Machine learning methods for classifying cells

Abstract

Methods are provided for classifying cells, such as T cells, using machine learning methods, which can be used to classify different subsets or types of cells in a mixed population of cells. TIFF2022546396000008.tif75138

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of priority to U.S. Provisional Patent Application No. 62 / 894,463, entitled "MACHINE LEARNING METHODS FOR CLASSIFYING CELLS," filed August 30, 2019, the contents of which are incorporated by reference in their entirety.

[0002] INCORPORATION BY REFERENCE OF SEQUENCE LISTING This application is filed with an electronic Sequence Listing, which is provided in a file named 735042012340SeqList.TXT, created on August 28, 2020, and is 22,780 bytes in size. The information in the electronic format of the Sequence Listing is incorporated by reference in its entirety.

[0003] Field The present disclosure, in some aspects, relates to methods for classifying cells, such as T cells, using machine learning methods, which can be used to classify different subsets or types of cells in a mixed population of cells. [Background technology]

[0004] background Adoptive cell therapy, which involves treatment with genetically engineered immune cells such as T cells (e.g., CD4+ T cells and / or CD8+ T cells) engineered to express recombinant receptors such as chimeric antigen receptors (CARs), is commonly used to treat a variety of diseases and conditions. Existing methods for characterizing cells, for example during ex vivo production or prior to administration of cell therapy, typically rely on direct handling or manipulation of the cells, such as by incubation with reagents that may interfere with cell quality or function, such as magnetic beads or other immunoaffinity-based reagents. Such methods can be time-consuming and may risk contamination of the therapeutic cell composition. Improved methods for efficiently sorting or characterizing cells, such as in the context of cell therapy, are needed. Provided herein are methods and devices that meet this need. Summary of the Invention

[0005] overview Provided herein is a method for classifying T cells, the method comprising: receiving image data associated with a first T cell of a population of cells containing T cells; determining one or more input features from the image data, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and applying the one or more input features as input to a process configured to classify the first T cell as belonging to a first group or a second group based on the one or more input features.

[0006] Provided herein is a method comprising: receiving image data associated with a first T cell of a population of cells containing T cells; determining a classification of the first T cell as belonging to a first group or a second group; determining one or more input features from the image data, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and generating a feature map having one or more dimensions, each dimension of the one or more dimensions being associated with one or more of the morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and training a convolutional neural network based on the feature map and the determined classification.

[0007] In some embodiments, the process includes applying a convolutional neural network trained using a method described herein, and applying one or more input features determined from image data associated with the first T cell as input to the process includes applying the one or more input features to the convolutional neural network.

[0008] Provided herein is a method of classifying T cells, the method comprising: receiving image data associated with a first T cell of a population of cells containing T cells; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and applying the one or more input features as input to a process to classify the first T cell as belonging to a first group or a second group, the process comprising a convolutional neural network trained on the one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof.

[0009] Provided herein is a method comprising: receiving image data associated with a first T cell of a population of cells containing T cells; determining a classification of the first T cell as belonging to a first group or a second group; determining one or more input features from the image data, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and training a neural network on the input features and the determined classification.

[0010] In some embodiments, the process includes applying a neural network trained according to the methods described herein, wherein applying one or more input features determined from the image data associated with the first T cell as input to the process includes applying the one or more input features to the neural network.

[0011] Provided herein is a method of classifying T cells, the method comprising: receiving image data associated with a first T cell of a population of cells containing T cells; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and applying the one or more input features as input to a process to classify the first T cell as belonging to a first group or a second group, the process comprising a neural network trained on the one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof.

[0012] Provided herein is a method comprising the steps of receiving image data associated with a first T cell of a population of cells containing T cells; determining a classification of the first T cell as belonging to a first group or a second group; and determining a hyperplane for a support vector machine, the hyperplane indicating a decision boundary between the first group and the second group, wherein the hyperplane is determined based on one or more input features determined from the image data, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof.

[0013] In some embodiments, the process includes applying a support vector machine using the hyperplane determined using the methods described herein, and applying one or more input features determined from the image data associated with the first T cell as input to the process includes applying the one or more input features to the support vector machine.

[0014] Provided herein is a method of classifying T cells, the method comprising: receiving image data associated with a first T cell of a population of cells containing T cells; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and applying the one or more input features as input to a process to classify the first T cell as belonging to a first group or a second group, the process comprising a support vector machine trained on the one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof.

[0015] Provided herein is a method comprising: receiving image data associated with a first T cell of a population of cells containing T cells; determining a classification of the first T cell as belonging to a first group or a second group; and determining a random forest for the classification process, wherein the random forest comprises one or more decision trees, and the classification process associates one or more input features determined from the image data associated with the first T cell with the classification of the first T cell as belonging to the first group or the second group, wherein the decision tree of the one or more decision trees is determined based on the one or more input features, and the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof.

[0016] In some embodiments, the process comprises a random forest determined using a method described herein, and applying one or more input features determined from image data associated with the image of the first T cell as input to the process comprises applying the one or more input features to the random forest.

[0017] Provided herein is a method of classifying T cells, the method comprising: receiving image data associated with a first T cell of a population of cells containing T cells; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and applying the one or more input features as input to a process to classify the first T cell as belonging to a first group or a second group, the process comprising a random forest classification process on the one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof.

[0018] Provided herein is a method that includes receiving image data associated with a first T cell of a population of cells containing T cells, the image data including one or more of phase image data, intensity image data, and overlay image data; determining a classification of the first T cell as belonging to a first group or a second group; generating a feature map having one or more dimensions from the image data; and training a convolutional neural network based on the feature map and the determined classification.

[0019] Provided herein is a method of classifying T cells, the method comprising: receiving image data associated with a first T cell of a population of cells containing T cells, the image data comprising one or more of phase image data, intensity image data, and overlay image data; and applying the image data as input to a process to classify the first T cell as belonging to a first group or a second group, the process comprising a convolutional neural network trained on image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the images comprising image data comprising one or more of phase image data, intensity image data, and overlay image data.

[0020] Provided herein is a method for classifying T cells, the method comprising: receiving image data associated with each cell of a plurality of T cells of a population of cells, the image data comprising one or more of phase image data, intensity image data, and overlay image data; and applying the image data from each cell of the plurality of T cells as input to a process to classify each cell of the plurality of T cells as belonging to a first group or a second group, the process comprising a convolutional neural network trained on image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the image data comprising one or more of phase image data, intensity image data, and overlay image data.

[0021] Provided herein is a method for classifying T cells, the method comprising: receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, the image data comprising one or more of phase image data, intensity image data, and overlay image data; and applying the image data from each of the plurality of T cells as input to a process configured to classify each of the plurality of T cells as belonging to a first group or a second group based on the image data.

[0022] Provided herein is a method comprising: receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, the image data comprising one or more of phase image data, intensity image data, and overlay image data; determining a classification of each of the plurality of T cells as belonging to a first group or a second group; and training a convolutional neural network based on the image data and the determined classification.

[0023] In some embodiments, the process comprises applying a convolutional neural network trained using a method described herein, wherein applying image data associated with each cell of the plurality of T cells as input to the process comprises applying the image data to the convolutional neural network.

[0024] In some embodiments, the convolutional neural network is further trained on image data associated with T cells known to belong to a first group and T cells known to belong to a second group, where the image data has been manipulated by zooming, tilting, and / or rotating the image data.

[0025] In some embodiments, a process such as those described herein is configured to classify about or at least 70, 75, 80, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100% of the T cells in a population of cells containing T cells as belonging to a first group or a second group. In some embodiments, each cell of the T cells in the population of cells containing T cells is classified by the process. In some embodiments, each cell of a plurality of T cells in a population of cells containing T cells is classified by the process. In some embodiments, the plurality of T cells comprises about, or at least, 70, 75, 80, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100% of the T cells in a population of cells comprising T cells.

[0026] Provided herein is a method for generating a cellular dataset for training a machine learning model, the method comprising: (a) providing a mixed population of cells containing at least a first and a second different cell type, wherein at least the first cell type expresses at least one surface molecule that is not expressed by other cell types in the mixed population of cells; (b) contacting the mixed population of cells with a first multimerization reagent reversibly bound to a plurality of first binding agents, each of the first binding agents comprising a monovalent binding site capable of binding the surface molecule expressed by the first cell type; and (c) isolating one or more cells of the first cell type that are bound to the first binding agent by immunoaffinity chromatography, thereby obtaining a first cellular dataset, wherein the isolation is performed under conditions that reversibly dissociate the first multimerization reagent from the first binding agent, and the first cellular dataset is substantially free of the first multimerization reagent and the first binding agent. In some embodiments, at least one first cell type expresses a recombinant surface molecule. In some embodiments, the recombinant surface molecule is a recombinant receptor. In some embodiments, the recombinant receptor is a chimeric antigen receptor or a T cell receptor. In some embodiments, prior to (c), the method produces a second population of cells that are not bound to the first binding agent, the second population of cells containing cells that do not express the surface molecule expressed by the first cell type, thereby obtaining a second dataset of cells. In some embodiments, each of the first and second cell types expresses at least one surface molecule that is not expressed by other cell types in the mixed population of cells.

[0027] In some embodiments, the method further comprises: (d) contacting the mixed population of cells with a second multimerization reagent reversibly bound to a plurality of second binding agents, each of the second binding agents comprising a monovalent binding site capable of binding a surface molecule expressed by the second cell type; and (e) isolating one or more cells of the second cell type that are bound to the second binding agent by immunoaffinity-based chromatography, thereby obtaining a second dataset of cells, wherein the isolation is performed under conditions for reversibly dissociating the second multimerization reagent from the second binding agent, wherein the second dataset of cells is substantially free of the second multimerization reagent and the second binding agent. In some embodiments, the mixed population of cells further comprises a third cell type that expresses at least one surface molecule that is not expressed by other cell types in the mixed population of cells, and the method further comprises: (f) contacting the mixed population of cells with a third multimerization reagent reversibly bound to a plurality of third binding agents, each of which comprises a monovalent binding site capable of binding a surface molecule expressed by the third cell type; and (e) isolating one or more cells of the third cell type that are bound to the third binding agent by immunoaffinity chromatography, whereby a third cell data set is obtained, wherein the isolation is performed under conditions that reversibly dissociate the third multimerization reagent from the third binding agent, and the third cell data set is substantially free of the third multimerization reagent and the third binding agent. In some embodiments, this is repeated for one or more additional different cell types.

[0028] In some embodiments, prior to (c), the method produces a second population of cells that are not bound to the first binding agent, the second population of cells containing cells that do not express a surface molecule expressed by the first cell type, and the method further comprises: (d) contacting the second population of cells with a second multimerization reagent reversibly bound to a plurality of second binding agents, each of the second binding agents comprising a monovalent binding site capable of binding a surface molecule expressed by the second cell type; and (e) isolating one or more cells of the second cell type that are bound to the second binding agent by immunoaffinity-based chromatography, thereby obtaining a second dataset of cells, the isolation being performed under conditions for reversibly dissociating the second multimerization reagent from the second binding agent, wherein the second dataset of cells is substantially free of the second multimerization reagent and the second binding agent. In some embodiments, prior to (e), the method produces a third population of cells that are not bound to the second binding agent, the third population of cells comprising cells that do not express a surface molecule expressed by the first cell type and the second cell type, and the method further comprises: (d) contacting the third population of cells with a third multimerization reagent reversibly bound to a plurality of third binding agents, each of the third binding agents comprising a monovalent binding site capable of binding a surface molecule expressed by the third cell type; and (e) isolating one or more cells of the third cell type that are bound to the third binding agent by immunoaffinity chromatography, thereby obtaining a third data set of cells, the isolation being performed under conditions for reversibly dissociating the third multimerization reagent from the third binding agent, the third data set of cells being substantially free of the third multimerization reagent and the third binding agent. In some embodiments, this is repeated for one or more different cell types.

[0029] In some embodiments, the first and second cell types are one of (i) CD4+ T cells and (ii) CD8+ T cells, and the second cell type is the other of (i) CD4+ T cells and (ii) CD8+ T cells.

[0030] In some embodiments, the contacting step is carried out by adding the cells to a chromatography column comprising a stationary phase on which a multimerization reagent reversibly bound to a binding agent is immobilized (e.g., a first multimerization reagent reversibly bound to a first binding agent, a second multimerization reagent reversibly bound to a second binding agent, or a third multimerization reagent reversibly bound to a third binding agent). In some embodiments, the isolating step comprises eluting the cells from the chromatography column.

[0031] In some embodiments, the first binding agent further comprises a binding partner capable of forming a reversible bond with the first multimerization reagent, wherein the binding partner comprises the sequence of amino acids set forth in SEQ ID NO:6, 7, 8, 9, or 10, and the first multimerization reagent comprises streptavidin, a streptavidin mutein, avidin, or an avidin mutein. In some embodiments, the second binding agent further comprises a binding partner capable of forming a reversible bond with the second multimerization reagent, wherein the binding partner comprises the sequence of amino acids set forth in SEQ ID NO:6, 7, 8, 9, or 10, and the second multimerization reagent comprises streptavidin, a streptavidin mutein, avidin, or an avidin mutein. In some embodiments, the third binding agent further comprises a binding partner capable of forming a reversible bond with the third multimerization reagent, wherein the binding partner comprises the sequence of amino acids set forth in SEQ ID NO: 6, 7, 8, 9, or 10, and the third multimerization reagent comprises streptavidin, a streptavidin mutein, avidin, or an avidin mutein. In some embodiments, the streptavidin mutein comprises the sequence of amino acids set forth in SEQ ID NO: 12, 13, 15, or 16. In some embodiments, the reversible bond is formed within a range of about 10 -2 ~about 10 -13 Dissociation constants (K D In some embodiments, the monovalent binding site is a Fab fragment, an sdAb, an Fv fragment, or a single-chain Fv fragment. In some embodiments, the binding between the monovalent binding site and the surface molecule is about 10 -3 ~about 10-7 Dissociation constants (K D In some embodiments, the binding between the monovalent binding site and the surface molecule is about 3×10 -5 sec -1 In some embodiments, the reversible dissociation comprises the addition of a competing reagent. In some embodiments, the competing reagent is biotin or a biotin analog.

[0032] In some embodiments, the method further comprises receiving image data associated with cells of a first cellular dataset, wherein the image data comprises one or more of phase image data, intensity image data, and overlay image data. In some embodiments, receiving image data associated with cells of a second cellular dataset, wherein the image data comprises one or more of phase image data, intensity image data, and overlay image data. In some embodiments, the method further comprises receiving image data associated with cells of a third cellular dataset, wherein the image data comprises one or more of phase image data, intensity image data, and overlay image data. In some embodiments, the method further comprises receiving image data associated with cells of the first cellular dataset; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof. In some embodiments, the method further comprises receiving image data associated with cells of a second cellular dataset; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof. In some embodiments, the method further comprises receiving image data associated with cells of a third cellular dataset; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof.

[0033] In some embodiments, the method further comprises generating a feature map having one or more dimensions from the image data; and training a convolutional neural network based on the feature map. In some embodiments, the method further comprises training the neural network on the input features. In some embodiments, the method further comprises determining a hyperplane associated with a support vector machine, the determining step being based on the one or more input features. In some embodiments, the method further comprises determining one or more decision trees of a random forest, the determining step being based on the one or more input features.

[0034] In some embodiments, image data is received from about 70, 75, 80, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100% of the T cells of a population of cells containing T cells, or from at least 70, 75, 80, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100% of the T cells. In some embodiments, the image data includes one or more of phase image data, intensity image data, and overlay image data. In some embodiments, the image data is obtained using differential digital holographic microscopy (DDHM). In some embodiments, the image data is obtained using an about 20x objective lens. In some embodiments, the image data is obtained using a CCD camera.

[0035] In some embodiments, the image data associated with the T cells is used to determine one or more input features, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof.In some embodiments, the one or more input features are a cell aspect ratio, a cell depth, a cell area, a cell descriptor, a cell identifier, an image identifier, an object identifier, a centroid along the X axis, a centroid along the Y axis, a cell circularity, a cell compactness, a normalized aspect ratio, a cell elongation, a cell diameter, a peak diameter, a hu moment invariant 1, a hu moment invariant 2, a hu moment invariant 3, a hu moment invariant 4, a hu moment invariant 5, a hu moment invariant 6, a hu moment invariant 7, a mean intensity contrast, a mean entropy, a mean intensity, a mean intensity homogeneity, an intensity contrast of an image of cells, an intensity correlation of an image of cells, an intensity entropy of an image of cells, an intensity homogeneity of an image of cells, a maximum cell intensity, a mean cell intensity, a minimum cell intensity Intensity skewness of the image of the cell, Intensity smoothness of the image of the cell, Intensity variance of the image of the cell, Intensity uniformity of the image of the cell, Plane at which the intensity of the cell is maximum, Indication that the cell is located along the boundary of the field of view, Indication that the refractive peak is located along the boundary of the field of view, Mass eccentricity of the cell, Maximum optical height of the cell in radians, Maximum optical height of the cell in microns, Mean optical height of the cell in radians, Mean optical height of the cell in microns, Normalized optical height of the cell, Minimum optical height of the cell in radians, Minimum optical height of the cell in microns, Variance of the optical height of the phase of the image of the cell in radians, Position of the image of the cell in microns The phase features may include one or more of: phase optical height variance, cell optical volume, cell refraction peak area, refraction peak area normalized by cell area, cell refraction peak number, cell refraction peak intensity, cell normalized refraction peak height, perimeter, cell phase image mean intensity contrast, cell phase image mean entropy, cell mean phase, cell mean phase uniformity, cell phase intensity contrast, cell phase correlation, cell phase entropy feature, cell phase homogeneity, cell phase skewness, cell phase smoothness, cell phase uniformity, cell mean radius, cell radius variance, and cell normalized radius variance.In some embodiments, the input features include about 70, 65, 60, 55, 50, 45, 40, 35, 30, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or 1 input features, or at least 70, 65, 60, 55, 50, 45, 40, 35, 30, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or 1 input features. In some embodiments, the input features include about 1 to about 70, about 1 to about 60, about 1 to about 50, about 1 to about 40, about 1 to about 30, about 1 to about 20, about 1 to about 15, about 1 to about 10, or about 1 to about 5 input features. In some embodiments, the input features include less than, or less than about, 70, 65, 60, 55, 50, 45, 40, 35, 30, 25, 20, 15, 10, or 5 input features. In some embodiments, the input features include less than, or less than about, 20, 15, 10, or 5 input features.

[0036] In some embodiments, the one or more input features are morphological features of the image data selected from one or more of cell aspect ratio, cell area, cell circularity, cell compactness, normalized aspect ratio, cell elongation, cell diameter, hu moment invariant 1, hu moment invariant 2, hu moment invariant 3, hu moment invariant 4, hu moment invariant 5, hu moment invariant 6, hu moment invariant 7, perimeter, cell radius mean, cell radius variance, and normalized radius variance. In some embodiments, the one or more input features is cell area. In some embodiments, the one or more input features are optical features of the image data selected from one or more of the following: cell diameter, maximum cell intensity, average cell intensity, minimum cell intensity, cell mass eccentricity, maximum cell optical height in radians, maximum cell optical height in microns, average cell optical height in radians, average cell optical height in microns, normalized cell optical height, minimum cell optical height in radians, minimum cell optical height in microns, cell optical volume, area of ​​the cell's refractive peak, area of ​​the refractive peak normalized by the cell's area, number of cell's refractive peaks, intensity of the cell's refractive peak, and normalized cell's refractive peak height. In some embodiments, one or more of the input features are intensity features of the image data selected from one or more of the following: mean intensity contrast, mean entropy, mean intensity, mean intensity uniformity, intensity contrast of the image of the cells, intensity correlation of the image of the cells, intensity entropy of the image of the cells, intensity uniformity of the image of the cells, intensity skewness of the image of the cells, intensity smoothness of the image of the cells, intensity variance of the image of the cells, plane of maximum cell intensity and intensity uniformity of the image of the cells.In some embodiments, the one or more input features are phase features of the image data selected from one or more of: a variance of the optical height of the phase of an image of cells in radians, a variance of the optical height of the phase of an image of cells in microns, a mean intensity contrast of the phase image of cells, a mean entropy of the phase image of cells, a mean phase of cells, a mean phase uniformity of cells, a phase intensity contrast of cells, a phase correlation of cells, a phase entropy feature of cells, a phase homogeneity of cells, a phase skewness of cells, a phase smoothness of cells, and a phase uniformity of cells. In some embodiments, the one or more input features are or include a phase correlation of cells. In some embodiments, the one or more inputs are system features of the image data selected from one or more of: a cell depth, an identified cell, an image identifier, a cell descriptor, a centroid along the X-axis, a centroid along the Y-axis, an object identifier, an indication that a cell is located along a boundary of the field of view, and an indication that a refractive peak is located along a boundary of the field of view. In some embodiments, the one or more input features are or include a cell depth.

[0037] In some embodiments, the first group and the second group are defined by one or more cellular attributes selected from live, dead, CD4+, CD8+, recombinant receptor positive, or recombinant receptor negative, and the first group and the second group comprise at least one different attribute. In some embodiments, one of the first or second groups comprises the attribute live, and the other group comprises the attribute dead. In some embodiments, the attribute dead comprises non-viable cells and debris. In some embodiments, the attribute live comprises single viable cells or clusters of viable cells. In some embodiments, the one or more input features comprise cell phase correlation, cell area, cell number of refractive peaks, cell phase skewness, peak diameter, refractive peak area normalized by cell area, cell radius variance, cell image intensity uniformity, cell compactness, cell phase intensity contrast, normalized radius variance, and cell circularity. In some embodiments, the one or more input features comprise cell phase correlation. In some embodiments, the one or more input features comprise cell area.

[0038] In some embodiments, one of the first or second group comprises the attribute CD4+, and the other group comprises the attribute CD8+. In some embodiments, the one or more input features comprise average phase uniformity of cells, peak diameter, normalized refractive peak height of cells, average phase of cells, refractive peak area normalized by area of ​​cells, minimum optical height of cells in microns, compactness of cells, circularity of cells, phase smoothness of cells, intensity homogeneity of the image of cells, plane at which intensity of cells is maximum, area of ​​refractive peak of cells, phase correlation of cells, depth of cells, intensity contrast of the image of cells, intensity uniformity of the image of cells, normalized radial variance, intensity smoothness of the image of cells, phase intensity contrast of cells, maximum optical height of cells, average intensity contrast of cells, average intensity uniformity of cells, intensity skewness of the image of cells, hu moment invariant 1, intensity variance of the image of cells, average entropy, and intensity correlation of the image of cells. In some embodiments, the one or more input features comprise cell depth, cell image intensity contrast, cell image intensity uniformity, normalized radial variance, cell image intensity smoothness, cell phase intensity contrast, cell maximum optical height in microns, cell mean intensity contrast, cell mean intensity uniformity, cell image intensity skewness, hu moment invariant 1, cell image intensity variance, mean entropy, and cell image intensity correlation. In some embodiments, the one or more input features comprise cell depth. In some embodiments, the one or more input features comprise cell mean phase uniformity, cell peak diameter, cell normalized refractive peak height, cell mean phase, refractive peak area normalized by cell area, cell minimum optical height in microns, cell compactness, cell circularity, cell phase smoothness, cell image intensity uniformity, cell image plane of maximum cell intensity, cell refractive peak area, and cell phase correlation.

[0039] In some embodiments, one of the first or second groups comprises attributable recombinant receptor-positive cells, and the other group comprises attributable recombinant receptor-negative cells.

