state is unknown with a reaction reagent containing magnetic beads having an attachment that specifically binds to the virus vector on the surface of the magnetic beads; generating a test binding complex between the test virus vector and the attached magnetic beads in the test reaction solution; applying a magnetic field from the outside of a test sample container containing the test reaction solution to immobilize the test binding complex in the container; replacing the test reaction solution with a washing solution and then removing the washing solution; and introducing an elution reagent into the test sample container to liberate the test virus vector from the test binding complex, thereby producing a test virus vector dispersion eluate in which the test virus vector is dispersed in the elution reagent.

[0063] As described above, the removal of contaminants, which is one of the problems of the prior art, can be achieved by removing and washing the matrix from the reaction solution in step S704, so that the concentration of contaminants in the generated test virus vector dispersion eluate is very low or substantially free of contaminants. Therefore, it is possible to provide a test virus vector sample that can be suitably used not only with the pore sensor of the present invention, but also with conventional methods such as transmission electron microscopy (TEM), ultracentrifugation analysis, flow cytometry, and mass spectrometry.

[0064] The present invention will be described in more detail below with reference to examples, but the present invention is not limited to the following examples in any way.

[0065] The results of experiments verifying the effectiveness of the present invention are described in Figures 12 to 14. Here, particles 1210, in which λDNA 1211 was physically bound to polystyrene beads whose surfaces were modified with mouse IgG antibodies, and particles 1220, in which λDNA was not bound to polystyrene beads whose surfaces were also modified with mouse IgG antibodies, were likened to viral vectors with different internal states and were distinguished using the method of the present invention. Hereinafter, particles 1210 will be referred to as DNA-bound particles, and particles 1220 will be referred to as DNA-free particles. The reaction reagent was prepared by suspending anti-mouse IgG antibody-modified magnetic beads and DNA-bound or DNA-free particles in 20 mM phosphate buffer (pH 7.5) supplemented with 150 mM NaCl, 0.5% BSA, 1.5% PEG-20K, and a surfactant. The elution reagent used was 100 mM glycine HCl (pH 2.7) supplemented with 10 mM NaCl and a surfactant.

[0066] As shown in FIG. 13 , sample 1311 represents a test sample containing DNA-free particles mixed with a reaction reagent, and sample 1315 represents a test sample containing DNA-bound particles mixed with a reaction reagent (corresponding to step S701). For each sample, steps S701 to S708 were performed as shown in FIG. 13 , and measurements were performed using the pore sensor according to the present invention, as shown in measurements 1351 and 1355 (corresponding to step S708). FIG. 14 shows measurement results 1410 for measurement 1351 and 1420 for measurement 1355. Measurement results 1410 and 1420 each represent a single pulse, showing the distribution of pulse height and pulse width for each pulse. These results were used to train machine learning model 1400 (corresponding to step S309), using teacher labels for the measured sample and teacher labels for "DNA-free" and "DNA-bound" for each measurement result. Both the pulse height and pulse width shown in FIG. 14 are examples of feature quantities in the present invention, but in training the machine learning model in this verification experiment, 40 feature quantities in addition to these were used.

[0067] To evaluate the performance of the trained machine learning model, cross-validation was performed for each pulse, i.e., for each particle to be identified in the test sample, using the leave-one-out method, and the results are shown as confusion matrix 1401. As a result, the F-measure was 0.875, indicating that highly accurate identification was possible.

[0068] Although each embodiment of the present invention has been described above, various inventions can be formed by appropriately combining the multiple components disclosed in each embodiment. For example, some components may be omitted from all the components shown in the embodiments. Furthermore, components of different embodiments may be appropriately combined.