[0040] In some embodiments, the population of cells containing T cells comprises a population of T cells enriched or purified from a biological sample or a population of mixed T cell subtypes. In some embodiments, the enriched or purified population is obtained by mixing T cell populations enriched or purified from biological samples. In some embodiments, the biological sample comprises a whole blood sample, a buffy coat sample, a peripheral blood mononuclear cell (PBMC) sample, an unfractionated T cell sample, a lymphocyte sample, a leukocyte sample, a blood purification therapy product, or a leukopheresis therapy product. In some embodiments, the population of cells containing T cells comprises primary cells obtained from a subject. In some embodiments, the population of cells containing T cells comprises a population of T cells transduced with a vector comprising a recombinant receptor. In some embodiments, the population of T cells from which one or more T cells are sorted comprises a population of T cells undergoing manufacturing to generate a therapeutic T cell composition. In some embodiments, the manufacturing comprises an incubation step after transduction of the T cell population with a vector comprising the recombinant receptor. In some embodiments, the recombinant receptor is a chimeric antigen receptor (CAR).

[0041] In some embodiments, the T cells known to belong to the first group and the T cells known to belong to the second group are or comprise T cells that have undergone manufacturing to generate a therapeutic T cell composition. In some embodiments, the manufacturing of the T cells known to belong to the first group and the T cells known to belong to the second group is the same as or nearly the same as the manufacturing of the population of T cells into which one or more T cells are sorted.

[0042] In some embodiments, one or more T cells of the population of cells containing T cells are sorted at different time points during the incubation process. In some embodiments, one or more T cells of the population of cells containing T cells are sorted continuously throughout the incubation period. In some embodiments, sorting one or more T cells of the population of cells containing T cells is performed in a closed system. In some embodiments, the closed system is sterile. In some embodiments, the closed system is automated. [The present invention 1001] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, the image data comprising one or more of phase image data, intensity image data, and overlay image data; applying the image data from each cell of the plurality of T cells as input to a process to classify each cell of the plurality of T cells as belonging to a first group or a second group, the process comprising a convolutional neural network trained on image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the image data comprising one or more of phase image data, intensity image data, and overlay image data. A method for classifying T cells, comprising: [The present invention 1002] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, the image data comprising one or more of phase image data, intensity image data, and overlay image data; applying the image data from each cell of the plurality of T cells as input to a process configured to classify each cell of the plurality of T cells as belonging to a first group or a second group based on the image data. A method for classifying T cells, comprising: [The present invention 1003] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, the image data comprising one or more of phase image data, intensity image data, and overlay image data; determining the classification of each cell of said plurality of T cells as belonging to a first group or a second group; training a convolutional neural network based on the image data and the determined classification; A method comprising: [The present invention 1004] The process includes applying a convolutional neural network trained using the method of the present invention 1003; applying the image data associated with each cell of the plurality of T cells as an input to the process includes applying the image data to the convolutional neural network. The method of the present invention 1002. [The present invention 1005] 10. The method of any of claims 1001, 1003 and 1004, wherein the convolutional neural network is further trained on image data associated with T cells known to belong to the first group and T cells known to belong to the second group, wherein the image data has been manipulated by zooming, tilting and / or rotating the image data. [The present invention 1006] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; applying the one or more input features for each cell of the plurality of T cells as input to a process configured to classify each cell of the plurality of T cells as belonging to a first group or a second group based on the one or more input features. A method for classifying T cells, comprising: [The present invention 1007] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells; determining the classification of each cell of said plurality of T cells as belonging to a first group or a second group; determining from the image data one or more input features for each cell of the plurality of T cells, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and generating a feature map having one or more dimensions, each dimension of the one or more dimensions associated with one or more of the morphological features of the image data, the optical features of the image data, the intensity features of the image data, the phase features of the image data, the system features of the image data, or any combination thereof; training a convolutional neural network based on the feature map and the determined classification; A method comprising: [The present invention 1008] the process includes applying a convolutional neural network trained using the method of the present invention 1007; applying the one or more input features determined from the image data associated with each cell of the plurality of T cells as input to the process comprises applying the one or more input features to the convolutional neural network. The method of the present invention 1006. [The present invention 1009] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; applying the one or more input features for each cell of the plurality of T cells as input to a process to classify each of the plurality of T cells as belonging to a first group or a second group, the process comprising a convolutional neural network trained on one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof. A method for classifying T cells, comprising: [The present invention 1010] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells; determining the classification of each cell of said plurality of T cells as belonging to a first group or a second group; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; training a neural network on said input features and said determined classification; A method comprising: [The present invention 1011] the process includes applying a neural network trained using the method of the present invention 1010; applying the one or more input features determined from the image data associated with each cell of the plurality of T cells as input to the process comprises applying the one or more input features to the neural network. The method of the present invention 1006. [The present invention 1012] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; applying the one or more input features for each cell of the plurality of T cells as input to a process to classify each cell of the plurality of T cells as belonging to a first group or a second group, the process comprising a neural network trained on one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof. A method for classifying T cells, comprising: [The present invention 1013] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells; determining the classification of each cell of said plurality of T cells as belonging to a first group or a second group; and determining a hyperplane for a support vector machine, the hyperplane representing a decision boundary between the first group and the second group; Including, the hyperplane is determined based on one or more input features determined from the image data for each cell of the plurality of T cells, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; method. [The present invention 1014] the process includes applying a support vector machine using a hyperplane determined using the method of the present invention 1013; applying the one or more input features determined from the image data associated with each cell of the plurality of T cells as input to the process comprises applying the one or more input features to the support vector machine. The method of the present invention 1006. [The present invention 1015] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; applying the one or more input features for each cell of the plurality of T cells as input to a process to classify each cell of the plurality of T cells as belonging to a first group or a second group, the process comprising a support vector machine trained on one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof. A method for classifying T cells, comprising: [The present invention 1016] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells; determining the classification of each cell of said plurality of T cells as belonging to a first group or a second group; and determining a random forest for a classification process, the random forest comprising one or more decision trees, the classification process relating one or more input features determined from the image data associated with each cell of the plurality of T cells to the classification of each cell of the plurality of T cells as belonging to the first group or as belonging to the second group; Including, a decision tree of the one or more decision trees is determined based on one or more input features for each cell of the plurality of T cells, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; method. [The present invention 1017] the process includes a random forest determined using the method of the present invention 1016; applying the one or more input features determined from the image data associated with the image of the first T cell as input to the process comprises applying the one or more input features to the random forest. The method of the present invention 1006. [The present invention 1018] receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; applying the one or more input features for each cell of the plurality of T cells as input to a process to classify each cell of the plurality of T cells as belonging to a first group or a second group, the process comprising a random forest classification process on one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof. A method for classifying T cells, comprising: [The present invention 1019] The method of any one of claims 1006 to 1018, wherein the image data includes one or more of phase image data, intensity image data, and overlay image data. [The present invention 1020] 1019. The method of any one of claims 1001 to 1019, wherein the image data is phase image data and intensity image data. [The present invention 1021] The method according to any one of claims 1001 to 1020, wherein the image data is phase image data. [The present invention 1022] 1022. The method of any of claims 1001 to 1021, wherein said plurality of T cells comprises about, or at least 70, 75, 80, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100% of the T cells in said population of cells comprising T cells. [The present invention 1023] (a) providing a mixed population of cells comprising at least a first and a second different cell type, wherein at least said first cell type expresses at least one surface molecule that is not expressed by other cell types in said mixed population of cells; (b) contacting the mixed population of cells with a first multimerization reagent reversibly bound to a plurality of first binding agents, each of the first binding agents comprising a monovalent binding site capable of binding the surface molecule expressed by the first cell type; and (c) isolating one or more cells of the first cell type that are bound to the first binding agent by immunoaffinity-based chromatography, thereby obtaining a first data set of cells, wherein the isolation is performed under conditions for reversibly dissociating the first multimerization reagent from the first binding agent. Including, the first cell dataset is substantially free of the first multimerization reagent and the first binding agent; A method for generating a dataset of cells for training machine learning models. [The present invention 1024] 1024. The method of claim 1023, wherein said at least one first cell type expresses a recombinant surface molecule. [The present invention 1025] 1025. The method of claim 1024, wherein said recombinant surface molecule is a recombinant receptor. [The present invention 1026] 1026. The method of claim 1025, wherein said recombinant receptor is a chimeric antigen receptor or a T cell receptor. [The present invention 1027] 1027. The method of any of claims 1023 to 1026, wherein prior to (c), the method produces a second population of cells that are not bound to the first binding agent, wherein the second population of cells comprises cells that do not express the surface molecule expressed by the first cell type, thereby obtaining a second data set of cells. [The present invention 1028] 1028. The method of any of claims 1023 to 1027, wherein said first and second cell types each express at least one surface molecule that is not expressed by other cell types in said mixed population of cells. [The present invention 1029] (d) contacting the mixed population of cells with a second multimerization reagent reversibly bound to a plurality of second binding agents, each of the second binding agents comprising a monovalent binding site capable of binding the surface molecule expressed by the second cell type; and (e) isolating one or more cells of the second cell type that are bound to the second binding agent by immunoaffinity-based chromatography, thereby obtaining a second data set of cells, wherein the isolating is performed under conditions for reversibly dissociating the second multimerization reagent from the second binding agent. further comprising the second cellular dataset is substantially free of the second multimerization reagent and the second binding agent; Any of the methods of 1023 to 1028 of the present invention. [The present invention 1030] said mixed population of cells further comprising a third cell type that expresses at least one surface molecule that is not expressed by other cell types in said mixed population of cells; The method comprises: (f) contacting the mixed population of cells with a third multimerization reagent reversibly bound to a plurality of third binding agents, each of the third binding agents comprising a monovalent binding site capable of binding the surface molecule expressed by the third cell type; and (e) isolating one or more cells of the third cell type that are bound to the third binding agent by immunoaffinity-based chromatography, thereby obtaining a third data set of cells, wherein the isolation is performed under conditions for reversibly dissociating the third multimerization reagent from the third binding agent. further comprising the third cellular dataset is substantially free of the third multimerization reagent and the third binding agent; The method of the present invention 1029. [The present invention 1031] 1029. The method of any of claims 1023 to 1029, repeated for one or more additional different cell types. [The present invention 1032] prior to (c), the method produces a second population of cells that are not bound to the first binding agent, the second population of cells comprising cells that do not express the surface molecule expressed by the first cell type; The method comprises: (d) contacting the second population of cells with a second multimerization reagent reversibly bound to a plurality of second binding agents, each of the second binding agents comprising a monovalent binding site capable of binding the surface molecule expressed by the second cell type; and (e) isolating one or more cells of the second cell type that are bound to the second binding agent by immunoaffinity-based chromatography, thereby obtaining a second data set of cells, wherein the isolating is performed under conditions for reversibly dissociating the second multimerization reagent from the second binding agent. further comprising the second cellular dataset is substantially free of the second multimerization reagent and the second binding agent; Any of the methods of 1023 to 1028 of the present invention. [The present invention 1033] prior to (e), the method produces a third population of cells that are not bound to the second binding agent, the third population of cells comprising cells that do not express the surface molecule expressed by the first cell type and the second cell type; The method comprises: (f) contacting the third population of cells with a third multimerization reagent reversibly bound to a plurality of third binding agents, each of the third binding agents comprising a monovalent binding site capable of binding the surface molecule expressed by the third cell type; and (g) isolating one or more cells of the third cell type that are bound to the third binding agent by immunoaffinity-based chromatography, thereby obtaining a third data set of cells, wherein the isolation is performed under conditions for reversibly dissociating the third multimerization reagent from the third binding agent. further comprising the third cellular dataset is substantially free of the third multimerization reagent and the third binding agent; The method of the present invention 1032. [The present invention 1034] The method of any of claims 1023-1028, 1032 and 1033, repeated for one or more additional different cell types. [This invention 1035] Any of the methods of claims 1023 to 1034, wherein the first and second cell types are one of (i) CD4+ T cells and (ii) CD8+ T cells, and the second cell type is the other of (i) CD4+ T cells and (ii) CD8+ T cells. [The present invention 1036] 1036. The method of any of claims 1023 to 1035, wherein the contacting step is carried out by adding the cells to a chromatography column comprising a stationary phase on which the first multimerization reagent reversibly bound to the first binding agent is immobilized. [This invention 1037] 1036. The method of any of claims 1029 to 1036, wherein the contacting step is carried out by adding the cells to a chromatography column comprising a stationary phase on which the second multimerization reagent reversibly bound to the second binding agent is immobilized. [The present invention 1038] Any of the methods of claims 1030 to 1037, wherein the contacting step is carried out by adding the cells to a chromatography column comprising a stationary phase on which the third multimerization reagent reversibly bound to the third binder is immobilized. [This invention 1039] 1039. The method of any one of claims 1023 to 1038, wherein said isolating step comprises eluting said cells from said chromatography column. [The present invention 1040] receiving image data associated with the cells of the first cellular dataset, the image data including one or more of phase image data, intensity image data, and overlay image data. Any of the methods of claims 1023 to 1039, further comprising: [This invention 1041] receiving image data associated with the cells of the second cellular dataset, the image data including one or more of phase image data, intensity image data, and overlay image data. Any of the methods of inventions 1027 to 1040, further comprising: [The present invention 1042] receiving image data associated with the cells of the third cellular dataset, the image data including one or more of phase image data, intensity image data, and overlay image data. Any of the methods of inventions 1030 to 1041, further comprising: [This invention 1043] receiving image data associated with cells of said first cellular dataset; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; Any of the methods of inventions 1023 to 1042, further comprising: [This invention 1044] receiving image data associated with cells of said second cellular dataset; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; Any of the methods of claims 1027 to 1043, further comprising: [This invention 1045] receiving image data associated with cells of said third cellular dataset; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; Any of the methods of 1030 to 1044 of the present invention further comprising: [The present invention 1046] generating a feature map from the image data, the feature map having one or more dimensions; training a convolutional neural network based on the feature map; 1043. The method of any of claims 1040, 1041 or 1042, further comprising: [This invention 1047] The method of any of 1043, 1044 or 1045, further comprising the step of training a neural network on said input features. [This invention 1048] determining a hyperplane associated with the support vector machine; the determining step is based on the one or more input features. The method of any of inventions 1043, 1044, or 1045. [This invention 1049] determining one or more decision trees of the random forest; the determining step is based on the one or more input features. The method of any of inventions 1043, 1044, or 1045. [The present invention 1050] The method of any of claims 1001 to 1022 and 1040 to 1049, wherein the image data is obtained using differential digital holographic microscopy (DDHM). [This invention 1051] The method according to any one of claims 1001 to 1022 and 1040 to 1050, wherein the image data is obtained using an objective lens of about 20x magnification. [This invention 1052] The method of any one of claims 1001 to 1022 and 1040 to 1051, wherein the image data is obtained using a CCD camera. [This invention 1053] The one or more input features include an aspect ratio of the cell, a depth of the cell, an area of ​​the cell, a cell descriptor, a cell identifier, an image identifier, an object identifier, a centroid along the X axis, a centroid along the Y axis, a circularity of the cell, a compactness of the cell, a normalized aspect ratio, an elongation of the cell, a diameter of the cell, a peak diameter, a hu moment invariant 1, a hu moment invariant 2, a hu moment invariant 3, a hu moment invariant 4, a hu moment invariant 5, a hu moment invariant 6, a hu moment invariant 7, a mean intensity contrast, a mean entropy, a mean intensity, a mean intensity uniformity, an intensity contrast of the image of the cell, an intensity correlation of the image of the cell, an intensity entropy of the image of the cell, an intensity homogeneity of the image of the cell, a maximum intensity of the cell, a mean intensity of the cell, a minimum intensity of the cell Intensity skewness of the image of the cell, intensity smoothness of the image of the cell, intensity variance of the image of the cell, intensity uniformity of the image of the cell, plane at which the intensity of the cell is maximum, indication that the cell is located along a boundary of the field of view, indication that a refractive peak is located along a boundary of the field of view, mass eccentricity of the cell, maximum optical height of the cell in radians, maximum optical height of the cell in microns, average optical height of the cell in radians, average optical height of the cell in microns, normalized optical height of the cell, minimum optical height of the cell in radians, minimum optical height of the cell in microns, variance of the optical height of the phase of the image of the cell in radians, variance of the optical height of the phase of the image of the cell in microns, optical volume of the cell. , the area of ​​the refraction peak of the cell, the refraction peak area normalized by the area of ​​the cell, the number of refraction peaks of the cell, the intensity of the refraction peak of the cell, the normalized refraction peak height of the cell, the perimeter, the average intensity contrast of the phase image of the cell, the average entropy of the phase image of the cell, the average phase of the cell, the average phase uniformity of the cell, the phase intensity contrast of the cell, the phase correlation of the cell, the phase entropy feature of the cell, the phase homogeneity of the cell, the phase skewness of the cell, the phase smoothness of the cell, the phase uniformity of the cell, the average radius of the cell, the radius variance of the cell, and the normalized radius variance of the cell. [This invention 1054] 10. The method of any of claims 1006-1022 and 1043-1053, wherein the input features include about or at least 70, 65, 60, 55, 50, 45, 40, 35, 30, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or 1 input features. [This invention 1055] The method of any of claims 1006 to 1022 and 1043 to 1054, wherein the input features include about 1 to about 70, about 1 to about 60, about 1 to about 50, about 1 to about 40, about 1 to about 30, about 1 to about 20, about 1 to about 15, about 1 to about 10, or about 1 to about 5 input features. [This invention 1056] 10. The method of any of claims 1006-1022 and 1043-1055, wherein the input features include less than or about 70, 65, 60, 55, 50, 45, 40, 35, 30, 25, 20, 15, 10, or 5 input features. [This invention 1057] The method of any of claims 1006-1022 and 1043-1056, wherein the input features include less than, or about, 20, 15, 10, or 5 input features. [This invention 1058] 8. The method of any one of claims 1006 to 1022 and 1043 to 1057, wherein the one or more input features are morphological features of the image data selected from one or more of: aspect ratio of the cell, area of ​​the cell, circularity of the cell, compactness of the cell, normalized aspect ratio, elongation of the cell, diameter of the cell, Hu moment invariant 1, Hu moment invariant 2, Hu moment invariant 3, Hu moment invariant 4, Hu moment invariant 5, Hu moment invariant 6, Hu moment invariant 7, perimeter, mean radius of the cell, variance of radius of the cell, and normalized radius variance. [This invention 1059] The method of any one of claims 1006 to 1022 and 1043 to 1058, wherein the one or more input features is an area of ​​the cell. [The present invention 1060] Any of the methods of inventions 1006 to 1022 and 1043 to 1059, wherein the one or more input features are optical features of the image data selected from one or more of: diameter of the cell, maximum intensity of the cell, average intensity of the cell, minimum intensity of the cell, mass eccentricity of the cell, maximum optical height of the cell in radians, maximum optical height of the cell in microns, average optical height of the cell in radians, average optical height of the cell in microns, normalized optical height of the cell, minimum optical height of the cell in radians, minimum optical height of the cell in microns, optical volume of the cell, area of ​​a refractive peak of the cell, area of ​​a refractive peak normalized by the area of ​​the cell, number of refractive peaks of the cell, intensity of a refractive peak of the cell, and normalized refractive peak height of the cell. [This invention 1061] Any of the methods of inventions 1006 to 1022 and 1043 to 1060, wherein the one or more of the input features are intensity features of the image data selected from one or more of mean intensity contrast, mean entropy, mean intensity, mean intensity uniformity, intensity contrast of the image of the cell, intensity correlation of the image of the cell, intensity entropy of the image of the cell, intensity homogeneity of the image of the cell, intensity skewness of the image of the cell, intensity smoothness of the image of the cell, intensity variance of the image of the cell, plane at which the intensity of the cell is maximum, and intensity uniformity of the image of the cell. [This invention 1062] 2. The method of any one of claims 1006 to 1022 and 1043 to 1061, wherein the one or more input features are phase features of the image data selected from one or more of: a variance of the phase optical height of the image of the cells in radians, a variance of the phase optical height of the image of the cells in microns, a mean intensity contrast of the phase image of the cells, a mean entropy of the phase image of the cells, a mean phase of the cells, a mean phase uniformity of the cells, a phase intensity contrast of the cells, a phase correlation of the cells, a phase entropy feature of the cells, a phase homogeneity of the cells, a phase skewness of the cells, a phase smoothness of the cells, and a phase uniformity of the cells. [This invention 1063] The method of any of claims 1006 to 1022 and 1043 to 1062, wherein said one or more input features is or comprises a phase correlation of said cells. [This invention 1064] Any of the methods of inventions 1006-1022 and 1043-1063, wherein the one or more inputs are system features of the image data selected from one or more of: cell depth, identified cell, image identifier, cell descriptor, centroid along the X axis, centroid along the Y axis, object identifier, an indication that the cell is located along a boundary of the field of view, and an indication that the refractive peak is located along a boundary of the field of view. [This invention 1065] The method of any of claims 1006 to 1022 and 1043 to 1064, wherein the one or more input features is or includes a cell depth. [The present invention 1066] Any of the methods of inventions 1001 to 1022 and 1050 to 1065, wherein the first group and the second group are defined by one or more cellular attributes selected from alive, dead, CD4+, CD8+, recombinant receptor positive, or recombinant receptor negative, and the first group and the second group comprise at least one different attribute. [This invention 1067] The method of any one of claims 1001 to 1022 and 1050 to 1066, wherein one of the first or second group includes the attribute survival, and the other group includes the attribute death. [The present invention 1068] The method of claim 1066 or claim 1067, wherein said attribute death comprises non-viable cells and debris. [The present invention 1069] The method of claim 1066 or claim 1067, wherein said attribute survival comprises a single survival cell or a cluster of survival cells. [The present invention 1070] Any of the methods of claims 1006 to 1022 and 1043 to 1069, wherein the one or more input features include a phase correlation of the cell, an area of ​​the cell, a number of refractive peaks of the cell, a phase distortion of the cell, a peak diameter, a refractive peak area normalized by the area of ​​the cell, a radial variance of the cell, an intensity uniformity of an image of the cell, a compactness of the cell, a phase intensity contrast of the cell, a normalized radial variance, and a circularity of the cell. [This invention 1071] The method of any of claims 1006 to 1022 and 1043 to 1070, wherein the one or more input features include a phase correlation of the cells. [This invention 1072] The method of any of claims 1006 to 1022 and 1043 to 1071, wherein the one or more input features include an area of ​​the cell. [This invention 1073] The method of any one of claims 1006 to 1022 and 1050 to 1066, wherein one of said first or second group comprises said attribute CD4+, and the other group comprises said attribute CD8+. [This invention 1074] 1073. Any of the methods of claims 1006-1022, 1043-1066, and 1073, wherein the one or more input features comprise an average phase uniformity of the cells, the peak diameter, a normalized refractive peak height of the cells, an average phase of the cells, a refractive peak area normalized by the area of ​​the cells, a minimum optical height of the cells in microns, a compactness of the cells, a circularity of the cells, a phase smoothness of the cells, an intensity homogeneity of the image of the cells, a plane at which the intensity of the cells is maximum, an area of ​​the refractive peak of the cells, a phase correlation of the cells, a depth of the cells, an intensity contrast of the image of the cells, an intensity uniformity of the image of the cells, a normalized radial variance, an intensity smoothness of the image of the cells, a phase intensity contrast of the cells, a maximum optical height of the cells, an average intensity contrast of the cells, an average intensity uniformity of the cells, an intensity skewness of the image of the cells, a hu moment invariant 1, an intensity variance of the image of the cells, an average entropy, and an intensity correlation of the image of the cells. [This invention 1075] Any of the methods of claims 1006-1022, 1043-1066, 1073, and 1074, wherein the one or more input features include a depth of a cell, an intensity contrast of an image of the cell, an intensity uniformity of an image of the cell, a normalized radial variance, an intensity smoothness of an image of the cell, a phase intensity contrast of the cell, a maximum optical height of the cell in microns, a mean intensity contrast of the cell, a mean intensity uniformity of the cell, an intensity skewness of the image of the cell, a hu moment invariant 1, an intensity variance of the image of the cell, a mean entropy, and an intensity correlation of the image of the cell. [This invention 1076] The method of any of claims 1006-1022, 1043-1066, and 1073-1075, wherein the one or more input features include cell depth. [This invention 1077] 1075. Any of the methods of claims 1006 to 1022, 1043 to 1066, 1073, and 1074, wherein the one or more input features comprise an average phase uniformity of the cell, a peak diameter, a normalized refractive peak height of the cell, an average phase of the cell, a refractive peak area normalized by the area of ​​the cell, a minimum optical height in microns, a compactness of the cell, a circularity of the cell, a phase smoothness of the cell, an intensity homogeneity of an image of the cell, a plane at which the intensity of the cell is maximum, an area of ​​a refractive peak of the cell, and a phase correlation of the cell. [This invention 1078] Any of the methods of claims 1006 to 1022 and 1050 to 1066, wherein one of the first or second group comprises cells positive for the attributable recombinant receptor, and the other group comprises cells negative for the attributable recombinant receptor. [This invention 1079] the population of cells comprising T cells A population of T cells enriched or purified from a biological sample or a population of mixed T cell subtypes, optionally obtained by mixing enriched or purified T cell populations from a biological sample. Any of the methods of 1001 to 1078 of the present invention, comprising: [The present invention 1080] The method of claim 1079, wherein the biological sample comprises a whole blood sample, a buffy coat sample, a peripheral blood mononuclear cell (PBMC) sample, an unfractionated T cell sample, a lymphocyte sample, a leukocyte sample, a blood purification therapy product, or a leukocyte apheresis therapy product. [This invention 1081] The method of any of claims 1001 to 1080, wherein said population of cells comprising T cells comprises primary cells obtained from a subject. [This invention 1082] The method of any of claims 1001 to 1081, wherein said population of cells comprising T cells comprises a population of T cells transduced with a vector comprising a recombinant receptor. [This invention 1083] Any of the methods of claims 1001 to 1082, wherein said population of T cells from which one or more T cells are sorted comprises a population of T cells that have undergone manufacturing to generate a therapeutic T cell composition. [This invention 1084] Any of the methods of inventions 1001 to 1083, wherein the T cells known to belong to the first group and the T cells known to belong to the second group are or comprise T cells undergoing manufacture to generate a therapeutic T cell composition. [This invention 1085] 1084. The method of claim 1084, wherein said production of said T cells known to belong to a first group and said T cells known to belong to a second group is identical or nearly identical to said production of said population of T cells into which one or more T cells are sorted. [The present invention 1086] 1086. The method of any of claims 1083 to 1085, wherein said producing comprises an incubation step after transduction of said T cell population with a vector comprising a recombinant receptor. [This invention 1087] The method of any one of claims 1024 to 1066 and 1078 to 1086, wherein the recombinant receptor is a chimeric antigen receptor (CAR). [This invention 1088] The method of claim 1086 or claim 1087, wherein said one or more T cells of the population of cells comprising a T cell are sorted at different time points during said incubation step. [This invention 1089] 108. The method of claim 1086 or 1087, wherein said one or more T cells of the population of cells comprising a T cell are continuously sorted throughout said incubation period. [The present invention 1090] The method of any one of claims 1001 to 1022 and 1050 to 1089, wherein sorting the one or more T cells of a population of cells comprising T cells is carried out in a closed system. [This invention 1091] 1090. The method of claim 1090, wherein said closed system is sterile. [This invention 1092] 1092. The method of claim 1090 or 1091, wherein said closed system is automated. [Brief explanation of the drawings]

[0043] [Figure 1A] Figures 1A and 1B show the viable cell count (VCC; x106 cells / mL), cell viability (%), and cell diameter (µm) assessed using continuous monitoring by differential DHM ("Continuous", line) or manual sampling ("Manual", dots) in experimental run 1 (Figure 1A) and run 2 (Figure 1B). The top panel shows the measurements for each, and the bottom panel shows the linear regression analysis and R2 and slope for comparing continuous monitoring and manual sampling. [Figure 1B] See legend to Figure 1A. [Figure 2A]Figures 2A-2D show the viable cell count (VCC; x106 cells / mL), cell viability (%), and cell diameter (µm) assessed using continuous monitoring by differential DHM ("Continuous", lines) or manual sampling ("Manual", dots) for CD4+ cells from Donor 1 in Experiment 1 (Figure 2A), Donor 2 in Experiment 1 (Figure 2B), or Donor 3 in Experiment 2 (Figure 2C), or CD8+ cells from Donor 3 in Experiment 2 (Figure 2D). The top panels show the measurements for each, and the bottom panels show the linear regression analysis and R2 and slope for comparing continuous monitoring and manual sampling. [Figure 2B] See legend to Figure 2A. [Figure 2C] See legend to Figure 2A. [Figure 2D] See legend to Figure 2A. [Figure 3] Figure 3 shows the viable cell count (VCC; x106 cells / mL) and cell viability (%) assessed using continuous monitoring by differential DHM in the automated expansion process compared to the manual expansion process. [Figure 4] Figure 4 shows the relative importance of features used by the gradient boosted random forest classifier to predict cell viability. [Figure 5A] Figures 5A-5C show the accuracy of a gradient-boosted random forest classifier for reproducing viability measurements over seeding time predicted by an alternative viability prediction model involving optical imaging (e.g., holograms) that also uses features extracted from cell images to predict viability on a cell-by-cell basis for test data by machine learning methods. Figure 5A shows the accuracy of a gradient-boosted random forest classifier compared to an alternative viability prediction model for predicting "survive" / "dead" classification as a function of seeding time. [Figure 5B]Figures 5A-5C show the accuracy of the gradient-boosted random forest classifier for reproducing viability measurements over seeding time predicted by an alternative viability prediction model involving optical imaging (e.g., holograms) that also uses features extracted from cell images to predict viability on a cell-by-cell basis for test data via machine learning methods. Figure 5B shows a comparison of the total "survival" percentage over time. "Actual": predictions made by the alternative viability prediction model; "Predicted": percentage predicted by the gradient-boosted random forest classifier. [Figure 5C] Figures 5A-5C show the accuracy of a gradient-boosted random forest classifier for reproducing viability measurements over seeding time predicted by an alternative viability prediction model involving optical imaging (e.g., holograms) that also uses features extracted from cell images to predict viability on a cell-by-cell basis for test data by machine learning methods. Figure 5C shows the relative difference between the predictions of the alternative viability prediction model and the predictions made by the gradient-boosted random forest classifier. [Figure 6] Figure 6A shows two images of an object classified as "alive" by an alternative viability prediction model but "dead" by the deep learning model. Figure 6B shows two images of an object classified as a "cluster" by an alternative viability prediction model but that appears to be an individual, viable T cell with appendages. [Figure 7] Figure 7 shows a bar graph depicting the relative importance of the 14 most commonly used features within the random forest model used to predict CD4 and CD8 T cell types. [Figure 8] Figure 8 shows an image segmentation mask with identified cells. The boxes represent identified cell objects from scikit-image. [Figure 9] Figure 9 shows a 3D plot of the extracted cell image. The relative z-contours are shown on the X and Y axes, and the identified caps are shown on the Z axis. The caps represent peaks on the surface of the cell image. DETAILED DESCRIPTION OF THE INVENTION

[0044] Detailed Description Provided herein are methods for classifying (e.g., predicting) individual cells (e.g., T cells) in a population of cells (e.g., a population of cells comprising T cells) using a classification process involving a machine learning model. In certain embodiments, the method includes using a machine learning model trained to classify individual cells (e.g., T cells) in a population of cells (e.g., a population of cells comprising T cells) as belonging to a group, where the group is defined by one or more cellular attributes. Cellular attributes include, but are not limited to, cell health (e.g., alive, dead), cell cycle status, differentiation status, activation status, cell subtype (CD4+, CD8+ T cell), and / or engineering status (e.g., whether the T cell expresses a recombinant receptor (e.g., chimeric antigen receptor (CAR), T cell receptor (TCR))). In some embodiments, the machine learning model is trained to classify (e.g., predict) cell subtype identity (e.g., CD4+, CD8+). In some embodiments, the machine learning model is trained to classify (e.g., predict) whether a cell is healthy (e.g., alive or non-apoptotic). In some embodiments, the machine learning model is trained to classify (e.g., predict) whether a cell is activated. In some embodiments, the machine learning model is trained to classify (e.g., predict) whether a cell is differentiated. In some embodiments, the machine learning model is trained to classify (e.g., predict) whether a cell has entered the cell cycle or is in the cell cycle. In some embodiments, the machine learning model is trained to classify (e.g., predict) whether a cell expresses a recombinant molecule (e.g., a CAR, a TCR).

[0045] Existing methods for determining or identifying whether cells (e.g., T cells) belong to a certain group can be identified using techniques such as immunohistochemistry, immunocytochemistry, or cell sorting techniques such as FACS, MACS, or chromatography. However, these methods can damage or destroy the cells of interest. This can be problematic when cells are intended for use in, for example, therapeutic cell products. For example, repeated product sampling during production, in which damaging or destructive methods are used to classify cells, can deplete the entire therapeutic cell population, unintentionally increasing the duration of production, and also increasing the possibility of contamination.

[0046] The provided methods relate to methods that can classify cells (e.g., T cells) as belonging to a particular group without the need for processing the cells, e.g., by immunohistochemistry, immunocytochemistry, or cell sorting. The provided methods are based on determining or identifying differences in cells (e.g., T cells) from captured images (e.g., optical images) of cells, such as cells within a population of cells (e.g., T cells within a population of cells containing T cells), without the need for further processing of the cells, e.g., by immunohistochemistry, immunocytochemistry, or cell sorting. Thus, in some embodiments, a population of cells can be imaged (e.g., via microscopy techniques as described below), and the image data output from the imaging process can be used as input to a classification process (e.g., a machine learning model) as described herein.

[0047] The provided methods offer advantages over other methods of classifying cells. The methods for classifying cells provided herein, including their embodiments, can classify cells of a population of cells (e.g., T cells of a population of cells containing T cells) using a non-destructive and non-damaging method. For example, in some embodiments, the imaged cells are returned to the cell population (e.g., a therapeutic cell product) under undamaged and / or sterile conditions. The provided methods can also provide an objective and unbiased approach to classifying cells. This differs from many available methods, including those involving purely manual (e.g., human) evaluation of image data or input features to classify cells, which can be labor-intensive, subjective, and prone to human error. Machine learning models have been shown to be useful for objectively classifying data.

[0048] In aspects of the provided methods, the cell population to be evaluated is a mixed population of cells containing cells with various cellular attributes and characteristics. In certain embodiments, the population of cells is a T cell population. In some embodiments, a cell population, such as a T cell population, is a mixed population containing cells of various subtypes (e.g., CD4+ and CD8+ T cells) in various differentiation states (e.g., naive or memory), various activation stages (resting or non-activated and activated), viable or non-viable, apoptotic or non-apoptotic, expressing or not expressing recombinant molecules (e.g., via transduction or transfection), or any combination of any of the foregoing. In some embodiments, the mixed population contains CD4+ and CD8+ T cells, and some of the cells in the mixed population also express recombinant molecules (e.g., CARs) (e.g., via transduction). In some embodiments, the mixed population contains a single T cell subtype (e.g., CD4+ or CD8+), and some of the cells express recombinant molecules (e.g., CARs) (e.g., via transduction). In such an embodiment, the method can be used when it is desired to know or determine such particular attributes present in a cell composition containing a population of cells.

[0049] In some embodiments, the methods provided herein are used in conjunction with manufacturing processes, such as those described herein, for producing therapeutic cell populations (e.g., therapeutic cell products) useful in cell therapy. The classification processes provided herein can be used, for example, to optimize manufacturing processes and / or evaluate product quality. For example, the classification methods provided herein can be used to determine the total number and / or identity of viable cells in a population, including the subtype and / or the ratio of successfully engineered cells (e.g., CAR-expressing cells), and can inform the duration and / or conditions of the manufacturing process to ensure a useful therapeutic cell product. Using a classification process incorporating machine learning models to predict such values ​​provides an objective means of optimizing the manufacturing of therapeutic cell products and improving product consistency.

[0050] In some embodiments, the groups into which cells (e.g., T cells) can be classified are defined by one cellular attribute (e.g., survival). In some embodiments, the groups are defined by at least two cellular attributes (e.g., survival and CD4+; survival, CD4+, and CAR+). In some embodiments, the group is a first group defined by one or more cellular attributes. In some embodiments, the group is a second group defined by one or more cellular attributes, where at least one attribute is different from the one or more attributes defining the first group. In some embodiments, three or more or multiple groups are defined by one or more cellular attributes, where at least one attribute in each of the three or more or multiple groups is different from each of the other three or more or multiple groups.

[0051] In some embodiments, the classification results (e.g., predictions) can be used to determine the characteristics of the cell population. For example, the ratio of the first subtype (e.g., CD4+) to the second subtype (CD8+) in the cell population, the proportion of cells engineered with a recombinant molecule such as a recombinant receptor (e.g., CAR) (e.g., the ratio of CAR+ cells to CAR- cells, including cell identities (e.g., CD4+CAR+, CD4+CAR-, CD8+CAR+, CD8+CAR-)), and / or the total number of viable cells in the sampled population can be determined. For example, a first group can be defined by the cell attribute of being CD4+, a second group can be defined by the cell attribute of being CD8+, and cells with unknown attributes (e.g., T cells) can be evaluated by a trained machine learning model to classify the cells with unknown attributes as belonging to the first group or the second group.

[0052] In the provided methods, the machine learning model is trained using image data from cells, such as T cells. In some embodiments, the image data includes imaging of the cells, for example, by using a microscopy technique. In some embodiments, the microscopy technique is differential digital holographic microscopy (DDHM). In some embodiments, the microscopy technique generates image data associated with the cells (e.g., T cells) that includes phase data, intensity data, and / or overlay data. In some embodiments, the machine learning model is trained using phase image data, intensity image data, and / or overlay image data associated with the cells. In some embodiments, the machine learning model is trained using phase image data and intensity image data associated with the cells. In certain embodiments of the provided methods, the cells are T cells.

[0053] In some embodiments, the machine learning model is trained using one or more input features determined from image data associated with cells (e.g., T cells). In some embodiments, the one or more input features include morphological features (e.g., related to the size or shape of the cell), optical features (e.g., optical properties of the cell image, such as related to the refraction of the cell), intensity (e.g., intensity properties, e.g., RGB or grayscale values ​​of the image), phase (e.g., properties of the phase image, such as the mean or variance or optical height), and / or system features (e.g., properties of the display or readout of the image, such as the position of the cell along the depth axis). In some embodiments, the input features correlate with cellular attributes. In some embodiments, the machine learning model is trained on one or more input features. In some embodiments, the machine learning model is trained using one or more morphological, optical, intensity, phase, and / or system features as described herein.

[0054] In some embodiments, the machine learning model is trained using a supervised, semi-supervised, or unsupervised method. In some embodiments, the training method is supervised. For example, the image data on which the machine learning model is trained is labeled, e.g., the image data is associated with cells (e.g., T cells) that are known to belong to a group based on the knowledge that the cells have one or more cellular attributes that define the group.

[0055] Training data, such as image data (e.g., phase, intensity, overlay) and / or input features determined from image data associated with cells in a cellular dataset can be used as inputs for training a machine learning model. In some embodiments, the cellular dataset is a pure population of cells in which substantially all of the cells in the population, e.g., 85%, 86%, 87%, 88%, 89%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99%, or more, exhibit the same one or more cellular attributes (e.g., subtypes such as CD4+ or CD8+). In some cases, a pure population of cells with desired cellular attributes can be isolated from a sample (e.g., blood purification therapy, leukapheresis therapy, blood, PBMC sample). In this way, cells in a cellular dataset containing a pure population are known to belong to a specific group, and the group is defined by one or more cellular attributes. In some embodiments, training data can be obtained directly from a dataset of cells containing a pure population, and the training data can be used to train a machine learning model to classify cells having one or more cellular attributes (e.g., belonging to a particular group). Thus, when a machine learning model receives input, e.g., image data, input features associated with cells of unknown cellular attributes, the machine learning model can classify the cells. In some embodiments, the pure population can undergo additional processing, such as the manufacturing procedures described herein. In some embodiments, a dataset of cells containing cells of a pure population of cells is processed in the same or substantially the same manner as the cells to be classified (e.g., cells with unknown cellular attributes).

[0056] Provided herein is a method for generating a cellular dataset that can be used to train a machine learning model. The provided method for generating a cellular dataset is based on the observation herein that, in some cases, using a specific pure population of cells as a cellular dataset for training to classify cells in a mixed population of cells may not accurately predict the attributes of the mixed population of cells (e.g., Example 6). Unlike a cellular dataset generated entirely from a pure population of cells, a mixed cell population is considered to be subject to cell signaling and / or other environmental influences that may occur under conditions where multiple cell types exist (e.g., due to the presence of cytokines or growth factors produced by one cell type that may affect the attributes of another cell type in the mixed population of cells). However, unlike a cellular dataset from a pure population, a mixed population contains cells belonging to multiple groups (e.g., cells in a mixed population do not all share one or more identical cellular attributes), so the mixed population cannot be used directly to obtain training data. In an aspect of the provided method for generating a cellular dataset for training, the cellular dataset is generated by separating cells from the mixed population of cells based on cellular attributes to provide a substantially pure population of cells for training. For example, the mixed population of cells may contain CD4+ and CD8+ cells, and / or CD4+ and CD8+ cells transduced or transfected to express a recombinant molecule (e.g., CAR, TCR), and the method includes selecting one of the CD4+ or CD8+ cells (or one of the CD4+ or CD8+ cells transduced or transfected to express a recombinant molecule). Thus, in some embodiments, to generate a dataset of cells from the mixed population, cells with desired attributes are separated from the mixed population. By separating the cells, a dataset of cells containing cells with known cell attributes (e.g., belonging to a particular group) can be generated.

[0057] In embodiments for generating a cellular dataset, a method is performed to select cell types from a mixed population of cells for one or more desired attributes, and the selection method is performed under conditions such that the resulting cell population is substantially free of the reagents used in the selection process, without excessively manipulating or treating the cells to remove the reagents. Because certain manipulations of cells and / or the presence of reagents can risk losing cells in the training cellular dataset or can alter the phenotype or function of the cells, thereby changing the cells' underlying input characteristics, such methods are provided to ensure that the cellular data used for training is as close as possible to representing the cells present in the mixed population. As an example, the use of magnetic beads for cell selection may require the use of a magnet to remove the beads from the cells, which may result in extra incubation or processing that can affect the cellular phenotype or function, may result in cell loss due to incomplete release of the magnetic field, and / or may result in residual beads in the final selected cell population, thereby interfering with machine learning methods. Another advantage of the provided methods is that the method for selecting or separating cells from a mixed population of cells does not require the cellular dataset to be subjected to any additional processing to match the cellular processing of the mixed population of cells. Rather, the cells of the cellular dataset are cells that have been processed, processed, or handled in the same or substantially the same manner as the cells to be classified (e.g., cells with unknown cellular attributes).

[0058] In certain embodiments, the cellular dataset is generated by separating or selecting cells from a mixed population using reversible affinity-based selection, such as by immunoaffinity chromatography, which allows the affinity reagent used for selection to efficiently dissociate from the cells. Reversible immunoaffinity chromatography methods, in some aspects, include those described in U.S. Patent Application Publication Nos. US2015 / 0024411 and US2017 / 0037369, both of which are incorporated by reference herein in their entireties. In some embodiments, the method uses a separation matrix to which streptavidin muteins are immobilized that are reversibly bound to monovalent antibodies (e.g., Fabs) specific for markers (e.g., surface molecules) of a desired cell attribute, such as a cell subtype (e.g., CD4+ or CD8+).

[0059] In some embodiments, the cellular dataset is generated using positive selection. In some embodiments, the cellular dataset is generated using negative selection. Considering that more than one cell type exists in a mixed population, in some embodiments, two or more or multiple cellular datasets are generated by separating cells of the mixed population. Thus, in some aspects, multiple separation steps can be performed to isolate cells with specific cellular attributes, thereby generating a cellular dataset. In some embodiments, a further separation step is performed on the negative fraction (e.g., cells not selected by affinity-based methods). In some embodiments, a further separation step is performed on the positive fraction (e.g., cells selected by affinity-based methods). It is envisioned that any number of separation steps can be used to generate a cellular dataset for training a machine learning model.

[0060] In some embodiments, a cellular dataset containing a subset of cells from a mixed population (e.g., belonging to a specific group) can be used to obtain training data for training a machine learning model to classify cells with one or more cellular attributes (e.g., belonging to a specific group). Thus, when a machine learning model receives input, e.g., image data, input features associated with cells with unknown cellular attributes, the machine learning model can classify the cells. In some embodiments, the mixed cell population can undergo additional processing, such as the manufacturing procedures described herein. In some embodiments, the mixed population of cells from which the cellular dataset is generated is processed in the same or substantially the same manner as the cells to be classified (e.g., cells with unknown cellular attributes).

[0061] In some embodiments, the method of classifying cells comprises testing the trained machine learning model for accuracy. In some embodiments, the method comprises validating the trained machine learning model. In some embodiments, cross-validation is used to validate the model.

[0062] In some embodiments, a method for classifying cells of a population of cells includes receiving image data from cells of the population, e.g., by using DDHM, and applying one or more of the phase image data, intensity image data, and / or overlay image data of the image data to a machine learning model trained on one or more of the phase image data, intensity image data, and / or overlay image data of the image data from cells known to belong to the first group and cells known to belong to the second group, to classify the cells of the population as belonging to a first or second group. In some embodiments, a method for classifying cells of a population of cells includes receiving image data from cells of the population, e.g., by using DDHM, and applying the phase image data and intensity image data of the image data to a machine learning model trained on the phase image data and intensity image data of the image data from cells known to belong to the first group and cells known to belong to the second group, to classify the cells of the population as belonging to a first or second group. In some embodiments, there may be more than two groups into which T cells can be classified. For example, the machine learning model may be trained to distinguish between more than three or multiple groups. In some embodiments, the machine learning model is a deep learning model. In some embodiments, the machine learning model is an artificial neural network. In some embodiments, the machine learning model is a convolutional neural network. Convolutional neural networks are particularly well suited to handling image data. The advantage of convolutional neural networks is that they reduce the number of parameters of deep neural networks with many units without compromising the quality of the model. Therefore, convolutional neural networks are computationally efficient. Further advantages of convolutional neural networks include their ability to automatically detect important features for classification without human supervision, for example, to detect features that may not be detectable by humans.

[0063] In some embodiments, a method of classifying cells of a population of cells includes receiving image data from cells of the population, e.g., by using a DDHM; determining one or more input features from the image data, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, and / or system features of the image data; and applying the one or more input features as inputs to a machine learning model trained on the one or more input features determined from image data associated with cells known to belong to the first group and cells known to belong to the second group, to classify the cells of the population as belonging to a first group or a second group, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, and system features of the image data. In some embodiments, a method for classifying cells of a population of cells including T cells includes receiving image data from cells of the population, e.g., by using DDHM; determining one or more input features from the image data, wherein the one or more input features include morphological features of the image data, optical features of the image data, intensity features of the image data, and / or phase features of the image data; and applying the one or more input features as inputs to a machine learning model trained on the one or more input features determined from image data associated with cells known to belong to the first group and cells known to belong to the second group, to classify the cells of the population as belonging to a first group or a second group, wherein the one or more input features include morphological features of the image data, optical features of the image data, intensity features of the image data, and / or phase features of the image data. In some embodiments, there may be more than two groups into which T cells can be classified. In some embodiments, the machine learning model is an artificial neural network. In some embodiments, the machine learning model is a random forest. In some embodiments, the machine learning model is a support vector machine.In some embodiments, the machine learning model is a deep learning model.