[0069] 100 Sensor 101 Test object 110 Chamber 111 Inlet 112 Electrode 120 Chamber 121 Inlet 122 Electrode 140 Pore 141 Diaphragm 150 Amplifier 151 Ammeter 152 Voltage source 200 Baseline 201 Time (horizontal axis) 202 Current value (vertical axis) 208 Pulse height 209 Pulse width 210 Pulse signal 310 Full particle 311 Delivery gene 320 Empty particle 330 Partial particle 331 Delivery gene with deletion 340 Contaminating viral vector 341 Contaminant 410 Sample 411 Viral vector 412 Delivery gene 420 Sample 421 Full particle 422 Empty particle 423 Unpackaged delivery gene 430 Sample 431 Empty viral vector 440 Sample 441 Viral vector 442 Empty viral vector 445 Contaminants 450 Sample 453 Partial particles 454 Contaminating viral vector 455 Contaminants 501 Viral vector production sample 502 Viral vector production sample 503 Viral vector production sample 511 Purification process intermediate sample 512 Purification process intermediate sample 513 Purification process intermediate sample 521 Purified sample 522 Purified sample 523 Purified sample 551 Purification process 552 Purification process 553 Purification process 561 Purification process 562 Purification process 563 Purification process 591 Empty particle 592 Full particle 593 Contaminants 594 Matrix 0 595 Matrix 1 596 Matrix 2 611 Measurement results for sample 1 612 Pulse height 613 Pulse height histogram 614 Pulse width 615 Pulse width histogram 616 Pulse waveform 621 Measurement results for sample 2 622 Pulse height 623 Pulse height histogram 624 Pulse width 625 Pulse width histogram 626 Pulse waveform 810 Reaction vessel 811 Viral vector 812 Contaminants 813 Magnetic beads 814 Matrix 820 Reaction vessel during immune reaction 821 Bound body 830 Reaction vessel when magnetic field is applied 832 Magnetic field-immobilized bound body 833 Permanent magnet 840 Reaction vessel when washing solution is removed 841 Magnetic field-immobilized bound body when washing solution is removed 850 Reaction vessel during elution 851 Elution reagent 852 Magnetic field-immobilized magnetic beads853 Viral vector 861 Electrolyte 862 Viral vector dispersion eluate 863 Pore 865 Pulse signal 870 Machine learning model 912 Trained machine learning model 921 Trained machine learning model 923 Trained machine learning model 1000 Machine learning model 1102 Training data 1103 PUC machine learning algorithm 1104 Training data 1121 Mixed sample Y2 1131 Training sample Z2 1210 DNA-bound particle 1211 λDNA 1220 DNA-free particle 1311 Sample 1312 DNA-free particle 1313 Magnetic bead 1315 Sample 1316 DNA-bound particle 1317 Magnetic bead 1321 Sample 1325 Sample 1331 Sample 1335 Sample 1341 Sample 1345 Sample 1351 Measurement 1355 Measurement 1400 Machine learning model 1401 Confusion matrix 1410 Measurement result 1411 Example of pulse waveform of particle without DNA 1420 Measurement result 1421 Example of pulse waveform of particle bound to DNA

Claims

1. A method for testing viral vectors, comprising: identifying the internal state of viral vectors in a test sample using a pore sensor having one pore, two chambers connected to both sides of the pore, and a pair of electrodes installed in each of the two chambers; and using a reaction reagent containing attached magnetic beads having a substance that specifically binds to viral vectors attached to the magnetic bead surface; when a reaction solution is prepared by mixing the test sample and the reaction reagent, a bond between the viral vector and the attached magnetic beads is formed in the reaction solution; applying a magnetic field from outside a container containing the reaction solution to fix the bond in the container, and then replacing the reaction solution with a washing solution and removing the washing solution; furthermore, introducing an elution reagent into the container to liberate the viral vectors from the bond, thereby producing a viral vector dispersion eluate in which the viral vectors are dispersed in the elution reagent; introducing the viral vector dispersion eluate into one of the chambers of the pore sensor and an electrolyte into the other chamber of the pore sensor, and then applying a voltage to the pair of electrodes; A method for testing viral vectors, characterized by identifying the internal state of multiple viral vectors that have passed through the pores from the characteristics of the shape of a pulse signal caused by a transient current change that occurs in the electrode pair when the viral vectors in the viral vector dispersion eluate pass through the pores.

2. The method for testing viral vectors according to claim 1, wherein the internal state includes a plurality of predetermined types, and the method further comprises classifying the identified viral vector into one of the plurality of types of the internal state.

3. The method for testing viral vectors described in claim 2, further characterized in that the abundance ratio of the viral vectors for each internal state in the test sample is estimated by aggregating the number of particles of the viral vectors for each classified internal state.

4. A method for testing a viral vector according to claim 1 or 2, characterized in that the attached substance is an antibody that specifically binds to the viral vector.

5. A method for testing a viral vector according to claim 1 or 2, characterized in that the attached substance is a lectin that specifically binds to the sugar chain expressed on the surface of the viral vector.

6. The method for testing a viral vector according to claim 2, characterized in that the internal state includes at least two types: a state in which a delivery gene is packaged in the viral vector, and a state in which a gene is not packaged in the viral vector.