[0064] In some embodiments, the methods for classifying cells provided herein are used throughout one or more steps or periods of a process for producing or manipulating cells ex vivo for use as adoptive cell therapy, e.g., as described herein. In some embodiments, the methods for classifying cells provided herein are used, e.g., in a process described herein, during a specific step of the manufacturing process after the cells have been transduced or transfected with a recombinant molecule such as a CAR, e.g., during the step of incubating the engineered T cell population (e.g., Section II-C-3). In some embodiments, the methods for classifying cells described herein, when used in connection with a manufacturing process, e.g., as described herein, can be used to guide incubation conditions, including one or more of the stimulation conditions, manipulation conditions, culture conditions, and harvest conditions, of the manufacturing process. In some embodiments, the methods for classifying cells described herein, when used in connection with a manufacturing process, e.g., as described herein, can be useful for determining the need to modify incubation conditions, including stimulation conditions, or can be used to identify when cells are ready to be harvested. For example, identifying cell type, cell health, and viable concentration may be useful to determine perfusion and / or feeding schedules, the need to reduce or add cytokines during stimulation conditions, to detect the concentration of activated or differentiated T cells, to detect whether a mixed population contains an appropriate concentration of viable cells and / or viable cell subtypes to be transduced or transfected, to detect the concentration of successfully transduced cells, and / or to detect the concentration of viable cells, viable transduced / transfected cells, viable transduced / transfected cell subtypes to determine harvest.In some embodiments, the cell sorting methods described herein, when used in conjunction with a manufacturing process, e.g., as described herein, can support a fully automated cell culture system, e.g., without the need for manual (e.g., human operator) input. In some embodiments, the fully automated system is a closed-loop system. In some embodiments, the fully automated closed-loop system efficiently and reliably produces sterile, viable populations of engineered cells, including specific cell types (e.g., CD4+, CD8+, CAR+, TCR+), in specific ratios or total cell numbers for cell therapy (e.g., adoptive cell therapy, autologous cell therapy).

[0065] In some embodiments, the cell classification methods described herein, when used in conjunction with a manufacturing process, for example, as described herein, result in a manufacturing process that generates or produces genetically engineered cells suitable for cell therapy in a manner that is faster, more efficient, and results in a more consistent product than alternative manufacturing processes. In certain embodiments, the cell classification methods provided herein, when used in conjunction with a manufacturing process, for example, as described herein, result in a higher success rate for generating or producing compositions of engineered cells from a broader population of subjects than may be possible using alternative processes. In certain embodiments, the engineered cells produced or generated by the provided methods may have better health, viability, activation, and greater expression of recombinant receptors than cells produced by alternative methods. Thus, in some aspects, the speed and efficiency of the provided methods for generating engineered cells for cell therapy allows for easier planning and coordination of cell therapy treatments, such as autologous therapy, for a broader population of subjects than may be possible with some alternative methods.

[0066] All publications referenced in this application, including patent documents, scientific papers, and databases, are incorporated by reference in their entirety for all purposes, as if each individual publication were individually incorporated by reference. To the extent that a definition set forth herein conflicts or is otherwise inconsistent with a definition set forth in a patent, patent application, published patent application, or other publication incorporated herein by reference, the definition set forth herein shall take precedence over the definition incorporated herein by reference.

[0067] The section headings used herein are for organizational purposes only and are not to be construed as limiting the subject matter described.

[0068] I. How to classify T cells The methods provided herein allow for the classification of cells as belonging to a group defined by one or more cellular attributes through the use of a classification process incorporating machine learning models. In some embodiments, the classification process can incorporate one or more types of machine learning models, such as those described below. For example, cells, including cells that have undergone a manufacturing process to genetically engineer cells, can exhibit differences that can be used to classify cells as belonging to a first group and exclude them from belonging to a second group. For example, morphological characteristics of T cells can be used to classify T cells as a CD4+ subtype rather than a CD8+ subtype. In some embodiments, the cells are T cells. In some embodiments, the T cells are or have been used in a manufacturing process, such as a process described herein. Group, class, and category can be used interchangeably herein.

[0069] Provided herein is a method for classifying cells (e.g., T cells), the method comprising: receiving image data associated with one or more such cells (e.g., a first, second, or third cell) of a population of cells, such as by using an imaging technique, e.g., digital holographic imaging; determining from the image data one or more features (also referred to as "input features") that are depicted, described, represented by, or deducible from the image data (e.g., by mathematical means); and applying the one or more input features as input to a machine learning process configured to classify the cells as belonging to a first group or a second group based on the one or more input features. In some embodiments, the one or more features of the image data can be one or more of a morphological feature of the image data, an optical feature of the image data, an intensity feature of the image data, a phase feature of the image data, a system feature of the image data, or any combination thereof.

[0070] Provided herein is a method for classifying cells (e.g., T cells), the method comprising: receiving image data associated with one or more such cells (e.g., a first, second, or third cell) of a population of cells, such as by using an imaging technique, e.g., digital holographic imaging; and applying the image data as input to a machine learning process configured to classify the cells as belonging to a first group or a second group based on one or more input features. In some embodiments, the image data comprises one or more of phase data, intensity data, overlay data, or any combination thereof. In some embodiments, the image data comprises phase data and / or intensity data.

[0071] In certain embodiments, the provided methods can be used to classify different groups of T cells, such as T cells of different subsets (e.g., CD4+ or CD8+), of different viability or health status (e.g., viable or non-viable), or of different engineered status (e.g., transduced or not transduced with a recombinant molecule, e.g., a CAR). Features of the provided methods are described in the following subsections.

[0072] A. Cell Imaging Technology The provided methods include imaging cells of a population of cells. Imaging techniques contemplated for use in connection with the methods of classifying cells provided herein include any and all microscopy techniques capable of imaging cells, e.g., T cells, from a population of cells including T cells, to generate image data and / or information determined therefrom (e.g., input features; see, e.g., Section IB) that can be used as input to a classification process (e.g., a machine learning model), e.g., for classification or training purposes. In some embodiments, the imaging technique includes steps such as image acquisition, image processing (e.g., preprocessing), and / or image segmentation to generate image data, e.g., image data associated with cells. In some embodiments, the image processing includes centroid detection. In some embodiments, the image processing includes cell detection based on peak intensity when out of focus. In some embodiments, the image segmentation includes edge detection. In some embodiments, the image segmentation is performed on phase image data. In some embodiments, a computer system is used to interpret the output (e.g., results) of the imaging technique to generate image data.

[0073] Imaging techniques may use a digital device to capture, for example, record, the output (e.g., result) of the imaging process. In some embodiments, the digital device is a charge-coupled device (e.g., a CCD camera). In some embodiments, the digital device is a complementary metal oxide semiconductor device (e.g., a CMOS camera). Thus, in some embodiments, image data is captured using a digital device. In some embodiments, the digital device interfaces with a computer to store and / or analyze the results (e.g., output) of the imaging process.

[0074] In some embodiments, the imaging technique allows for obtaining image data from a single cell (e.g., a T cell). In some embodiments, the imaging technique allows for obtaining image data from two or more or multiple cells (e.g., T cells) simultaneously. For example, the imaging technique may include a field of view that allows for multiple cells to be imaged at once, and image data associated with each cell (e.g., a T cell) can be stored, analyzed, and / or used independently (e.g., on a computer system) from image data associated with other cells (e.g., T cells) within the field of view. In some embodiments, image segmentation is used to determine image data associated with each cell of the multiple cells. In some embodiments, the imaging technique includes a field of view of approximately 320 μm x 320 μm. In some embodiments, the imaging technique includes a horizontal resolution of approximately 1 μm. In some embodiments, the imaging technique includes a depth of 120 μm.

[0075] In some embodiments, the imaging technique generates image data containing phase data. In some embodiments, the imaging technique generates image data containing intensity data. In some embodiments, the imaging technique generates image data containing phase and intensity data. In some embodiments, the imaging technique generates image data containing superposition data. In some embodiments, the imaging technique generates image data containing phase, intensity, and superposition data. In some embodiments, the imaging technique generates a hologram. In some embodiments, the imaging technique generates three-dimensional information. For example, the imaging technique may generate an image including x, y, and z coordinate values. In some embodiments, the image data is stored or processed on a computer.

[0076] In order to obtain image data associated with cells (e.g., T cells) in a non-destructive manner without causing damage, the cells can be contained in a liquid such as a culture medium. In some embodiments, the cells (e.g., T cells) can be suspended in a liquid, such as a culture medium, for imaging. Thus, in some embodiments, the imaging technique can image cells, such as T cells, contained and / or suspended in a liquid.

[0077] In some embodiments, the imaging technique is bright-field microscopy, fluorescence microscopy, differential interference contrast (DIC) microscopy, phase contrast microscopy, digital holographic microscopy (DHM), in-line differential digital holographic microscopy, differential digital holographic microscopy (differential DHM or DDHM), or combinations thereof.In some embodiments, the imaging technique is DDHM.

[0078] 1. Digital holographic microscopy An exemplary microscopy technique for use in connection with the classification methods described herein is digital holographic microscopy (DHM), more specifically, differential DHM (DDHM). Digital holographic microscopy allows for the study of living cells without the need for markers or dyes and allows for quantification of the objects being studied. Digital holographic microscopy is a technique that allows for the recording of 3D samples or objects without the need to scan the sample layer by layer. In this respect, DHM is a superior technique to confocal microscopy. In DHM, holographic representations are recorded by a digital camera, such as a CCD or CMOS camera, and can then be stored or processed on a computer. Different DHM techniques, including DDHM, as well as microscope configurations and elements, are described in US 2014 / 0193850, US 7,362,449, and EP 1631788, which are incorporated herein by reference in their entireties.

[0079] To create a holographic representation, or hologram, traditionally, a highly coherent light source, such as laser light, is used to illuminate a sample. In the most basic setup, the light from the light source is then split into two beams: an object beam and a reference beam. The object beam is sent to the sample through an optical system and interacts with the sample, thereby changing the phase and amplitude of the light depending on the optical properties and 3D shape of the object. The object beam, reflected from or transmitted through the sample, is then made to interfere with the reference beam (e.g., by a set of mirrors and / or beam splitters), resulting in an interference pattern that is digitally recorded. Since holograms are more accurate when the object and reference beams have comparable amplitudes, an absorbing element can be introduced into the reference beam, reducing the amplitude of the reference beam to the level of the object beam but not changing the phase of the reference beam, or at most changing its phase entirely, i.e., independent of where and how the reference beam passes through the absorbing element. The recorded interference pattern contains information about the phase and amplitude variations that depend on the optical properties and 3D shape of the object.

[0080] Another method for creating holograms is to use in-line holographic techniques. In-line DHM is similar to more traditional DHM, but at least the light beam is not split by a beam splitter or other external optical elements. In-line DHM is most preferably used to study particles in fluids, such as less dense solutions of cells. This allows some of the at least partially coherent light to pass through the sample without interacting with the particles (reference beam) and interfere with the light that has interacted with the particles (object beam), resulting in an interference pattern that is digitally recorded and processed. In-line DHM is used in transmission mode, requires light with a relatively large coherence length, and cannot be used when the sample is too viscous or dense.

[0081] Another DHM technique, called differential DHM (DDHM), is disclosed in European Patent EP 1 631 788, which is incorporated herein by reference in its entirety. DDHM differs from other techniques in that it does not utilize a reference beam or an object beam at all. In a preferred configuration of DDHM, the sample is illuminated by an illumination means consisting of at least partially coherent light in reflection or transmission mode. The reflected or transmitted sample beam is sent through an objective lens and subsequently split into two by a beam splitter and sent along different paths in, for example, a Michelson or Mach-Zehnder differential interferometer. In one of the paths, a beam-bending element or angle-changing means, such as a transparent wedge, is inserted. The two beams are then made to interfere with each other in the focal plane of a focusing lens, and the interference pattern in this focal plane is digitally recorded and stored, for example, by a CCD camera connected to a computer. The beam-bending element slightly shifts the two beams in a controlled manner, and the interference pattern depends on the amount of shift. The beam bending element is then rotated, thereby changing the amount of shift. A new interference pattern is also recorded. This can be done N times, and from these N interference patterns, the gradient of the phase (or spatial derivative) at the focal plane of the focusing lens can be approximately calculated. This is called the phase stepping method, although other methods of obtaining the phase gradient are also known, such as Fourier transform data processing techniques. The gradient of the phase can be integrated to give the phase as a function of position. The amplitude of the light as a function of position can possibly, but not necessarily, be calculated from a weighted average of the amplitudes of the N recorded interference patterns. Since the phase and amplitude are thus known, the same information can be obtained as in direct holographic methods (using reference and object beams), and subsequent 3D reconstruction of the object can be performed.

[0082] In DDHM, the illumination means can include spatially and temporally partially coherent light. This contrasts with other DHM methods, which use only highly correlated laser light. Spatially and temporally partially coherent light can be generated, for example, by an LED. LEDs are less expensive than lasers and generate light with a spectrum centered around a known wavelength that is spatially and temporally partially coherent, i.e., not as coherent as laser light but still sufficiently coherent to generate a holographic image of the quality required for the application at hand. LEDs are available for many different wavelengths and have the advantage of being very small and easy to use or, if necessary, replace. Therefore, providing a method that can use spatially and temporally partially coherent light to generate a holographic image would result in a more cost-effective device for implementing such a method. In some embodiments, the illumination means is a red LED.

[0083] The images may undergo object segmentation and further analysis to obtain multiple morphological features that quantitatively describe the imaged object (e.g., cultured cells, cell debris). Thus, for example, various features (e.g., cell morphology, cell viability, cell concentration) may be evaluated or calculated directly from DDHM using the steps of image acquisition, image processing, image segmentation, and feature extraction. In some embodiments, a digital recording device is used to record the holographic image. In some embodiments, a computer containing an algorithm for analyzing the holographic image may be used. In some embodiments, a monitor and / or computer may be used to display the results of the holographic image analysis. In some embodiments, the analysis is automated (i.e., can be performed without user input). Examples of DDHM systems include, but are not limited to, the Ovizio iLine F (Ovizio Imaging Systems NV / SA, Brussels, Belgium).

[0084] Any type of DHM can be used according to the methods provided herein (including their embodiments). In some embodiments, DHM, in-line DHM or differential DHM is used in combination with other microscopy (e.g., fluorescence microscopy). In some embodiments, DDHM is used in connection with the methods provided herein. In some embodiments, DDHM is performed as described above. In some embodiments, image data is collected from imaging techniques including DDHM.

[0085] B. Image Data and Input Features Image data, such as those collected using the imaging techniques described in Sections IA and IA-1 above, can be used as input to a classification process (e.g., a machine learning model) for classifying T cells. In some embodiments, one or more features can be determined from the image data. In some embodiments, one or more features can be used as input to a classification process (e.g., a machine learning model) for classifying T cells. Such features may be referred to herein as "input features." The image data can include intensity data (e.g., RGB, RGBA, grayscale values ​​corresponding to pixels), phase data, overlay data, spatial data, or any other suitable data corresponding to the output of an imaging process (e.g., DDHM). The image data can be represented as an array of values ​​along any suitable number and type of dimensions. For example, in some examples, the image data can be represented as a two-dimensional array, with each element holding three 8-bit values ​​and each value representing the RGB intensity of the corresponding pixel (or other image element). The image data can represent a 2D image, a 3D stereoscopic image, a hologram, or any other suitable image. The image data can correspond to a visual image, but need not correspond to one. The image data may correspond to visible light, although in some examples the image data may correspond to any suitable light source (e.g., infrared or ultraviolet light). Additionally, any suitable imaging method may be used.

[0086] In some embodiments, the image data is collected using a DDHM. In some embodiments, the image data is received from a DDHM. In some embodiments, the image data collected using or received from a DDHM is processed to extract features, such as those described below.

[0087] In the provided methods, the image data itself can be used to train a machine learning model, such as a convolutional neural network, and the image data from the cells to be classified can be used as input to the model (e.g., a convolutional neural network model). In other provided embodiments, a plurality of input features derived from the image data can be used to train a machine learning model using a support vector machine (SVM), a neural network, a random forest, or the like, and the input features from the cells to be classified can be used as input to the model. In some embodiments, the input features determined from the image data can include any one or more of morphological features, optical features, intensity features, phase features, or system features.

[0088] In some embodiments, one or more morphological features can be determined from the image data corresponding to the cells, e.g., by applying appropriate filters, statistical methods, and / or signal processing methods to the image data. The morphological features describe one or more characteristics of the physical shape of the cells. Morphological features can include, for example, cell aspect ratio (e.g., the ratio of the length of a first axis of a cell to the length of a second axis of the cell), which in some embodiments can be normalized; cell area (e.g., surface area or projected surface area); cell circularity (e.g., the ratio of the area of ​​a cell to the perimeter of the cell); cell compactness (e.g., the ratio of the area of ​​a cell to the variance of an object in the axis of the cell); cell elongation (e.g., the ratio of the width to the height of a rectangle enclosing the cell); cell diameter (e.g., the diameter of a circle having the same area as the cell); Hu moment (e.g., a weighted average of pixel intensities associated with a cell (e.g., one of Hu moment invariants 1-7)); perimeter (e.g., the perimeter of a cell); or radius (e.g., the distance from the center of gravity of the cell to the perimeter of the cell), which in some embodiments can determine normalized radius variance by dividing the radius variance by the diameter. In some examples, such morphological features can be associated with two-dimensional images of cells or three-dimensional images of cells. The above-described morphological features can be applied as inputs to a classification process for classifying cells, as described below.

[0089] In some embodiments, one or more optical features can be determined from image data corresponding to a cell, e.g., by applying appropriate filters, statistical methods, and / or signal processing methods to the image data. The optical features describe one or more optical properties of the image of the cell. The optical features can include, for example, peak diameter (e.g., diameter of a circle having the same area as the refractive peak of the cell); mass eccentricity of the cell (e.g., distance between the geometric center of the cell and the center of gravity of the cell, weighted by the optical height (e.g., in pixels)); minimum, maximum, or average intensity of the cell (e.g., RGB intensity value); minimum, maximum, or average optical height of the cell (e.g., height in microns or radians across the cell surface); normalized optical height of the cell (e.g., optical height of the cell (e.g., in microns) divided by the cell diameter); optical volume of the cell (e.g., a value proportional to the cell volume and refractive index of the cell); area of ​​the refractive peak of the cell (e.g., in microns); number of refractive peaks of the cell; refractive peak area normalized by the area of ​​the cell (e.g., peak area divided by the product of the cell area and the average intensity); normalized refractive peak height of the cell (e.g., intensity of the refractive peak of the cell divided by the product of the cell surface and the average intensity of the cell); and / or intensity of the refractive peak of the cell. In some examples, the optical features described above can be associated with a two-dimensional image of the cell or a three-dimensional image of the cell. The optical features described above can be applied as input to a classification process to classify cells, as described below.

[0090] In some embodiments, one or more intensity features can be determined from image data corresponding to a cell, for example, by applying appropriate filters, statistical processing methods, and / or signal processing methods to the image data. The intensity feature describes one or more characteristics of the intensity of the cell. In some examples, the intensity of a cell can include an intensity value of an image of the cell (e.g., an RGB value of the image). In some examples, the intensity of a cell can include an indication of presence at a location (e.g., a value of 1 or 0 indicating whether a cell is present at that location). Intensity features can include, for example, the intensity value (e.g., average intensity) of the image of the cell; the intensity contrast (e.g., the difference in intensity between a pixel in an image and a neighboring pixel) of the image of the cell, where in some embodiments the intensity contrast of the image of the cell is the average intensity contrast; the intensity entropy (e.g., average entropy) of the image of the cell; the intensity uniformity (e.g., a measure of the uniformity of intensity across the cell surface), where in some embodiments the intensity uniformity of the image of the cell is expressed as an average measure of the uniformity of intensity across the cell surface (e.g., average intensity uniformity); the intensity correlation (e.g., the correlation between the intensity of a pixel in an image and a neighboring pixel); the intensity homogeneity (e.g., the spatial proximity of the distribution to the diagonal) of the image of the cell; the intensity skewness (e.g., intensity symmetry) of the image of the cell; the intensity variance (e.g., the variance of the intensity across the cell surface) of the image of the cell; the plane (e.g., in microns) at which the intensity of the cell is maximum; and / or the intensity smoothness (e.g., average smoothness) of the image of the cell. In some examples, such intensity features can be associated with a two-dimensional image of the cell or a three-dimensional image of the cell. The intensity features described above can be applied as input to a classification process to classify cells, as described below.

[0091] In some embodiments, one or more phase features can be determined from the image data corresponding to the cells, e.g., by applying appropriate filters, statistical methods, and / or signal processing methods to the image data, where the phase features describe one or more properties of the phase image of the cells. Phase features can include, for example, the optical height of a phase image of a cell (e.g., the mean or variance of the optical height in microns or radians); the phase intensity contrast of a cell (e.g., the average intensity contrast of a phase image of a cell or the intensity contrast of a phase image of a cell between pixels across the surface of a cell); the phase entropy of a cell (e.g., the average entropy of a phase image of a cell); the average phase of a cell (e.g., the phase average taken across the surface of a cell); the phase uniformity of a cell (e.g., a measure of phase uniformity across the surface of a cell), in some embodiments, the phase uniformity of a cell is an average measure of phase uniformity across the surface of a cell (e.g., the average phase uniformity of a cell); the phase correlation of a cell (e.g., a measure of how correlated a pixel is with its neighboring pixels in a phase image of a cell); the phase homogeneity of a cell (e.g., a measure of spatial proximity of the distribution across the diagonal in a phase image of a cell); the phase skewness of a cell (e.g., a measure of asymmetry in a phase image of a cell); and / or the phase smoothness of a cell (e.g., a measure of smoothness in a phase image of a cell). In some examples, such phase features can be associated with a two-dimensional image of a cell or a three-dimensional image of a cell. The topological features described above can be applied as input to a classification process to classify cells, as described below.

[0092] In some embodiments, one or more system features can be determined from image data corresponding to a cell, for example, by applying appropriate filters, statistical processing methods, and / or signal processing methods to the image data. The system features describe characteristics of the display or readout of an image of a cell, such as one or more positional or statistical characteristics of the image of the cell. System features can include, for example, cell depth (e.g., the position of the cell along the depth axis); cell identifier (e.g., a unique ID number assigned to the cell in the image); image identifier (e.g., a unique identifier for an image containing an object); object identifier (e.g., a unique identifier for an object in the image); cell descriptor (e.g., a categorical description such as live, dead, debris, cluster, generic, debris with peaks, etc.); centroid (e.g., a centroid coordinate along the X- or Y-axis in the image); and / or cell boundary indication (e.g., whether a cell or a cell's refractive peak is present along a boundary in the image). In some examples, such system features can be associated with a two-dimensional image of a cell or a three-dimensional image of a cell. The system features described above can be applied as inputs to a classification process for classifying cells, as described below.

[0093] In some embodiments, the input features include one or more or all of 70 input features, wherein the input features include a cell aspect ratio, a cell depth, a cell area, a cell descriptor, a cell identifier, an image identifier, an object identifier, a centroid along the X axis, a centroid along the Y axis, a cell circularity, a cell compactness, a normalized aspect ratio, a cell elongation, a cell diameter, a peak diameter, a hu moment invariant 1, a hu moment invariant 2, a hu moment invariant 3, a hu moment invariant 4, a hu moment invariant 5, a hu moment invariant 6, a hu moment invariant 7, a mean intensity contrast, a mean entropy, a mean intensity, a mean intensity uniformity, an intensity contrast of an image of a cell, an intensity correlation of an image of a cell, an intensity entropy of an image of a cell, an intensity homogeneity of an image of a cell, a maximum cell intensity, a mean cell intensity, a minimum cell intensity Intensity skewness of the image of the cell, Intensity smoothness of the image of the cell, Intensity variance of the image of the cell, Intensity uniformity of the image of the cell, Plane at which the intensity of the cell is maximum, Indication that the cell is located along the boundary of the field of view, Indication that the refractive peak is located along the boundary of the field of view, Mass eccentricity of the cell, Maximum optical height of the cell in radians, Maximum optical height of the cell in microns, Mean optical height of the cell in radians, Mean optical height of the cell in microns, Normalized optical height of the cell, Minimum optical height of the cell in radians, Minimum optical height of the cell in microns, Phase optical height variance of the image of the cell in radians, Cell in microns These include the variance of the optical height of the image phase, the optical volume of the cell, the area of ​​the refraction peak of the cell, the refraction peak area normalized by the area of ​​the cell, the number of refraction peaks of the cell, the intensity of the refraction peak of the cell, the normalized refraction peak height of the cell, the perimeter, the average intensity contrast of the phase image of the cell, the average entropy of the phase image of the cell, the average phase of the cell, the average phase uniformity of the cell, the phase intensity contrast of the cell, the phase correlation of the cell, the phase entropy feature of the cell, the phase homogeneity of the cell, the phase skewness of the cell, the phase smoothness of the cell, the phase uniformity of the cell, the average radius of the cell, the variance of the radius of the cell, and the normalized radius variance of the cell.

[0094] In some embodiments, one or more of the input features can be applied as input to a classification process for classifying cells, as described below. In some embodiments, one or more of the input features can be applied as input to a machine learning model trained to classify cells. In some embodiments, the one or more input features applied as input to the machine learning model trained to classify cells correspond to the same one or more input features used to train the machine learning model. For example, if a machine learning model is trained using a subset of input features from cells known to belong to a certain group, the input features applied to the machine learning model for the cell to be classified are the same input features that the model was trained with.

[0095] In some examples, it is not necessary to use all of the input features determined from the image data associated with a cell to classify the cell. For example, initial statistical tests can be performed to determine which input features are highly correlated with cell attributes. These input features can then be used as inputs to the machine learning model. Statistical tests can also be performed to identify highly correlated input features. Additionally, statistical tests can be performed to identify input features that exhibit low variance. These input features can be excluded as inputs to the machine learning model. It should be understood that these or similar analyses can be performed on input features used as training data for the machine learning model. In some embodiments, the method described in this paragraph is referred to as data preprocessing. In some cases, the data preprocessing step in model creation avoids the generation of a model that produces misleading or inaccurate results. In some embodiments, preprocessing prevents out-of-range values, missing values, impossible data combinations, highly correlated variables, etc. from being incorporated into the model. In some cases, preprocessing results in the identification of useful features.

[0096] In some embodiments, at least 70, 65, 60, 55, 50, 45, 40, 35, 30, 25, 20, 15, 10, 9, 8, 7, 6, 5, 4, 3, 2, or 1 input features are applied as inputs to the machine learning model, hi some embodiments, about 1 to about 70, about 1 to about 60, about 1 to about 50, about 1 to about 40, about 1 to about 30, about 1 to about 20, about 1 to about 15, about 1 to about 10, or about 1 to about 5 input features are applied as inputs to the machine learning model. In some embodiments, fewer than or less than about 70, 65, 60, 55, 50, 45, 40, 35, 30, 25, 20, 15, 10, or 5 input features are applied to the machine learning model. In some embodiments, fewer than or less than about 20, 15, 10, or 5 input features are applied to the machine learning model.

[0097] C. Machine Learning In some embodiments, image data and / or input features, e.g., as described in Section IB, may be used as input to a machine learning model to classify cells (e.g., T cells) of a population of cells (e.g., a population of cells containing T cells). In some embodiments, each T cell in a population of cells containing T cells can be classified. In some embodiments, a plurality of T cells contained in a population of cells containing T cells can be classified. In some embodiments, about 70, 75, 80, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100% or at least 70, 75, 80, 85, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100% of the T cells in a population of cells containing T cells are classified. Classification can be achieved by a computer performing a classification process, such as a classification process incorporating a machine learning model capable of classifying cell types. The machine learning model may include any suitable machine learning model, such as the exemplary models described below.

[0098] In some embodiments, classification may include identifying a cell as belonging specifically to a first category (e.g., CD4+ subtype) or a second category (e.g., CD8+ subtype). In some cases, classification may include assigning a probability to a category (e.g., a 70% probability that the T cell belongs to the CD4+ subtype); and / or comparing the probability to a threshold to determine whether the cell belongs to the category with sufficient probability.

[0099] In some embodiments, the machine learning model comprises one or more of an artificial neural network (e.g., a convolutional neural network), a regression model, an example-based model, a regularization model, a decision tree, a random forest, a Bayesian model, a clustering model, an associative model, a deep learning model, a dimensionality reduction model, a support vector machine, and / or an ensemble model (e.g., boosting). However, any suitable machine learning model or combination of models can be used to implement various embodiments.

[0100] In some embodiments, an artificial neural network (also referred to herein as a neural network or NN) can be utilized to classify cells as belonging to a first category (e.g., CD4+ subtype) or a second category (e.g., CD8+ subtype). The neural network can utilize a network of one or more computational units (e.g., perceptrons) to connect one or more inputs with one or more outputs. For example, the inputs can include image data and / or input features associated with the candidate cells to be classified, and the output can include an indication of whether the candidate cells belong to the first category (e.g., CD4+) or the second category (e.g., CD8+). In some examples, some or all of the above input features can be presented as inputs to the neural network. The neural network calculates a function of the inputs to arrive at the output, and the function arrives at the output through calculation by one or more computational units.

[0101] The computational unit may incorporate intermediate values ​​(weights). The weights can be adjusted through a training process in which the neural network is tuned to accurately generate a desired output from one or more inputs. For example, an untrained neural network may lack the ability to correctly classify cells as CD4+ or CD8+ cells. However, by continuously applying known inputs (e.g., image data and / or input features belonging to cells known to be CD4+ or CD8+ cells) that correspond to the known output, the weights can be modified to minimize the error between the neural network's output and the desired output. Thus, the neural network improves over time to more accurately classify cells as CD4+ or CD8+ cells. After training, the neural network becomes able to accurately classify unknown inputs (e.g., image data and / or input features for cells not known to be either CD4+ or CD8+). That is, image data and / or input features associated with candidate cells (e.g., one, some, or all of the optical, morphological, intensity, phase, or system features described above) can be presented to a trained neural network, and the neural network can accurately provide an output corresponding to whether the candidate cells are CD4+ or CD8+.

[0102] Those skilled in the art will appreciate that many types of neural networks and their configurations (e.g., number and / or arrangement of perceptrons, such as single-layer or multi-layer) can be utilized without departing from the scope of the present invention. Similarly, various methods for training a neural network can be used without departing from the scope of the present invention. For example, in some embodiments, the neural network described above can be a convolutional neural network (CNN). In a CNN, a convolutional layer includes one or more convolutional filters. These convolutional filters can be applied to an input (e.g., image data) to generate one or more feature maps; a training process, such as that described below, can be used to arrive at feature maps that minimize the error between the desired output and the predicted result, thus improving the CNN's ability to classify candidate cells (e.g., T cells). In some examples, a CNN can receive as input image data (e.g., intensity image data, phase image data, overlay image data) associated with a candidate cell to be classified and output (e.g., predict) whether the candidate cell belongs to a first group (e.g., CD4+) or a second group (e.g., CD8+). In some embodiments, the CNN is trained using image data. In some embodiments, the CNN classifies image data associated with an unknown type of cell.

[0103] In some embodiments, a convolutional neural network can be used to determine the health of cells in a population of cells. For example, image data associated with the cells of a population can be used as input to the convolutional neural network to determine whether the cells of the population are alive or dead, and / or the proportion, total number, and / or concentration of live and dead cells in the cell population. In some embodiments, the cell attribute of alive includes a single live cell and / or a cluster of live cells. In some embodiments, the cell attribute of dead includes a non-viable cell and / or debris. In some embodiments, the convolutional neural network can classify a single live cell, a cluster of live cells, a non-viable cell, and debris. In some embodiments, the cell to be classified is a T cell in a population of cells containing T cells.

[0104] In some embodiments, a convolutional neural network can be used to determine the subtype identity of cells in a cell population. In some embodiments, image data associated with the cells of the population can be used as input to the convolutional neural network to determine the subtype identity of cells and / or the proportion, total number, and / or concentration of cell subtypes in the cell population. For example, image data associated with the cells of the population can be used as input to the convolutional neural network to determine whether the cells are CD4+ or CD8+, and / or the proportion, concentration, or total number of cells of each subtype in the cell population. In some embodiments, the cells to be classified are T cells in a population of cells containing T cells. In some embodiments, the convolutional neural network can classify CD4+ and CD8+ T cells.

[0105] In some embodiments, a convolutional neural network can be used to determine whether a cell in a population of cells expresses a recombinant molecule. In some embodiments, the recombinant molecule is a recombinant receptor. In some embodiments, the recombinant receptor is a chimeric antigen receptor (CAR). In some embodiments, the recombinant receptor is a T cell receptor (TCR). In some embodiments, image data associated with the cells of the population can be used as input to the convolutional neural network to determine the cells that express the recombinant molecule and / or the proportion, total number, and / or concentration of cells that express the recombinant molecule in the cell population. In some embodiments, the convolutional neural network can classify CAR+ and CAR- cells. In some embodiments, the cell to be classified is a T cell of a population of cells containing T cells.

[0106] In some embodiments, a convolutional neural network can be used to determine whether a cell in a population of cells is in an activated state. In some embodiments, image data associated with the cells of the population can be used as input to the convolutional neural network to determine whether the cells are activated and / or the proportion, total number, and / or concentration of activated cells in the cell population. In some embodiments, the convolutional neural network can classify activated cells and non-activated cells. In some embodiments, the cell to be classified is a T cell in a population of cells containing T cells.

[0107] In some examples, a support vector machine (SVM) can be utilized (alone or in combination with other techniques) to classify cells as belonging to a first category (e.g., CD4+ subtype) or a second category (e.g., CD8+ subtype). As will be appreciated by those skilled in the art, an SVM can include a hyperplane used to linearly separate a dataset. By testing a dataset containing cells of unknown type against a hyperplane configured to separate cells of the first category from cells of the second category, candidate cells of the dataset can be classified as belonging to either the first category or the second category.

[0108] The hyperplane of the SVM as described above can be determined through a training method. Various methods for training an SVM are well known to those skilled in the art and are within the scope of the present disclosure. For example, a training dataset can be provided that includes inputs corresponding to known outputs (e.g., image data and / or input features belonging to cells known to be CD4+ or CD8+ cells). The hyperplane can be determined using any of a variety of methods known to those skilled in the art that perform linear classification of the training dataset. In some examples, it is preferable to determine the hyperplane so that the geometric margin is maximized. After training, the SVM can use the hyperplane to classify unknown inputs (e.g., image data and / or input features for cells not known to be either CD4+ or CD8+). That is, using the determined hyperplane, image data and / or input features associated with candidate cells (e.g., one, some, or all of the optical, morphological, intensity, phase, or system features described above) can be presented to the SVM, and the SVM can accurately provide an output corresponding to whether the candidate cells are CD4+ or CD8+.

[0109] In some examples, ensemble machine learning methods can be used to classify cells. Random forests are an example of an ensemble learning method that can be utilized in some embodiments to classify cells as belonging to a first category (e.g., CD4+ subtype) or a second category (e.g., CD8+ subtype). A random forest can include multiple decision trees, each of which is applied to one or more inputs to generate one or more corresponding outputs. In some embodiments, each decision tree in a random forest can accept image data and / or input features associated with a candidate cell (e.g., one, some, or all of the optical, morphological, intensity, phase, or system features described above) as input and generate a classification of the candidate cell as output. Preferably, the individual decision trees are sufficiently uncorrelated so that a common input applied to multiple decision trees results in a diversity of outputs. In some embodiments, image data and / or input features (e.g., one, some, or all of the optical, morphological, intensity, phase, or system features described above) are applied to each of multiple individual decision trees, and the corresponding outputs are adjusted to generate the output of the random forest. For example, in some embodiments, the output of a random forest may correspond to the classification output of a majority of the individual decision trees.

[0110] Various methods for training a random forest are well known to those skilled in the art and are within the scope of the present disclosure. For example, in some embodiments, a random forest can be trained on a dataset (e.g., a set of image data corresponding to cells known to be CD4+ cells or CD8+ cells) by applying randomly sampled input data (e.g., image data, one or more input features) to each individual decision tree of the random forest.

[0111] In some cases, a random forest can be trained on a dataset (e.g., input features derived from image data corresponding to cells known to be CD4+ cells or CD8+ cells) by applying randomly sampled input data (e.g., one or more input features) to each individual decision tree of the random forest. This training method can promote diversity among the individual decision trees and improve the accuracy of the random forest. Those skilled in the art will appreciate that many suitable configurations of random forests can be utilized as appropriate. The present disclosure is not limited to any type or configuration of a random forest, its component decision trees, any method of training a decision tree, or any method of training any of the above.

[0112] In some embodiments, boosting classifiers provide further examples of ensemble learning methods that can be utilized to classify cells as belonging to a first category (e.g., CD4+ subtype) or a second category (e.g., CD8+ subtype). For example, AdaBoost can be described as a boosting classifier that combines the outputs of "weak" classifiers into a weighted sum. In some embodiments, AdaBoost can be used to classify candidate cells as belonging to a first group or a second group based on image data and / or input features corresponding to the candidate cells. Similarly, AdaBoost can be trained on training data including image data and / or input features corresponding to cells known to have one or more cellular attributes (e.g., cells known to belong to a group defined by one or more cellular attributes). Similarly, other boosting methods, such as XGBoost, can be trained and applied in a similar manner to classify cells as belonging to a first category (e.g., CD4+ subtype) or a second category (e.g., CD8+ subtype). Boosting methods, such as XGBoost, can be combined with other machine learning methods. For example, XGBoost can be used in combination with random forests to classify cells.

[0113] In some embodiments, random forests can be used to determine the health of cells in a population of cells. For example, input features determined from image data associated with the cells of a population can be used as inputs to the random forest to determine whether the cells of the population are alive or dead, and / or the proportion, total number, and / or concentration of live and dead cells in the cell population. In some embodiments, the cell attribute of alive includes a single live cell and / or a cluster of live cells. In some embodiments, the cell attribute of dead includes a non-live cell and / or debris. In some embodiments, the random forest can classify a single live cell, a cluster of live cells, a non-live cell, and debris. In some embodiments, the cell to be classified is a T cell of a population of cells containing T cells.

[0114] In some embodiments, one or more input features can be used by a random forest classifier to classify the health status of cells in a cell population, for example, T cells in a cell population containing T cells.In some embodiments, one or more input features include cell phase correlation, cell area, the number of cell refraction peaks, cell phase distortion, peak diameter, the refraction peak area normalized by cell area, radius variance, the intensity uniformity of cell image, cell compactness, cell phase intensity contrast, normalized radius variance and cell circularity.In some embodiments, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 or 12 input features selected from cell phase correlation, cell area, the number of cell refraction peaks, cell phase distortion, peak diameter, the refraction peak area normalized by cell area, radius variance, the intensity uniformity of cell image, cell compactness, cell phase intensity contrast, normalized radius variance and cell circularity can be used by a machine learning model such as a random forest classifier to classify whether cells are alive or dead. In some embodiments, at least one, two, three, four, five, six, or all of the input features selected from cell phase correlation, cell area, number of cell refractive peaks, cell phase distortion, peak diameter, refractive peak area normalized by cell area, radius variance, cell image intensity uniformity, cell compactness, cell phase intensity contrast, normalized radius variance, and cell circularity, in any combination, can be used by a machine learning model, such as a random forest classifier, to classify whether a cell is alive or dead. In some embodiments, one, two, three, or four input features selected from cell phase correlation, cell area, number of cell refractive peaks, phase distortion, and peak diameter, in any combination, can be used by a machine learning model, such as a random forest classifier, to classify whether a cell is alive or dead. In some embodiments, one or more input features include cell phase correlation. In some embodiments, one or more input features include cell area.In some embodiments, the one or more features include several refractive peaks of the cell.

[0115] In some embodiments, random forests can be used to determine the subtype identity of cells in a cell population. In some embodiments, input features determined from image data associated with the cells of a population can be used as input to random forests to determine the subtype identity of cells and / or the proportion, total number, and / or concentration of cell subtypes in a cell population. For example, input features determined from image data associated with the cells of a population can be used as input to random forests to determine whether the cells are CD4+ or CD8+, and / or the proportion, concentration, or total number of cells of each subtype in a cell population. In some embodiments, the cells to be classified are T cells of a population of cells containing T cells. In some embodiments, random forests can classify CD4+ and CD8+ T cells.

[0116] In some embodiments, one or more input features can be used by a random forest classifier to classify cell subtypes in a population of cells, for example, T cells in a population of cells containing T cells. In some embodiments, the one or more input features include an average measure of phase uniformity across the cell surface (e.g., the average phase uniformity of the cells), peak diameter, normalized refractive peak height of the cells, average phase of the cells, refractive peak area normalized by the area of ​​the cells, minimum optical height of the cells, cell compactness, circularity of the cells, phase smoothness of the cells, intensity uniformity of the image of the cells, plane (e.g., in microns) at which the intensity of the cells is maximum, area of ​​the refractive peak of the cells, phase correlation of the cells, depth of the cells, intensity contrast of the image of the cells, intensity uniformity of the image of the cells, normalized radial variance, intensity smoothness of the image of the cells, phase intensity contrast of the cells, maximum optical height of the cells in microns, average intensity contrast of the cells, average intensity uniformity of the cells, intensity skewness of the image of the cells, hu moment invariant 1, intensity variance of the image of the cells, average entropy, and intensity correlation of the image of the cells. In some embodiments, the average measure of phase uniformity across the cell surface (e.g., average phase uniformity of the cell), peak diameter, normalized refractive peak height of the cell, average phase of the cell, refractive peak area normalized by the area of ​​the cell, minimum optical height of the cell, compactness of the cell, circularity of the cell, phase smoothness of the cell, intensity homogeneity of the image of the cell, plane (e.g., in microns) at which the intensity of the cell is greatest, area of ​​the refractive peak of the cell, phase correlation of the cell, depth of the cell, intensity contrast of the image of the cell, intensity uniformity of the image of the cell, normalized radial variance, intensity smoothness of the image of the cell, in any combination. One, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen, or more or all of the input features selected from: degree, phase intensity contrast of the cell, maximum optical height in microns of the cell, mean intensity contrast of the cell, mean intensity uniformity of the cell, intensity skewness of the image of the cell, hu moment invariant 1, intensity variance of the image of the cell, mean entropy, and intensity correlation of the image of the cell can be used by a machine learning model, such as a random forest classifier, to classify whether a cell is a particular subtype (e.g., CD4 or CD8 T cell).In some embodiments, the average measure of phase uniformity across the cell surface (e.g., average phase uniformity of the cell), peak diameter, normalized refractive peak height of the cell, average phase of the cell, refractive peak area normalized by the area of ​​the cell, minimum optical height of the cell, compactness of the cell, circularity of the cell, phase smoothness of the cell, intensity uniformity of the image of the cell, plane (e.g., in microns) at which the intensity of the cell is greatest, area of ​​the refractive peak of the cell, phase correlation of the cell, depth of the cell, intensity contrast of the image of the cell, intensity uniformity of the image of the cell, normalized radial variance, intensity of the image of the cell, At least one, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen, or all of the input features selected from smoothness, phase intensity contrast of cells, maximum optical height in microns of cells, mean intensity contrast of cells, mean intensity uniformity of cells, intensity skewness of images of cells, hu moment invariant 1, intensity variance of images of cells, mean entropy, and intensity correlation of images of cells can be used by a machine learning model, such as a random forest classifier, to classify whether cells are of a particular subtype (e.g., CD4 or CD8 T cells).

[0117] In some embodiments, one or more input features can be used by a random forest classifier to classify cell subtypes in a population of cells, for example, T cells in a population of cells containing T cells. In some embodiments, the one or more input features include cell depth, cell image intensity contrast, cell image intensity uniformity, normalized radius variance, cell image intensity smoothness, cell phase intensity contrast, cell maximum optical height in microns, mean intensity contrast, cell image mean intensity uniformity, cell image intensity skewness, hu moment invariant 1, cell image intensity variance, mean entropy, and cell image intensity correlation. In some embodiments, one, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen input features selected from cell depth, cell image intensity contrast, cell image intensity uniformity, normalized radial variance, cell image intensity smoothness, cell phase intensity contrast, cell maximum optical height in microns, cell mean intensity contrast, cell mean intensity uniformity, cell image intensity skewness, hu moment invariant 1, cell image intensity variance, mean entropy, and cell image intensity correlation, in any combination, can be used by a machine learning model, such as a random forest classifier, to classify whether a cell is a particular subtype (e.g., a CD4 or CD8 T cell). In some embodiments, at least one, two, three, four, five, six, or all input features selected from cell depth, cell image intensity contrast, cell image intensity uniformity, normalized radial variance, cell image intensity smoothness, cell phase intensity contrast, cell maximum optical height in microns, cell mean intensity contrast, cell mean intensity uniformity, cell image intensity skewness, hu moment invariant 1, cell image intensity variance, mean entropy, and cell image intensity correlation, in any combination, can be used by a machine learning model, such as a random forest classifier, to classify whether a cell is a particular subtype (e.g., a CD4 or CD8 T cell).In some embodiments, one, two, three, four, five, or six input features selected from cell depth, cell image intensity contrast, cell image intensity uniformity, normalized radial variance, cell image intensity smoothness, and cell phase intensity contrast, in any combination, can be used by a machine learning model, such as a random forest classifier, to classify whether a cell is a certain subtype (e.g., CD4 or CD8 T cell). In some embodiments, at least one, two, three, four, five, or six input features selected from cell depth, cell image intensity contrast, cell image intensity uniformity, normalized radial variance, cell image intensity smoothness, and cell phase intensity contrast, in any combination, can be used by a machine learning model, such as a random forest classifier, to classify whether a cell is a certain subtype (e.g., CD4 or CD8 T cell). In some embodiments, one or more features selected from cell depth, cell image intensity contrast, and cell image intensity uniformity, in any combination, can be used by a machine learning model, such as a random forest classifier, to classify whether a cell is a certain subtype (e.g., CD4 or CD8 T cell). In some embodiments, the one or more input features include cell depth. In some embodiments, the one or more characteristics include intensity contrast, hi some embodiments, the one or more characteristics include intensity uniformity.

[0118] In some embodiments, a random forest can be used to determine whether a cell in a population of cells expresses a recombinant molecule. In some embodiments, the recombinant molecule is a recombinant receptor. In some embodiments, the recombinant receptor is a chimeric antigen receptor (CAR). In some embodiments, the recombinant receptor is a T cell receptor (TCR). In some embodiments, input features determined from image data associated with cells of the population can be used as input to a random forest to determine the percentage, total number, and / or concentration of cells expressing the recombinant molecule and / or cells expressing the recombinant molecule in the cell population. In some embodiments, the random forest can classify CAR+ and CAR- cells. In some embodiments, the cell to be classified is a T cell of a population of cells containing T cells.