7. A machine learning system for testing viral vectors, configured to identify the internal state of a viral vector contained in a test sample by analyzing a pulse signal measured each time the viral vector contained in the test sample passes through the pore using a pore sensor having one pore, two chambers connected to both sides of the pore, and a pair of electrodes installed in each of the two chambers; the system further comprises the steps of: mixing a teacher sample with a reaction reagent containing magnetic beads having an attachment attached to the magnetic surface that specifically binds to a teacher viral vector having a known internal state to create a teacher reaction solution, whereby a teacher conjugate between the teacher viral vector and the attached magnetic beads is generated in the teacher reaction solution; applying a magnetic field from the outside of a teacher sample container containing the teacher reaction solution to fix the teacher conjugate in the container, and then replacing the teacher reaction solution with a washing solution and removing the washing solution; and introducing an elution reagent into the teacher sample container to release the teacher viral vector from the teacher conjugate, thereby producing a teacher viral vector dispersion elution solution in which the teacher viral vector is dispersed in the elution reagent; A machine learning system characterized by being configured to introduce the teacher virus vector dispersion eluate into one of the chambers of the pore sensor and an electrolyte into the other of the chambers of the pore sensor, apply a voltage to the electrode pair, and use as teacher data the characteristics of the shape of the teacher pulse signal due to the current transient change that occurs in the electrode pair when the teacher virus vector in the teacher virus vector dispersion eluate passes through the pore, and train using the known internal state as a teacher label to create a trained machine learning model.

8. The machine learning system further comprises: mixing a test sample containing a test viral vector whose internal state is unknown with the reaction reagent to prepare a test reaction solution, whereby a test binding entity between the test viral vector and the attached magnetic beads is generated in the test reaction solution; applying a magnetic field from the outside of a test sample container containing the test reaction solution, so that the test binding entity is fixed in the container, replacing the test reaction solution with the washing solution, and then removing the washing solution; further introducing the elution reagent into the test sample container to liberate the test viral vector from the test binding entity, thereby producing a test viral vector dispersion eluate in which the test viral vector is dispersed in the elution reagent; introducing the test viral vector dispersion eluate into one of the chambers of the pore sensor and an electrolyte into the other of the chambers of the pore sensor, and then applying a voltage to the electrode pair; The machine learning system of claim 7, wherein the system is configured to estimate the internal state of the test virus vector contained in the test sample by inputting into the trained machine learning model the characteristics of the shape of an unknown pulse signal due to a current transient change that occurs in the electrode pair when the test virus vector in the test virus vector dispersion eluate passes through the pore.

9. A method for preparing a test viral vector dispersion eluate, comprising: mixing a test sample containing a test viral vector whose internal state is unknown with a reaction reagent containing magnetic beads having a substance attached to the surface of the magnetic beads that specifically binds to the viral vector to produce a test reaction solution, thereby producing a test binding complex between the test viral vector and the magnetic beads in the test reaction solution; applying a magnetic field from the outside of a test sample container that contains the test reaction solution, so that the test binding complex is fixed in the container, replacing the test reaction solution with a washing solution and then removing the washing solution; and introducing an elution reagent into the test sample container to liberate the test viral vector from the test binding complex, thereby producing a test viral vector dispersion eluate in which the test viral vector is dispersed in the elution reagent.

10. A method for testing viral vectors, comprising identifying the internal state of a test viral vector in a test sample containing a test viral vector whose internal state is unknown, using a reaction reagent containing magnetic beads having a substance attached to the surface of the magnetic beads that specifically binds to the viral vector, and an elution reagent that separates the viral vector bound to the magnetic beads; the method comprises carrying out the following steps (1) and (2): (1) mixing a teacher sample containing a teacher viral vector whose internal state is known with the reaction reagent to prepare a teacher reaction solution, generating a teacher conjugate between the teacher viral vector and the magnetic beads in the teacher reaction solution; applying a magnetic field from the outside of a teacher container containing the teacher reaction solution, and, with the teacher conjugate fixed in the teacher container, replacing the teacher reaction solution with a washing solution and then removing it; and further introducing the elution reagent into the teacher container to release the teacher viral vector from the teacher conjugate in the elution reagent; (2) When the test sample and the reaction reagent are mixed to prepare a test reaction liquid, a test binding entity between the test viral vector and the attached magnetic beads is generated in the test reaction liquid, a magnetic field is applied from the outside of a test vessel containing the test reaction liquid, and while the test binding entity is fixed in the test vessel, the test reaction liquid is replaced with a washing liquid and then removed, and the elution reagent is further introduced into the test vessel, thereby liberating the test viral vector from the test binding entity in the elution reagent; the method then comprises: (3) measuring the teacher viral vector and the test viral vector dispersed in matrices having the same composition, and comparing the characteristics of the measurement results to estimate the internal state of the test viral vector in the test sample.

Images