[0119] In some embodiments, random forest can be used to determine whether the cells in a cell population are in an activated state.In some embodiments, the input features determined from the image data associated with the cells of the population can be used as input to random forest to determine whether the cells are activated and / or the proportion, total number, and / or concentration of activated cells in the cell population.In some embodiments, random forest can classify activated cells and non-activated cells.In some embodiments, the cell to be classified is a T cell of a cell population containing T cells.

[0120] In some embodiments, a preprocessing step can be applied to image data associated with cells (e.g., T cells) to generate parameters (e.g., values) corresponding to individual input features (e.g., any of the optical, morphological, intensity, phase, or system features described above), and those parameters are then provided as input to a machine learning model (e.g., any of the machine learning models described above). For example, the preprocessing step can operate on the image data of a candidate cell, e.g., a 3D array of RGB or grayscale values, with each RGB or grayscale value corresponding to a pixel in that image. In some embodiments, the preprocessing includes removing or reducing noise (e.g., background noise) in the image data associated with the cells. In some embodiments, the preprocessing includes normalizing the image data associated with the cells. The preprocessing step can calculate intensity values, optical height values, contrast values, or any other suitable parameters (such as those described above) from the image, and those parameters can then be applied to the machine learning model.

[0121] In some embodiments, more than one machine learning model can be used to classify cells.For example, when the cell population to be classified goes through a manufacturing process, for example, the process described herein, the classification method can be used at any time or step in the manufacturing process to classify the cells of the population.It is believed that some machine learning model can perform better to classify cells at different time or step.In some embodiments, for example, one or more machine learning models as described above can be used at different time or during different steps of the manufacturing process, for example.

[0122] 1. Training method In some embodiments, the machine learning model can be trained using training data. In some embodiments, the training data is or includes image data associated with cells (e.g., T cells). In some embodiments, the training data is or includes input features determined from image data associated with cells (e.g., T cells). In some embodiments, the training data is associated with cells contained in a cellular dataset, for example, as described below. The training of the algorithm can be achieved through supervised, unsupervised, or semi-supervised training methods.

[0123] In some embodiments, the training of the machine learning model is supervised. In this case, the training data (e.g., image data, input features) used to train the machine learning model is derived from cells known to have one or more cellular attributes indicating that the cells belong to a particular group, where the group is defined by one or more cellular attributes. For example, the training data may include image data and / or input features collected from cells (e.g., T cells) with known cellular attributes (e.g., CD4+, CD8+, transduced cells). In this manner, the model is trained (e.g., learns) that certain data (e.g., image data, input features) correspond to cells (e.g., T cells) with particular cellular attributes (e.g., CD4+, CD8+, transduced cells). Thus, when new data (e.g., image data or input features) from cells with unknown cellular attributes are presented to the model, the model can classify (e.g., predict) that the cells belong to a group defined by one or more cellular attributes based on previous training. In some embodiments, the image data is obtained by DDHM. In some embodiments, the input features are determined from image data obtained by DDHM.

[0124] In some embodiments, the image data associated with cells having known cellular attributes is manipulated to generate additional data sets for supervised training of machine learning models.In some embodiments, the image data associated with cells having known cellular attributes is manipulated by zooming, tilting, and / or rotating.This strategy can be useful for enhancing learning ability when dealing with small training data sets or training data sets that lack diversity.In some embodiments, the image data is obtained by DDHM.

[0125] Alternatively, in some embodiments, the training of the machine learning model is unsupervised.In this case, the training data (e.g., image data, input features) used to train the model are derived from cells whose cellular attributes are unknown.Therefore, when the model is presented with data (e.g., image data, input features) of unknown origin (e.g., unknown cell type, unknown transduction success), the model associates new data with the cluster generated by unsupervised data.The model can classify new data as part of the existing cluster generated by unsupervised training data, or as part of a cluster different from the existing cluster, or as an outlier.The association of new data with the cluster can be achieved by the model through the use of centroid-based clustering techniques, such as k-means clustering.

[0126] The data set that can be used for training machine learning model can be divided or split in any suitable manner to generate the data set that is used for training, the set that is used for testing, and / or the set that is used for validation.For example, in some embodiments, the data set is divided or split so that 80% of data is used for training machine learning model, and 20% is used for testing machine learning model.Exemplary division or splitting of data for training and testing machine learning model can be found in Section VI below.

[0127] In some embodiments, training data is balanced.In some embodiments, test data is balanced.For example, if the population of cells belonging to the first group is larger than the population of cells belonging to the second group, the first group can be randomly sampled to create a population of cells that is the same size as the second group.

[0128] In some embodiments, the machine learning model is trained to identify cells (e.g., T cells) as belonging to a first or second group, where the first group is defined by one or more cellular attributes and the second group is defined by one or more cellular attributes, at least one attribute of which differs from the one or more attributes defining the first group. In some embodiments, the machine learning model is trained to classify cells (e.g., T cells) as belonging to one of three or more or multiple groups, where the groups are defined by one or more cellular attributes and at least one attribute in each of the three or more or multiple groups differs from each of the other three or more or multiple groups.

[0129] In some embodiments, the machine learning model is trained on training data that includes image data (e.g., one or more of phase data, intensity data, and overlay data) from a plurality of cells known to have one or more specific cellular attributes, thereby belonging to one of two or more groups, and the machine learning model is trained to distinguish between cells in these groups. In some embodiments, the machine learning model is trained on image data that includes phase data, intensity data, and overlay data. In some embodiments, the machine learning model is trained on image data that includes phase data and intensity data. As described above, in some embodiments, the image data is manipulated by zooming, tilting, and / or rotating. Thus, in some cases, additional training data can be generated, for example, from existing image data. In some embodiments, the image data is obtained using DDHM. In some embodiments, the convolutional neural network is trained using the image data described herein.

[0130] In some embodiments, the machine learning model is trained on training data that includes one or more input features (e.g., one or more of the features described above) determined from image data from a plurality of cells known to have one or more particular cellular attributes, whereby the training data belongs to one of two or more groups, and the machine learning model is trained to distinguish between cells in these groups.

[0131] In some embodiments, the machine learning model may select an input feature subset from a larger set of input features using classification based on the selected subset. The extracted input feature subset may be used for learning, as determined from feature selection and ranking, feature combination, or other processes. Exemplary features that may be used are described above.

[0132] In some embodiments, input features are derived from image data obtained using DDHM. In some embodiments, an initial analysis is performed to determine which input features correlate with cellular attributes. These input features can then be used as inputs to train a machine learning model. In some embodiments, statistical tests can be performed to identify highly correlated input features. In some embodiments, statistical tests can be performed to identify input features that display low variance. In some embodiments, statistical tests can be performed to identify input features that display low variance and to determine correlated input features. In some embodiments, input features with low variance are excluded as inputs to train the machine learning model. In some embodiments, only one input feature of a pair of correlated input features is used as an input to train the machine learning model. In some embodiments, a random forest model, neural network, or SVM is trained using the input features. In some embodiments, the input features used for training and / or classification undergo initial analysis as described herein. In some embodiments, the features that undergo initial analysis can be used to improve model performance.

[0133] The machine learning model described herein can be used to output a classification of a candidate T cell as belonging to a category based on one or more input features. Such input features can include one or more of the input features described above and can be derived from image data of the candidate T cell. In some embodiments, the machine learning model classifies the cell into one of one or more clusters that group cells with known classifications (e.g., CD4+, CD8+, CAR+, CAR-) based on at least one or more or multiple input features (e.g., as described above). In some embodiments, the machine learning model can generate multiple clusters that group variables with known classifications (e.g., CD4+, CD8+, CAR+, CAR-) by applying one or more density-based clustering algorithms.

[0134] D. How to generate a cell dataset Also provided herein are methods for generating cellular datasets for training machine learning models. Training data (e.g., image data, input features) for training machine learning models can be derived from cells contained in the cellular dataset. The cellular datasets are generated as described herein to contain cells with known cellular attributes (e.g., cells belonging to a specific group). In some embodiments, the cellular datasets include cells known to belong to a specific group. For example, a first cellular dataset generated according to the methods described herein can contain CD4+ cells, and a second cellular dataset generated according to the methods described herein can contain CD8+ cells. In this manner, the first cellular dataset contains cells belonging to a first group, where the first group is defined by the cellular attribute of expressing CD4 (e.g., being surface positive), and the second cellular dataset contains cells belonging to a second group, where the second group is defined by the cellular attribute of expressing CD8 (e.g., being surface positive). The methods provided herein for generating cellular datasets enable the generation of cellular datasets containing cell populations known to belong to a specific group defined by one or more cellular attributes. As noted above, cell attributes include, but are not limited to, cell health (e.g., alive, dead), cell cycle status, differentiation status, activation status, cell subtype (CD4+, CD8+ T cell), and / or engineering status (e.g., whether the T cell expresses a recombinant receptor (e.g., chimeric antigen receptor (CAR), T cell receptor (TCR))).

[0135] The cellular dataset that can obtain training data from can be generated using different techniques.The selection of the technique for generating the cellular dataset that is useful for training machine learning model depends on many factors, including but not limited to, how the cell population to be classified is processed.For example, it can be beneficial to train machine learning model on the training data from the cells in the cellular dataset that are processed in the same way as the cell population to be classified, for example, it can increase accuracy and / or robustness.In some embodiments, training data is derived from the cellular dataset that is processed in the same or substantially the same way as the cell population to be classified.

[0136] 1. Pure cell populations Pure cell population can serve as the cellular data set for training machine learning model.For example, in some embodiments, by selecting cells from sample (for example, blood purification therapy, leukopheresis therapy, blood, PBMC sample) based on desired attribute, thereby isolating the group of selected cells with the same attribute, pure population is generated, thus creating a cellular data set.The selected cell group only contains the cells with the attribute, so it can be called pure population.

[0137] The advantage of the data set of cells that contain pure population is that all cells of the population are known to belong to a certain group, and can directly obtain training data (for example, image data, input features) from the data set of cells that contain pure population, and can be used for supervised training of machine learning model.For example, T cell subtypes (for example, CD4+ or CD8+) can be isolated, collected, and / or processed (for example, stimulated; transduced; cultured; incubated; for example, produced according to the method used herein) from the mixed population of cells in sample (for example, blood purification therapy, leukopheresis therapy, blood, PBMC sample).In some embodiments, pure population is collected and / or processed in the same or substantially the same manner as the collection and / or processing of the population to be classified.

[0138] To obtain training data (e.g., image data, input features) without the need for further processing (e.g., selecting, isolating, separating cells from a pure population), a cellular dataset containing cells of a pure population can be directly evaluated (e.g., imaged via the imaging techniques described herein). In some embodiments, a cellular dataset containing cells of a pure population can be directly evaluated using DDHM.

[0139] 2. Mixed Cell Populations As an alternative to a pure population, a cellular dataset can be derived from a mixed population of cells. A mixed cell population differs from a pure population in that the cells contained in the mixed population do not necessarily have the same cellular attributes. In some embodiments, the mixed cell population is a sample (e.g., blood purification therapy, leukapheresis therapy, blood, PBMC sample). In some embodiments, the mixed cell population contains cells comprising one or more cellular attributes, and each cell of the mixed population does not necessarily comprise the same cellular attribute. For example, T cell subtypes (e.g., CD4+ and CD8+) can be isolated from a mixed population of cells in a sample (e.g., blood purification therapy, leukapheresis therapy, blood, PBMC sample), and the isolated populations are recombined, e.g., in a specific ratio, to generate a mixed population of cells. In some embodiments, the mixed cell population is collected and / or processed in the same or substantially the same manner as the collection and / or processing of the population to be sorted.

[0140] Unlike pure populations, training data obtained from mixed populations cannot be directly used for supervised training because the identity of the cells (e.g., the group to which the cells belong) is unknown. Thus, in some embodiments, to obtain useful training data from a mixed cell population, cells of the mixed population having the same one or more cellular attributes can be separated from the mixed population, thereby generating a cellular dataset. In some embodiments, the separated cells (e.g., cells of the cellular dataset) are evaluated (e.g., imaged via the imaging techniques described herein) to obtain training data (e.g., image data, input features).

[0141] In some embodiments, the method for separating cells having one or more of the same attributes from a mixed cell population to generate a cellular dataset does not change the physiological or functional state of the cells. In some embodiments, the method for separating cells from a mixed population does not result in cells that retain the agents and reagents used to separate the cells. Such an embodiment may occur in certain methods for separating cells from a mixed population, minimizing or reducing the ability of cells to be altered in some way, which may result in training data obtained from the cellular dataset that is not useful for teaching a machine learning model what data (e.g., image data, input features) are useful for classification. Similarly, if the cells of the cellular dataset retain the agents and reagents used to separate (e.g., isolate) the cells, the training data obtained from this cellular dataset may not be useful for teaching a machine learning model what data (e.g., image data, input features) are useful for classification. In some embodiments, the separation method does not change the physiological or functional state of the cells. In some embodiments, the separation method results in separated cells that are free or substantially free of the agents and reagents used to perform the separation. In some embodiments, substantially free includes when less than or about 30%, 25%, 20%, 15%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, or 1% of the agents and reagents used to perform the separation are present in the separated cell population. Separation methods that may be suitable for separating mixed populations are described in Section ID-2-a below and in U.S. Patent Application Publication Nos. US2015 / 0024411 and US2017 / 0037369, both of which are incorporated herein by reference in their entireties.

[0142] Generating a data set of cells from a mixed population can be particularly useful for training machine learning models when the cell population to be classified is itself a mixed population.For example, when the cell population to be classified contains multiple cell subtypes (e.g., CD4+ and CD8+ T cells), and / or when cells are transduced or transfected so that only some cells in the population to be classified express recombinant molecules (e.g., CAR, TCR), the data set of cells from a mixed population can be useful for training machine learning models.Cell signaling and / or other environmental conditions in a mixed cell population may be different from those in a pure population due to the existence of different cell types.The ability to generate a data set of cells from a mixed cell population has the advantage that the cells in the mixed population experience the same conditions as the population to be classified.

[0143] a. Immunoaffinity-based isolation In some embodiments, the separation method involves separating different cell types from a mixed population based on the expression or presence of one or more specific molecules, such as surface molecules, e.g., surface proteins, on the cell surface. In some embodiments, the expression of a surface molecule indicates that the cell is surface-positive for that molecule. Conversely, in some embodiments, if the cell does not express the surface molecule, the cell is surface-negative for the molecule.

[0144] In some embodiments, separation is based on immunoaffinity.For example, in some aspects, isolation comprises the separation of cells and cell populations based on the expression or expression level of one or more molecules, typically by incubation with antibody or binding agent that specifically binds to cell surface molecules, and this molecule, generally followed by washing step and separation of the cells that bind to antibody or binding agent from the cells that do not bind to antibody or binding agent.

[0145] Such separation steps can be based on positive selection, in which cells bound to the antibody or binder are retained for further use, and / or negative selection, in which cells that do not bind to the antibody or binder are retained. In some instances, both fractions are retained for further use. In some aspects, when antibodies or binders that specifically identify cell types in a heterogeneous population are not available, negative selection can be particularly useful, as separation is best performed based on molecules expressed by cells other than the desired population.

[0146] Separation does not necessarily result in 100% enrichment or removal of a particular cell population or cells expressing a particular marker. For example, positive selection or enrichment of a particular type of cell, such as cells expressing a surface molecule, refers to increasing the number or proportion of such cells, but does not necessarily result in the complete absence of cells that do not express the surface molecule. Similarly, negative selection, removal, or depletion of a particular type of cell, such as cells expressing a surface molecule, refers to reducing the number or proportion of such cells, but does not necessarily result in the complete removal of all such cells. For example, in some aspects, selection of one of the CD4+ or CD8+ populations will enrich either the CD4+ or CD8+ population, but may also contain some remaining or a small percentage of other unselected cells, and in some cases, may include the other of the CD4 or CD8 population still present in the enriched population.

[0147] In some examples, multiple separation steps are performed, in which the positively or negatively selected fraction from one step is subjected to another separation step, such as subsequent positive or negative selection.In some examples, a single separation step can simultaneously deplete cells that express multiple markers, such as by incubating cells with multiple antibodies or binding agents, each specific for the marker targeted for negative selection.Similarly, multiple cell types can be simultaneously positively selected by incubating cells with multiple antibodies or binding agents that are expressed on various cell types.

[0148] In some embodiments, affinity-based selection employs immunoaffinity chromatography. The immunoaffinity chromatography method, in some aspects, includes one or more chromatography matrices described in U.S. Patent Application Publication No. US2015 / 0024411. In some embodiments, the chromatography method is fluid chromatography, typically liquid chromatography. In some embodiments, chromatography can be performed in a flow-through mode, e.g., by gravity flow or by a pump at one end of a column containing the chromatography matrix, with the fluid sample exiting the column at the other end. Furthermore, in some aspects, chromatography can be performed in an "up-and-down" mode, e.g., by a pipette at one end of a column containing a chromatography matrix packed in a pipette tip, with the fluid sample entering and exiting the chromatography matrix / pipette tip at the other end of the column. In some embodiments, chromatography can also be performed in a batch mode, in which the chromatography material (stationary phase) is incubated with the cell-containing sample, e.g., by shaking, rotation, or repeated contact and removal of the fluid sample, e.g., by a pipette.

[0149] In some embodiments, the chromatography matrix is ​​a stationary phase. In some embodiments, the chromatography is column chromatography. In some embodiments, any suitable chromatography material can be used. In some embodiments, the chromatography matrix has the form of a solid or semi-solid phase. In some embodiments, the chromatography matrix can comprise a polymeric resin or a metal oxide or semi-metal oxide. In some embodiments, the chromatography matrix is ​​a non-magnetic or non-magnetizable material. In some embodiments, the chromatography matrix is ​​a natural polymer, e.g., a derivatized silica or cross-linked gel, such as a form of polysaccharide. In some embodiments, the chromatography matrix is ​​an agarose gel. Agarose gels for use in chromatography matrices are known in the art and in some aspects include Sepharose materials such as Superflow™ agarose or Superflow™ Sepharose®, which are commercially available in different bead and pore sizes. In some embodiments, the chromatography matrix is ​​a specific cross-linked agarose matrix to which dextran is covalently attached, such as those known in the art, e.g., in some aspects, Sephadex®, Superdex®, or Sephacryl®, which are available in different bead and pore sizes.

[0150] In some embodiments, the chromatography matrix is ​​made of a synthetic polymer, such as polyacrylamide, styrene-divinylbenzene gel, copolymers of acrylates and diols or copolymers of acrylamide and diols, copolymers of polysaccharides and agarose, such as polyacrylamide / agarose complexes, polysaccharides and N,N'-methylenebiscarylamide, or derivatized silica coupled to a synthetic or natural polymer.

[0151] In some embodiments, the chromatography matrix, such as agarose beads or other matrix, has a size of at least or at least about 50 μm, 60 μm, 70 μm, 80 μm, 90 μm, 100 μm, 120 μm, or 150 μm or more. The exclusion limit of the size exclusion chromatography matrix is ​​selected to be less than the maximum width of the target cells, e.g., T cells, in the sample. In some embodiments, the volume of the matrix is ​​at least 0.5 mL, 1 mL, 1.5 mL, 2 mL, 3 mL, 4 mL, 5 mL, 6 mL, 7 mL, 8 mL, 9 mL, 10 mL, or more. In some embodiments, the chromatography matrix is ​​packed into a column.

[0152] In some embodiments, the chromatography matrix, which is an immunoaffinity chromatography matrix, comprises a binding agent, such as an antibody or an antigen-binding fragment, such as a Fab, immobilized on the chromatography matrix. In some embodiments, the antibody is a full-length antibody or a variable heavy chain (V) capable of specifically binding an antigen, such as a (Fab) fragment, a F(ab')2 fragment, a Fab' fragment, an Fv fragment, or a variable heavy chain (V) capable of specifically binding an antigen. H ) region, single-chain antibody fragments including single-chain variable fragments (scFv), and single-domain antibodies (e.g., sdAb, sdFv, nanobodies) and antigen-binding fragments thereof. In some embodiments, the antibody is a Fab fragment. In some embodiments, the antibody can be monovalent, bivalent, or multivalent. In some embodiments, the binding agent comprises a monovalent binding site. In some embodiments, the monovalent binding site is a Fab fragment, sdAb, Fv fragment, or single-chain Fv fragment.

[0153] In some aspects, the antibody or antigen-binding fragment thereof can be produced by or derived from a hybridoma, such as: OKT3 (αCD3), 13B8.2 (αCD4), or OKT8 (αCD8). In some embodiments, any of the above antibodies can contain one or more mutations within the framework of the heavy and light chain variable regions without targeting the highly variable CDR regions. Examples of such antibodies, in some aspects, include anti-CD4 antibodies, such as those described in U.S. Patent No. 7,482,000 and Bes et al. (2003) J. Biol. Chem., 278: 14265-14273.

[0154] In some embodiments, the antibody or antigen-binding fragment thereof is an anti-idiotypic antibody or antigen-binding fragment thereof (anti-ID). In some embodiments, the anti-ID binds to a target antigen receptor, e.g., a CAR.

[0155] In some embodiments, the antibody or antigen-binding fragment thereof specifically binds to a surrogate marker. The surrogate marker can be used to detect cells into which a polynucleotide, such as a polynucleotide encoding a recombinant receptor, has been introduced. In some embodiments, the surrogate marker can indicate or confirm cellular modification. In some embodiments, the surrogate marker is a protein co-expressed with the recombinant receptor, such as a CAR, on the cell surface. In certain embodiments, such a surrogate marker is a surface protein modified to have little or no activity. In certain embodiments, the surrogate marker is encoded on the same polynucleotide as the recombinant receptor. In some embodiments, the nucleic acid sequence encoding the recombinant receptor is operably linked to a nucleic acid sequence encoding a marker, or a nucleic acid encoding a self-cleaving peptide or a peptide that causes ribosome skipping, such as a 2A sequence, for example, T2A, P2A, E2A, or F2A, optionally separated by an internal ribosome entry site (IRES). In some cases, an exogenous marker gene can be utilized in conjunction with engineered cells to enable cell detection or selection, and in some cases, to facilitate cell suicide.

[0156] Exemplary surrogate markers can include truncated forms of cell surface polypeptides, such as truncated forms that are non-functional, do not or cannot transduce signals or signals normally transduced by the full-length form of the cell surface polypeptide, and / or do not or cannot be internalized. Exemplary truncated cell surface polypeptides include truncated growth factors or other receptors, such as truncated human epidermal growth factor receptor 2 (tHER2), truncated epidermal growth factor receptor (EGFRt, exemplary EGFRt sequences shown in SEQ ID NO: 23 or 24), or prostate-specific membrane antigen (PSMA), or modified forms thereof. EGFRt may contain an epitope recognized by the antibody cetuximab (Erbitux®) or other therapeutic anti-EGFR antibodies or binding molecules, which can be used to identify or select cells engineered with EGFRt constructs and recombinant receptors, such as chimeric antigen receptors (CARs), and / or to eliminate or separate cells expressing the receptor. See U.S. Patent No. 8,802,374 and Liu et al., Nature Biotech. 2016 April; 34(4): 430-434. In some aspects, the marker, e.g., a surrogate marker, comprises all or a portion (e.g., a truncated form) of CD34, NGFR, CD19, or a truncated CD19, e.g., a truncated non-human CD19, or epidermal growth factor receptor (e.g., tEGFR). In some embodiments, the nucleic acid encoding the marker is operably linked to a polynucleotide encoding a cleavable linker sequence, e.g., a linker sequence such as T2A. For example, the marker and optional linker sequence can be any of those disclosed in PCT Publication No. WO2014031687. For example, the marker can be a truncated EGFR (tEGFR), optionally linked to a linker sequence such as a T2A cleavable linker sequence.Exemplary polypeptides for truncated EGFR (e.g., tEGFR) include the sequence of amino acids set forth in SEQ ID NO:23 or 24, or a sequence of amino acids that exhibits at least 85%, 86%, 87%, 88%, 89%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99% or more sequence identity to SEQ ID NO:23 or 24.

[0157] In some embodiments, antigen-binding fragments such as Fab fragments can be generated from such antibodies using methods known in the art, for example, in some aspects, cloning, which allows amplification of the hypervariable sequences of the heavy and light chains and combination with sequences encoding appropriate constant domains. In some embodiments, the constant domains are of the human subclass IgG1 / κ. Such antibodies can be fused at the carboxy terminus to a peptide streptavidin-binding molecule such as that shown in SEQ ID NO: 10. Examples of such antibodies are described in Stemberger et al. (2102) PLoS One, 7: 35798 and International PCT Application No.

[0158] The antibody or antigen-binding fragment, such as a Fab, may in some aspects be any antibody known in the art, a specific k off It can be any of the above, including antibodies with a specific dissociation constant and / or antibodies with a specific dissociation constant.

[0159] In some embodiments, the dissociation constant (K D ) is about 10 -2 M ~ about 10 -11 M, or about 10 -2 M ~ about 10 -10 M, or about 10 -2 M ~ about 10 -9 M, or about 10 -2 M ~ about 10 -8 M, or about 10 -2 M ~ about 10 -7 M, or about 10 -2 M ~ about 10 -6 M, or about 10 -2 M ~ about 10-5 , or about 10 -2 M ~ about 10 -4 M, or about 10 -2 ~about 10 -3 or any number range therebetween. In some embodiments, the dissociation constant (K D ) is about 10 -3 M ~ about 10 -10 M, or about 10 -3 M ~ about 10 -9 M, or about 10 -3 M ~ about 10 -8 M, or about 10 -3 M ~ about 10 -7 M, or about 10 -3 M ~ about 10 -6 M, or about 10 -3 M ~ about 10 -5 M, or about 10 -3 M ~ about 10 -4 or any number range therebetween. In some embodiments, the dissociation constant (K D ) is about 10 -3 M ~ about 10 -7 M, or about 10 -3 M ~ approx. 0.5×10 -7 M, or about 10 -3 M ~ about 10 -6 M, or about 10 -3 M ~ approx. 0.5×10 -6 M, or about 10 -3 M ~ about 10 -5 M, or about 10 -3 M ~ approx. 0.5×10 -5 M, or about 10 -3 M ~ about 10 -4 , or about 10 -3 M ~ approx. 0.5×10 -4 or any number range therebetween.

[0160] In some embodiments, the dissociation rate constant (k off ) is approximately 3 × 10 -5 seconds -1The dissociation rate constant (k ) can be a constant that characterizes the dissociation reaction of the complex formed between the binding site of the binder and the surface molecule on the surface of the target cell. The association rate constant (k ) can be a constant that characterizes the dissociation reaction of the complex formed between the binding site of the binder and the surface molecule on the surface of the target cell. on ) can have any value. To ensure sufficient reversible binding between the surface molecule and the binding agent, it is recommended to use a value of about 3×10 -5 seconds -1 or more, about 5 x 10 -5 seconds -1 or more, for example, about 1 × 10 -4 seconds -1 or more, about 1.5 x 10 -4 seconds -1 or more, approximately 2.0 x 10 -4 seconds -1 or more, about 2.5 x 10 -4 seconds -1 or more, about 3 x 10 -4 seconds -1 or more, about 3.5 x 10 -4 seconds -1 or more, about 4 x 10 -4 seconds -1 or more, about 5 × 10 -4 seconds -1 or more, about 7.5 x 10 -4 seconds -1 or more, about 1 x 10 -3 seconds -1 or more, about 1.5 x 10 -3 seconds -1 or more, about 2 x 10 -3 seconds -1 or more, about 2.5 x 10 -3 seconds -1 or more, about 3 x 10 -3 seconds -1 or more, about 4 x 10 -3 seconds -1 , about 5×10 -3 seconds -1 or more, about 7.5 x 10 -3 seconds -1 or more, about 1 x 10 -2 seconds -1or more, about 5 x 10 -2 seconds -1 or more, about 1 x 10 -1 seconds -1 or more, or about 5 × 10 -1 seconds -1 or higher value of k off It is advantageous to select a value of k off speed, k on Speed ​​or K D As used herein with respect to (see below), the term "about" is meant to include a tolerance of ±20.0%, including ±15.0%, ±10.0%, ±8.0%, ±9.0%, ±7.0%, ±6.0%, ±5.0%, ±4.5%, ±4.0%, ±3.5%, ±3.0%, ±2.8%, ±2.6%, ±2.4%, ±2.2%, ±2.0%, ±1.8%, ±1.6%, ±1.4%, ±1.2%, ±1.0%, ±0.9%, ±0.8%, ±0.7%, ±0.6%, ±0.5%, ±0.4%, ±0.3%, ±0.2%, ±0.1%, or ±0.01%. It should be noted that the values ​​of the rate constants and thermodynamic constants used herein refer to conditions of atmospheric pressure, i.e., 1.013 bar, and room temperature, i.e., 25° C. In some embodiments, the strength of binding, i.e., the dissociation constant (K d ) but with low affinity, e.g., about 10 -3 ~about 10 -7 K of M d or high affinity, e.g., about 10 -7 ~Approx. 1×10 -10 K of M d Regardless of whether the binding site is within the range of 1000 or 10000, the target cell can be reversibly bound as long as the dissociation of the binding agent from the surface molecule via the binding site occurs sufficiently quickly.

[0161] In some embodiments, the binding agent has a single (monovalent) binding site capable of specifically binding to a surface molecule. In some embodiments, the binding agent has at least two (i.e., multiple binding sites, including three, four, or five identical binding sites) capable of binding to a surface molecule. In any of these embodiments, binding of a surface molecule via (each of) the binding site is greater than or equal to about 3×10 -5 sec -1 or more k off Thus, a binding agent can be monovalent (e.g., a monovalent antibody fragment or a monovalent artificial binding molecule (proteinaceous or otherwise), such as a mutein based on a polypeptide of the lipocalin family (also known as "Anticalin®"), or a bivalent molecule, such as an antibody or fragment in which both binding sites are retained, e.g., an F(ab')2 fragment. In some embodiments, k off Speed ​​is 3x10 -5 seconds -1 or more, the surface molecule can be a multivalent molecule, such as a pentameric IgE molecule. In some embodiments, providing for (traceless) isolation of biological material via the reversible cell immunoaffinity chromatography techniques described herein is achieved by, at the molecular level, a ratio of (3×10) of binding of a binder via a binding site to a surface molecule on a target cell. -5 seconds -1 or more)k off Rather, as described, for example, in U.S. Pat. No. 7,776,562 or International Publication No. WO 02 / 054065, low affinity binding between a surface molecule and the binding site of a binder, along with the avidity effect mediated through a multimerization reagent, allows for reversible and traceless isolation of target cells. As noted above, such low binding affinity is approximately 1.0×10 for binding of a binder to a surface molecule on the target cell surface via the binding site. -3 M ~ approx. 1.0×10 -7 Dissociation constants (K DIn these embodiments, a complex can form between the binding site of the multimerization reagent and the binding partners (see below) contained in at least two binding agents, allowing for reversible immobilization and subsequent elution of target cells from the affinity chromatography matrix.

[0162] In some embodiments, the addition of a competing reagent can disrupt the bond (complex) formed between the binding partner and the binding site of the multimerization reagent, which subsequently results in the dissociation of the binding agent from the target cell.

[0163] In some embodiments, to generate a cellular dataset, cells can be separated from a mixed population of cells, where the cells of the cellular dataset do not contain the reagents and active factors used to separate the cells. The separation methods provided herein allow for the generation of a cellular dataset that does not contain the active factors and reagents used to separate (e.g., isolate) the cells from the mixed population.

[0164] In some embodiments, the binding agent is immobilized. In some embodiments, the binding agent is fused or linked to a binding partner that interacts with a multimerization reagent immobilized on the matrix. In some embodiments, the binding capacity of the chromatography matrix is ​​at least 1×10 7 cells / mL, 5×10 7 cells / mL, 1×10 8 cells / mL, 5×10 8 cells / mL, 1×10 9 It is sufficient to adsorb or is capable of adsorbing 100,000 cells / ml, or more, where the cells express a cell surface molecule that is specifically recognized by the binding site of the binding agent.

[0165] In some embodiments, the interaction between the multimerization reagent and the binding partner forms a reversible bond such that binding of the binding agent to the matrix is ​​reversible. In some embodiments, the reversible bond can be mediated by a streptavidin mutant binding partner and a multimerization reagent immobilized on a matrix that is streptavidin, a streptavidin mutein, avidin, or an avidin mutein.

[0166] In some embodiments, the reversible binding of the binding agent is via a peptide ligand binding reagent and a streptavidin mutein interaction.

[0167] In some embodiments, affinity-based selection uses a multimerization reagent comprising a streptavidin mutein, such as Strep-Tactin® or Strep-TactinXT® (see, e.g., U.S. Patent No. 6,103,493, International Published PCT Application Nos. WO2013011011, WO2014 / 076277). In some embodiments, the streptavidin mutein is functionalized, coated, and / or immobilized on a matrix.

[0168] In some embodiments, the streptavidin mutein exhibits a higher binding affinity for a peptide ligand containing the sequence of amino acids set forth in any of SEQ ID NOs:1-6, such as SEQ ID NO:5 and / or SEQ ID NO:6 (e.g., Strep-tagII®), than unmodified or wild-type streptavidin, e.g., unmodified or wild-type streptavidin, e.g., as set forth in SEQ ID NO:11 or SEQ ID NO:14. In some embodiments, the streptavidin variant exhibits a binding affinity, as measured by an affinity constant, for such peptides that is 5-fold, 10-fold, 50-fold, 100-fold, 200-fold or more greater than the binding affinity of wild-type streptavidin for the same peptide.

[0169] A streptavidin mutein contains one or more amino acid differences compared to unmodified streptavidin, such as wild-type streptavidin or a fragment thereof. The term "unmodified streptavidin" refers to the starting polypeptide to which one or more modifications are made. In some embodiments, the starting or unmodified polypeptide can be the wild-type polypeptide set forth in SEQ ID NO:11. In some embodiments, unmodified streptavidin is a fragment of wild-type streptavidin that is truncated at the N-terminus and / or C-terminus. Such minimal streptavidins include any that begin at the N-terminus in the region of amino acid positions 10-16 of SEQ ID NO:11 and end at the C-terminus in the region of amino acid positions 133-142 of SEQ ID NO:11. In some embodiments, unmodified streptavidin has the amino acid sequence set forth in SEQ ID NO:14. In some embodiments, unmodified streptavidin as shown in SEQ ID NO:14 can further contain an N-terminal methionine at the position corresponding to Ala13 in the numbering shown in SEQ ID NO:11. References to the number of residues in streptavidin provided herein refer to the residue numbering in SEQ ID NO:11.

[0170] The terms "streptavidin mutein," "streptavidin mutant," or variations thereof, refer to a streptavidin protein containing one or more amino acid differences compared to unmodified or wild-type streptavidin, such as the streptavidin set forth in SEQ ID NO:11 or SEQ ID NO:14. The one or more amino acid differences can be amino acid mutations, such as one or more amino acid replacements (substitutions), insertions, or deletions. In some embodiments, a streptavidin mutein can have at least 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, or 20 amino acid differences compared to wild-type or unmodified streptavidin. In some embodiments, the amino acid replacements (substitutions) are conservative or non-conservative mutations. Streptavidin muteins containing one or more amino acid differences exhibited a 2.7×10 4 M -1 In some embodiments, the streptavidin variant exhibits a binding affinity as an affinity constant greater than 1.4×10 for the peptide ligand (Trp Ser His Pro Gln Phe Glu Lys; also known as Strep-tag® II and set forth in SEQ ID NO:6). 4 M -1 The binding affinity is expressed as the larger affinity constant. In some embodiments, binding affinity can be determined by methods known in the art, such as any of those described below.

[0171] In some embodiments, the streptavidin mutein contains mutations at one or more residues 44, 45, 46, and / or 47. In some embodiments, the streptavidin mutant contains residues Val44-Thr45-Ala46-Arg47 as shown in the exemplary streptavidin mutein set forth in SEQ ID NO: 12 or SEQ ID NO: 15. In some embodiments, the streptavidin mutein contains residues Ile44-Gly45-Ala-46-Arg47 as shown in the exemplary streptavidin mutein set forth in SEQ ID NO: 13 or 16. In some embodiments, the streptavidin mutein exhibits at least 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99% or more sequence identity to the sequence of amino acids set forth in SEQ ID NO: 12, 13, 15, 16, 20, 21, or 22, or the sequence of amino acids set forth in SEQ ID NO: 12, 13, 15, or 16, and has a sequence identity of 2.7×10 to the peptide ligand (Trp Arg His Pro Gln Phe Gly Gly; also known as Strep-tag® and set forth in SEQ ID NO: 5). 4 M -1 1.4×10 for larger and / or peptide ligands (Trp Ser His Pro Gln Phe Glu Lys; also known as Strep-tag® II and shown in SEQ ID NO:6) 4 M -1 The amino acid sequences that exhibit greater binding affinity are shown.

[0172] In some embodiments, the streptavidin mutein is a variant as described in International Published PCT Application No. WO2014 / 076277. In some embodiments, the streptavidin mutein contains at least two cysteine ​​residues in the region of amino acid positions 44-53 relative to the amino acid positions set forth in SEQ ID NO:11. In some embodiments, cysteine ​​residues are present at positions 45 and 52, forming a disulfide bridge connecting these amino acids. In such embodiments, amino acid 44 is typically glycine or alanine, amino acid 46 is typically alanine or glycine, and amino acid 47 is typically arginine. In some embodiments, the streptavidin mutein contains at least one mutation or amino acid difference in the region of amino acid residues 115-121 relative to the amino acid positions set forth in SEQ ID NO:11. In some embodiments, the streptavidin mutein contains at least one mutation at amino acid positions 117, 120, and 121 and / or a deletion of amino acids 118 and 119 and a substitution at least at amino acid position 121.

[0173] In some embodiments, the resulting streptavidin mutein binds to the peptide ligand (Trp Arg His Pro Gln Phe Gly Gly; also known as Strep-tag® and set forth in SEQ ID NO:5) at 2.7×10 4 M -1 and / or 1.4 × 10 for peptide ligands (Trp Ser His Pro Gln Phe Glu Lys; also known as Strep-tag® II and shown in SEQ ID NO:6). 4 M -1 The streptavidin mutein can contain any of the above mutations in any combination, so long as it exhibits a binding affinity of greater than 1.

[0174] In some embodiments, the binding affinity of the streptavidin mutant for the peptide ligand binding reagent is 5×104 M -1 , 1×10 5 M -1 , 5×10 5 M -1 , 1×10 6 M -1 , 5×10 6 M -1 , or 1 × 10 7 M -1 Larger, but generally, 1 x 10 13 M -1 , 1×10 12 M -1 , or 1 × 10 11 M -1 Smaller than.

[0175] In some embodiments, the streptavidin mutants also exhibit binding to other streptavidin ligands, such as, but not limited to, biotin, iminobiotin, lipoic acid, desthiobiotin, diaminobiotin, HABA (hydroxyazobenzene-benzoic acid), or / and dimethyl-HABA. In some embodiments, the streptavidin mutein exhibits binding affinity for another streptavidin ligand, such as biotin or desthiobiotin, that is greater than the binding affinity of the streptavidin mutein for the peptide ligand (Trp Arg His Pro Gln Phe Gly; also known as Strep-tag® and set forth in SEQ ID NO:5) or the peptide ligand (Trp Ser His Pro Gln Phe Glu Lys; also known as Strep-tag® II and set forth in SEQ ID NO:6).

[0176] In some embodiments, the streptavidin mutein is a multimer. Multimers can be produced using any method known in the art, such as those described in U.S. Patent Application Publication No. US2004 / 0082012. In some embodiments, mutein oligomers or polymers can be prepared by introducing carboxyl residues into polysaccharides, such as dextran. In some aspects, the streptavidin mutein is then conjugated to the carboxyl groups of the dextran backbone via the primary amino groups of internal lysine residues and / or the free N-terminus using conventional carbodiimide chemistry in a second step. In some embodiments, the conjugation reaction is carried out at a molar ratio of about 60 moles of streptavidin mutant per mole of dextran. In some embodiments, oligomers or polymers can also be obtained by cross-linking via a bifunctional linker, such as glutaric dialdehyde, or by other methods known in the art.

[0177] In the context of a chromatography matrix, a matrix such as agarose beads or other matrix is ​​functionalized or conjugated with a multimerization reagent such as a streptavidin mutein such as any of those described above, e.g., any of those set forth in SEQ ID NOs: 12, 13, 15, 16, 20, 21, or 22. In some embodiments, an antibody or antigen-binding fragment such as a Fab is fused or linked, directly or indirectly, to a peptide ligand capable of binding to a streptavidin mutant such as any of those described above. In some embodiments, the peptide ligand is any of those described above, such as a peptide containing a sequence of amino acids set forth in any of SEQ ID NOs: 1-10 or 17-19. In some embodiments, the chromatography matrix column is contacted with a binding agent to immobilize or reversibly bind such binding agent to the column.

[0178] In some embodiments, immunoaffinity chromatography matrices can be used in the enrichment and selection methods described herein by contacting the matrix with a sample containing the cells to be enriched or selected (e.g., a mixed population). In some embodiments, selected cells are eluted or released from the matrix by disrupting the binding partner / multimerization reagent interaction. In some embodiments, the binding partner / multimerization reagent is mediated by the interaction of a peptide ligand with a streptavidin mutant, and release of selected cells can occur due to the presence of reversible binding. For example, in some embodiments, the binding between the peptide ligand binding partner and the streptavidin mutein binding reagent is high, as described above, but lower than the binding affinity of the streptavidin binding reagent for biotin or a biotin analog. Thus, in some embodiments, biotin (vitamin H) or a biotin analog can be added to compete for binding and disrupt the binding interaction between the streptavidin mutein binding reagent on the matrix and the peptide ligand binding partner associated with an antibody specifically bound to a cell marker on the surface. In some embodiments, the interaction can be reversed in the presence of low concentrations of biotin or an analog, e.g., 0.1 mM to 10 mM, 0.5 mM to 5 mM, or 1 mM to 3 mM, e.g., generally at least 1 mM or at least about 1 mM or at least 2 mM, e.g., 2.5 mM or about 2.5 mM. In some embodiments, elution in the presence of a competing reagent, such as biotin or a biotin analog, releases selected cells from the matrix. In some embodiments, the competing reagent is biotin or a biotin analog.

[0179] In some embodiments, the immunoaffinity chromatography in the provided methods is performed using at least two operably connected chromatography matrix columns. For example, a binding agent, e.g., an antibody, e.g., a Fab, comprising a monovalent binding site capable of binding to either CD4 or CD8 is bound to a first chromatography matrix in a first selection column, and a binding agent, e.g., an antibody, e.g., a Fab, comprising a monovalent binding site capable of binding to the other of CD4 or CD8 is bound to a second chromatography matrix in a second selection column. In some embodiments, the at least two chromatography matrix columns are present in a closed system or device, such as a sterile closed system or device.

[0180] Also provided herein, in some embodiments, is a closed system or device containing at least two operatively connected chromatography matrix columns.

[0181] In some embodiments, the closed system is automated. In some embodiments, components associated with the system can include an on-board microcomputer, a peristaltic pump, and various valves, such as pinch valves or stopcocks, to control fluid flow between various parts of the system. The on-board computer, in some aspects, controls all components of the instrument and directs the system to perform repetitive procedures in a standardized sequence. In some embodiments, the peristaltic pump controls the flow rate throughout the tubing set and, together with the pinch valves, ensures a controlled flow of buffer through the system.

[0182] The wash buffer can be any physiological buffer compatible with cells, such as phosphate-buffered saline. In some embodiments, the wash buffer contains bovine serum albumin, human serum albumin, or recombinant human serum albumin, e.g., at a concentration of 0.1% to 5% or 0.2% to 1%, e.g., about 0.5%. In some embodiments, the eluent is biotin or a biotin analog, e.g., desbiotin, in an amount of, e.g., at least 0.5 mM, 1 mM, 1.5 mM, 2 mM, 2.5 mM, 3 mM, 4 mM, or 5 mM.

[0183] In some embodiments, the separation of cells from a mixed population to generate a cellular dataset according to the methods described herein is performed sequentially. For example, a first positive selection can be performed in a first column, and a second selection of either the positive or negative fraction from the first selection can be performed in a second column. In some embodiments, the separation of cells from a mixed population to generate a cellular dataset according to the methods described herein is performed in parallel. For example, two positive selections can be performed simultaneously in two different columns. In some embodiments, parallel selection (e.g., separation, isolation) can be achieved by applying the mixed population to both columns simultaneously or nearly simultaneously. It is believed that any number of separation steps can be used to generate a cellular dataset for training a machine learning model.

[0184] II. Methods for generating engineered T cells In some embodiments, the cell classification methods provided herein can be used in connection with processes that include creating, generating, or producing cell therapies. In some embodiments, the cell therapies include cells, such as T cells, engineered with a recombinant receptor, such as a chimeric antigen receptor (CAR), e.g., CAR T cells. In some embodiments, the cell classification methods provided herein are used in connection with creating, generating, or producing cell therapies, and the cell therapies can be performed via a process that includes one or more processing steps, such as steps for cell isolation, separation, selection, activation or stimulation, transduction, incubation, culture, expansion, washing, suspending, diluting, concentration, and / or formulation. In some embodiments, the process does not include a step for expansion. In some embodiments, the process does not include a step for culture. For example, in some embodiments, the process does not include the steps described in Section II-D below. In some embodiments, the process includes one or more processing steps, such as steps for cell isolation, separation, selection, activation or stimulation, transduction, incubation, washing, suspending, diluting, concentration, and / or formulation. In some embodiments, a manufacturing process that does not include an expansion step is referred to as a non-expansion process or a minimal expansion process. A "non-expansion expansion" process is sometimes referred to as a "minimal expansion" process. In some embodiments, a non-expansion expansion process or a minimal expansion expansion process may result in cells that have undergone expansion even though the process does not include a step for expansion. In some embodiments, the harvested cells may have undergone an incubation or culture step that includes a media composition designed to reduce, inhibit, minimize, or eliminate expansion of the cell population as a whole. In some embodiments, the manufacturing process is a non-expansion expansion process or a minimal expansion expansion process.

[0185] In some embodiments, the method for producing or generating cell therapy includes isolating cells from a subject, preparing them under one or more stimulating conditions, treating, and culturing them. In some embodiments, the method includes the following processing steps performed in the following order: first, cells, such as primary cells, are isolated from a biological sample, for example, selected or separated; optionally, the isolated cells are stimulated in the presence of a stimulating reagent, followed by incubating the selected cells with viral vector particles for transduction; culturing the transduced cells, for example, to allow the cells to proliferate or not, or incubating the transduced cells, for example, in a non-expansion or minimal expansion process; and formulating the transduced cells into a composition. In some embodiments, the generated engineered cells are reintroduced into the same subject before or after cryopreservation. In some embodiments, the cells during one or more steps, including before and / or after isolation, selection, transduction, and / or culturing, can be cryopreserved and then thawed.

[0186] In some embodiments, the one or more processing steps can include one or more of: (a) washing a biological sample containing cells (e.g., a whole blood sample, a buffy coat sample, a peripheral blood mononuclear cell (PBMC) sample, an unfractionated T cell sample, a lymphocyte sample, a leukocyte sample, a blood purification therapy product, or a leukapheresis therapy product); (b) isolating, e.g., selecting, a desired subset or population of cells (e.g., CD4+ and / or CD8+ T cells) from the sample, e.g., by incubating the cells with selection or immunoaffinity reagents for immunoaffinity-based separation; (c) incubating isolated cells, such as selected cells, with viral vector particles; (d) culturing, cultivating, incubating, or optionally expanding the cells using the described methods; and (e) formulating the transduced cells, e.g., in a pharmaceutically acceptable buffer, cryopreservation medium, or other appropriate medium. In some embodiments, the method can further include (e) stimulating the cells by exposing them to a stimulatory condition, which can be performed before, during, and / or after incubation of the cells with the viral vector particles. In some embodiments, one or more additional steps of washing or suspension steps for cell dilution, concentration, and / or buffer exchange, etc., can also be performed before or after any of the above steps.

[0187] In some embodiments, the provided methods are performed such that one, several, or all steps in preparing cells for clinical use, for example, in adoptive cell therapy, are performed without exposing the cells to non-sterile conditions and without the need to use a sterile room or cabinet. In some embodiments of such processes, cells are isolated, separated, or selected, transduced, washed, optionally activated or stimulated, and prepared, all within a closed system. In some embodiments, the methods are performed in an automated manner. In some embodiments, one or more of the steps are performed separately from a closed system or device.

[0188] In some embodiments, the closed system is used to perform one or more of the other processing steps of the method for creating, producing, or producing a cell therapy. In some embodiments, one or more or all of the processing steps, such as isolation, selection and / or enrichment, processing, transduction, and incubation related to manipulation, and formulation steps, are performed using an integrated or self-contained system and / or an automated or programmable system, device, or instrument. In some aspects, the system or instrument includes a computer and / or computer program communicating with the system or instrument, allowing the user to program, control, evaluate, and / or adjust various aspects of the processing, isolation, manipulation, and formulation steps. In one example, the system is a system such as that described in International Publication No. WO2009 / 072003 or U.S. Patent Application Publication No. US20110003380A1. In one example, the system is a system such as that described in International Publication No. WO2016 / 073602. In some embodiments, the cell classification methods described herein are integrated with systems or devices such that the results of the classification process can be used to control and / or adjust various aspects of the processing, isolation, manipulation, and formulation processes. For example, the cell classification methods described herein can be used to determine cell health (alive, dead), cell type, and / or recombinant molecule-expressing cells, e.g., to determine (e.g., predict) the percentage, total number, and / or concentration of viable cells, cell types (e.g., CD4+, CD8+), and / or recombinant molecule-positive cells (e.g., CAR+, CAR-). This information can be used to inform processing, isolation, manipulation, formulation, and / or harvesting processes. The output of the classification process can be used by an operator (e.g., a human operator) to program, control, evaluate the results of, and / or adjust various aspects of the processing, isolation, manipulation, harvesting, and formulation processes, and can interface directly with systems or devices, or a combination of both.

[0189] In some embodiments, the methods of classifying cells provided herein are performed at any time during the manufacturing process. In some embodiments, the methods of classifying cells provided herein are performed at any step of the manufacturing process. In some embodiments, the methods of classifying cells provided herein are performed during one or more steps of the manufacturing process. In some embodiments, the classification methods provided herein are performed during a specific step of the manufacturing process, such as a step of the manufacturing process described herein. In some embodiments, the classification methods provided herein are performed during an incubation, culture, or expansion step of the manufacturing process, such as an incubation, culture, or expansion step described herein. In some embodiments, the classification methods provided herein are performed during an incubation step.

[0190] In some embodiments, the cell classification methods provided herein are used to guide the stimulation, operation, incubation, and culture conditions of a manufacturing process. In some embodiments, the classification methods provided herein are useful for determining the need to change stimulation, operation, incubation, and culture conditions. For example, determining cell health, identifying cell types, and identifying viable concentrations may be useful for determining perfusion and / or feeding schedules, the need to reduce or add cytokines during stimulation conditions, detecting the concentration of activated or differentiated T cells, determining whether a mixed population contains an appropriate concentration of viable cells and / or viable cell subtypes to be transduced or transfected, determining the concentration of successfully transduced cells, and / or determining the concentration of viable cells, viable transduced / transfected cells, and viable transduced / transfected cell subtypes for harvesting decisions. As described above, the cell classification methods described herein can output information that can be used by an operator (e.g., a human operator), a computer, and / or a system that can receive or interface with the output of the classification process.

[0191] In some embodiments, the cell sorting methods provided herein are performed in a closed system as described herein. In some embodiments, the cell sorting methods provided herein are performed in a sterile configuration. For example, an imaging device (e.g., a DDHM microscope) can be sterilely connected to a chamber in which a cell population is held. In some embodiments, the sterile connection is a closed-loop system between the imaging device (e.g., a microscope) and the chamber, where the cells of the population are transported to the imaging device for imaging and then returned to the chamber. In some embodiments, the chamber is an incubation chamber. In some embodiments, the chamber is a bioreactor. In some embodiments, the transfer of cells from the chamber to the imaging device (e.g., a DDHM microscope) and back to the chamber is automated. For example, the closed-loop connection can use a pump, allowing for damage-free circulation of cells from the chamber to the imaging device (e.g., a DDHM microscope) and back to the chamber at set intervals. In some embodiments, the transfer of cells from the chamber to the imaging device (e.g., a DDHM microscope) and back to the chamber involves an operator (e.g., a human operator). For example, an operator (e.g., a human operator) can determine the time interval for imaging cells in a closed-loop system. In some embodiments, the operator (e.g., a human operator) can image cells in a closed-loop system at one or more discrete time points, e.g., at the discretion of the operator.

[0192] In some embodiments, the cell sorting methods provided herein support a fully automated manufacturing process, e.g., without requiring manual (e.g., human operator) input. In some embodiments, the fully automated system is a closed-loop system. In some embodiments, the fully automated closed-loop system efficiently and reliably produces sterile, viable populations of engineered cells, including specific cell types (e.g., CD4+, CD8+, CAR+, TCR+), at specific ratios or total cell numbers for cell therapy (e.g., adoptive cell therapy, autologous cell therapy).

[0193] Sorting can be performed to ascertain (e.g., measure, quantify) cell health (cell viability, cell death), cell subtype (CD4+, CD8+ T cells), and / or operational status (e.g., whether T cells express a recombinant receptor (e.g., chimeric antigen receptor (CAR), T cell receptor (TCR)). In some embodiments, sorting is performed to determine the total number, percentage, and / or concentration of viable cells, the total number, percentage, and / or concentration of cell subtypes (e.g., CD4+, CD8+), and / or the total number, percentage, and / or concentration of cells expressing a recombinant receptor. In some embodiments, the methods of sorting described herein are incorporated into an automated system. For example, the methods of sorting cells described herein can be performed in a culture (e.g., incubation) such as a bioreactor containing cells. The classification method may be stored on and / or executed on a computer interfaced with an imaging device (e.g., a microscope) connected to the cell culture chamber. As a further example, the computer containing the classification method may interface with or otherwise communicate with an instrument or device used to control a particular process. In this manner, the output of the classification method may be used to control the instrument or device. For example, the output of the classification method may be used to determine feeding cycles, media composition (e.g., cytokine concentrations), and / or harvest times. Automated systems may require minimal or no manual input (e.g., by a human operator) to classify cultured cells. In some embodiments, manual input (e.g., by a human operator) is required.

[0194] In some embodiments, the automated system is adapted to a bioreactor, such as a bioreactor described herein, such that cells can be removed from the bioreactor, imaged (e.g., for the purpose of obtaining image data and / or input features as described above), and then returned to the bioreactor. In some embodiments, the sorting and culturing are performed in a closed-loop configuration. In some aspects, in a closed-loop configuration, the automated system and bioreactor are kept sterile. In embodiments, the automated system is sterile. In some embodiments, the automated system is an in-line system.

[0195] In some embodiments, the automated system comprises an imaging device suitable for imaging cells in liquid suspension. Any imaging technique suitable for imaging cells in suspension is contemplated. Non-limiting examples of useful imaging techniques include bright-field microscopy, fluorescence microscopy, differential interference contrast microscopy, phase contrast microscopy, digital holography microscopy (DHM), differential digital holography microscopy (DDHM), or a combination thereof. In certain embodiments, the automated system comprises a differential digital holography microscope. In certain embodiments, the automated system comprises a differential digital holography microscope that includes an illumination means (e.g., laser, LED). Descriptions of DDHM methods and uses can be found, for example, in US 7,362,449, EP 1,631,788, US 9,904,248, and US 9,684,281, which are incorporated herein by reference in their entirety, and in Section IA-1 above.

[0196] In some embodiments, the automated system includes a digital recording device for recording the output of the imaging technique (e.g., DDHM). In some embodiments, the device is a CCD camera or a CMOS camera. In some embodiments, the automated system includes a computer comprising an algorithm for analyzing the images. In some embodiments, the automated system includes a computer comprising an algorithm for extracting image data (e.g., phase, intensity, overlay, as described in Section IB above) from the recorded images. In some embodiments, the automated system includes a monitor and / or computer for displaying the images. In some embodiments, the automated system includes a computer comprising an algorithm for extracting input features, as described in Section IB above, from the recorded images and / or image data. In some embodiments, the analysis is automated (i.e., can be performed without user input). In some embodiments, the automated system includes a computer comprising the machine learning models described herein that can receive image data and / or input features associated with the imaged cells and perform classification. Examples of suitable automated systems for monitoring cells include, but are not limited to, the Ovizio iLine F (Ovizio Imaging Systems NV / SA, Brussels, Belgium).

[0197] In certain embodiments, the classification method is performed continuously. In some embodiments, the classification method is performed in real time. In some embodiments, the classification method is performed at discrete time points. In some embodiments, the classification method is performed at least every 15 minutes during the culturing process. In some embodiments, the classification method is performed at least every 30 minutes during the culturing process. In some embodiments, the classification method is performed at least every 45 minutes during the culturing process. In some embodiments, the classification method is performed at least every hour during the culturing process. In some embodiments, the classification method is performed at least every 2 hours during the culturing process. In some embodiments, the classification method is performed at least every 4 hours during the culturing process. In some embodiments, the classification method is performed at least every 6 hours during the culturing process. In some embodiments, the classification method is performed at least every 8 hours during the culturing process. In some embodiments, the classification method is performed at least every 10 hours during the culturing process. In some embodiments, the classification method is performed at least every 12 hours during the culturing process. In some embodiments, the classification method is performed at least every 14 hours during the culturing process. In some embodiments, the classification method is performed at least every 16 hours during the culturing process. In some embodiments, the classification method is performed at least every 18 hours during the culturing process. In some embodiments, the classification method is performed at least every 20 hours during the culturing process. In some embodiments, the classification method is performed at least every 22 hours during the culturing process. In some embodiments, the classification method is performed at least once a day during the culturing process. In some embodiments, the classification method is performed at least once every two days during the culturing process. In some embodiments, the classification method is performed at least once every three days during the culturing process. In some embodiments, the classification method is performed at least once every four days during the culturing process. In some embodiments, the classification method is performed at least once every five days during the culturing process.In some embodiments, the classification method is performed at least once every 6 days during the culturing process. In some embodiments, the classification method is performed at least once every 7 days during the culturing process. In some embodiments, the classification method is performed at least once every 8 days during the culturing process. In some embodiments, the classification method is performed at least once every 9 days during the culturing process. In some embodiments, the classification method is performed at least once every 10 days during the culturing process. In some embodiments, the classification method is performed at least once during the culturing process.

[0198] In some embodiments, for example, in processes including an expansion step, cells are monitored by an automated system including a classification method until a proliferation threshold is reached. In some embodiments, once the proliferation threshold is reached, the cells are harvested, for example, by an automated method or a manual method, for example, by a human operator. The proliferation threshold can depend on the total concentration, density, and / or number of cultured cells determined by the automated system. Alternatively, the proliferation threshold can depend on the concentration, density, and / or number of viable cells.

[0199] A. Cell Isolation and Selection In some embodiments, the processing step involves isolating cells or compositions thereof from a biological sample, such as one obtained or derived from a subject, such as a subject with a particular disease or condition, or a subject in need of cell therapy, or a subject to whom cell therapy will be administered. In some aspects, the subject is human, such as a patient in need of a particular therapeutic intervention, such as adoptive cell therapy, for which the cells have been isolated, processed, and / or manipulated. Thus, in some embodiments, the cells are primary cells, such as primary human cells. In some embodiments, the cells comprise CD4+ and CD8+ T cells. In some embodiments, the cells comprise CD4+ or CD8+ T cells. Samples include tissues, fluids, and other samples taken directly from a subject. Biological samples can be samples obtained directly from a biological source or processed samples. Biological samples include, but are not limited to, bodily fluids, such as blood, plasma, serum, cerebrospinal fluid, synovial fluid, urine, and sweat, tissue and organ samples, including processed samples from the following:

[0200] In some aspects, the sample is a blood or blood-derived sample, or is or is derived from a blood purification or leukapheresis product. Exemplary samples include whole blood, peripheral blood mononuclear cells (PBMCs), white blood cells, bone marrow, thymus, tissue biopsy, tumor, leukemia, lymphoma, lymph node, gut-associated lymphoid tissue, mucosa-associated lymphoid tissue, spleen, other lymphoid tissue, liver, lung, stomach, intestine, colon, kidney, pancreas, breast, bone, prostate, cervix, testis, ovary, tonsil, or other organ, and / or cells derived therefrom. Samples include samples from autologous and allogeneic sources in the context of cell therapy, e.g., adoptive cell therapy.

[0201] In some examples, cells from the subject's circulating blood are obtained, for example, by blood purification or leukapheresis. The sample, in some aspects, contains lymphocytes, including T cells, monocytes, granulocytes, B cells, other nucleated white blood cells, red blood cells, and / or platelets, and in some aspects, contains cells other than red blood cells and platelets.

[0202] In some embodiments, blood cells collected from a subject are washed, e.g., to remove the plasma fraction and place the cells in an appropriate buffer or medium for subsequent processing steps. In some embodiments, the cells are washed with phosphate-buffered saline (PBS). In some embodiments, the wash solution lacks calcium and / or magnesium and / or many or all divalent cations. In some aspects, the wash step is accomplished by a semi-automated "flow-through" centrifuge (e.g., Cobe 2991 cell processor, Baxter) according to the manufacturer's instructions. In some aspects, the wash step is accomplished by tangential flow filtration (TFF) according to the manufacturer's instructions. In some embodiments, the cells are washed after washing, e.g., with Ca. ++ / Mg ++ In certain embodiments, the components of the blood cell sample are removed and the cells are resuspended directly in culture medium.

[0203] In some embodiments, the preparation method includes a step for freezing, e.g., cryopreserving, the cells either before or after isolation, selection, and / or enrichment and / or incubation for transduction and manipulation, and / or after culturing and / or harvesting the manipulated cells. Exemplary methods for freezing, cryopreserving, or cryopreserving biological samples, such as T cells or T cell compositions, include those described in WO2018170188, which is incorporated by reference in its entirety. In some embodiments, the freezing and subsequent thawing step removes granulocytes and, to some extent, monocytes from the cell population. In some embodiments, the cells are suspended in a freezing solution, e.g., after a washing step to remove plasma and platelets. In some aspects, any of a variety of known freezing solutions and parameters can be used. In some embodiments, cells are frozen, e.g., cryopreserved or cryoprotected, in medium and / or solution having a final concentration of at or about 12.5%, 12.0%, 11.5%, 11.0%, 10.5%, 10.0%, 9.5%, 9.0%, 8.5%, 8.0%, 7.5%, 7.0%, 6.5%, 6.0%, 5.5%, or 5.0% DMSO, or between 1% and 15%, 6% and 12%, 5% and 10%, or 6% and 8% DMSO. In certain embodiments, cells are frozen, e.g., cryopreserved or cryoprotected, in a medium and / or solution having a final concentration of 5.0%, 4.5%, 4.0%, 3.5%, 3.0%, 2.5%, 2.0%, 1.5%, 1.25%, 1.0%, 0.75%, 0.5%, or 0.25% HSA, or 0.1% to 5%, 0.25% to 4%, 0.5% to 2%, or 1% to 2% HSA. One example includes using PBS containing 20% ​​DMSO and 8% human serum albumin (HSA), or other suitable cell freezing medium.This is then diluted 1:1 with culture medium to a final concentration of 10% DMSO and 4% HSA, respectively. The cells are then generally frozen to at or about -80°C at a rate of at or about 1°C per minute and stored in the vapor phase of a liquid nitrogen storage tank.

[0204] In some embodiments, the isolation of a cell or population involves one or more preparative and / or non-affinity-based cell separation steps. In some instances, cells are washed, centrifuged, and / or incubated in the presence of one or more reagents, for example, to remove undesirable components, enrich for desired components, or lyse or remove cells sensitive to a particular reagent. In some instances, cells are separated based on one or more characteristics, such as density, adhesive properties, size, sensitivity, and / or resistance to a particular component. In some embodiments, the method involves a density-based cell separation method, such as preparing white blood cells from peripheral blood by lysing red blood cells and centrifuging them through a Percoll or Ficoll gradient.

[0205] In some embodiments, at least a portion of the selection step involves incubation of cells with a selection reagent. For example, incubation with one or more selection reagents may be performed as part of a selection method using one or more selection reagents to select one or more different cell types based on the expression or presence of one or more specific molecules, such as surface markers, e.g., surface proteins, intracellular markers, or nucleic acids, within or on cells. In some embodiments, any known method using one or more selection reagents for separation based on such markers may be used. In some embodiments, the one or more selection reagents result in a separation based on affinity or immunoaffinity. For example, in some aspects, selection involves incubation with one or more reagents for separating cells and cell populations based on the cellular expression or expression level of one or more markers, typically by incubation with an antibody or binding partner that specifically binds to a cell surface marker, typically followed by a washing step and separation of cells that bind to the antibody or binding partner from cells that do not bind to the antibody or binding partner. In some embodiments, the selection and / or other aspects of the process are as described in International Publication No. WO / 2015 / 164675.

[0206] In some aspects of such processes, a volume of cells is mixed with a certain amount of a desired affinity-based selection reagent. Immunoaffinity-based selection can be performed using any system or method that results in favorable energy interactions between the cells being separated and a molecule that specifically binds to a marker on the cells, such as an antibody or other binding partner on a solid surface, e.g., a particle. In some embodiments, the method is performed using particles such as beads, e.g., magnetic beads, coated with a selection agent (e.g., an antibody) specific to a marker on the cells. The particles (e.g., beads) can be incubated or mixed with the cells in a container, such as a tube or bag, with shaking or mixing, at a certain cell density-to-particle (e.g., bead) ratio to help promote energetically favorable interactions. In other cases, the method involves cell selection, where all or part of the selection is performed within the internal cavity of a centrifuge chamber, e.g., under centrifugal rotation. In some embodiments, incubation of the cells with a selection reagent, such as an immunoaffinity-based selection reagent, is performed within the centrifuge chamber. In certain embodiments, the isolation or separation is carried out using a system, device, or instrument described in International Patent Application Publication No. WO2009 / 072003 or US20110003380A1. In one example, the system is a system such as that described in International Publication No. WO2016 / 073602.

[0207] In some embodiments, performing such a selection step, or portions thereof (e.g., incubation with antibody-coated particles, e.g., magnetic beads) within the cavity of a centrifuge chamber allows users to control certain parameters, such as the volume of various solutions, the addition of solutions during processing, and their timing, providing advantages over other available methods. For example, the ability to reduce the liquid volume within the cavity during incubation can increase the concentration of the particles (e.g., bead reagents) used for selection, and therefore the chemical potential of the solution, without affecting the total number of cells within the cavity. This, in turn, can enhance pairwise interactions between the cells being processed and the particles used for selection. In some embodiments, for example, when associated with the systems, circuits, and controls described herein, performing the incubation step within the chamber allows users to agitate the solution at desired time(s) during incubation, which can also improve interactions.

[0208] In some embodiments, at least part of the selection step is carried out in a centrifuge chamber, which includes incubating cells with a selection reagent.In some aspects of this process, a volume of cells is mixed with a selection reagent based on the desired affinity in a much smaller amount than is normally used when performing the same selection in a tube or container to select the same number of cells and / or the same volume of cells according to the manufacturer's instructions.In some embodiments, the amount of one or more selection reagents used is 5% or less, 10% or less, 15% or less, 20% or less, 25% or less, 50% or less, 60% or less, 70% or less, or 80% or less of the amount of one or more same selection reagents used for selecting cells in a tube or container-based incubation for the same number of cells and / or the same volume of cells according to the manufacturer's instructions.

[0209] In some embodiments, for selection, e.g., based on the immunoaffinity of cells, the cells are incubated in a composition containing a selection buffer along with a selection reagent, e.g., a molecule, e.g., an antibody, that specifically binds to a surface marker on the cells desired to be enriched and / or depleted but does not specifically bind to surface markers on other cells in the composition, which may optionally be bound to a scaffold such as a polymer or surface, e.g., beads, e.g., magnetic beads, e.g., magnetic beads bound to monoclonal antibodies specific for CD4 and CD8. In some embodiments, the selection reagent is added to the cells in the cavity of the chamber in a substantially smaller amount (e.g., 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, or 80% or less) than would normally be used or required to achieve approximately the same or similar efficiency of selection of the same number of cells or the same volume of cells when selection is performed in a tube with shaking or rotation, as described above. In some embodiments, the volume is, for example, 10 mL to 200 mL, e.g., at least 10 mL, 20 mL, 30 mL, 40 mL, 50 mL, 60 mL, 70 mL, 80 mL, 90 mL, 100 mL, 150 mL, or 200 mL, or at least about 10 mL, 20 mL, 30 mL, 40 mL, 50 mL, 60 mL, 70 mL, 80 mL, 90 mL, 100 mL, 150 mL, or 200 mL, or about 10 mL, 20 mL, A selection buffer is added to the cells and selection reagent, and incubation is performed to achieve a target volume of incubation of 10 mL, 30 mL, 40 mL, 50 mL, 60 mL, 70 mL, 80 mL, 90 mL, 100 mL, 150 mL, or 200 mL, or 10 mL, 20 mL, 30 mL, 40 mL, 50 mL, 60 mL, 70 mL, 80 mL, 90 mL, 100 mL, 150 mL, or 200 mL of reagent. In some embodiments, the selection buffer and selection reagent are premixed before addition to the cells. In some embodiments, the selection buffer and selection reagent are added to the cells separately.In some embodiments, the selection incubation is performed using conditions of periodic gentle mixing, which can help promote energetically favorable interactions, thereby allowing for the use of less total selection reagents while achieving high selection efficiency.

[0210] In some embodiments, the total duration of incubation with the selection reagent is between 5 minutes and 6 hours, or between about 5 minutes and about 6 hours, such as between 30 minutes and 3 hours, for example at least 30 minutes, 60 minutes, 120 minutes, or 180 minutes, or at least about 30 minutes, 60 minutes, 120 minutes, or 180 minutes.

[0211] In some embodiments, the force or speed is generally relatively low, e.g., a speed lower than that used to pellet the cells, e.g., 600 rpm to 1700 rpm, or about 600 rpm to about 1700 rpm (e.g., 600 rpm, 1000 rpm, or 1500 rpm, or 1700 rpm, or about 600 rpm, 1000 rpm, or 1500 rpm, or 1700 rpm, or at least 600 rpm, 1000 rpm, or Incubation is generally performed under mixing conditions, such as in the presence of rotation at a speed of 1500 rpm, 1700 rpm, or 1800 rpm, e.g., at or about 80 g to 100 g (e.g., 80 g, 85 g, 90 g, 95 g, or 100 g or about 80 g, 85 g, 90 g, 95 g, or 100 g or at least 80 g, 85 g, 90 g, 95 g, or 100 g) of sample or at a relative centrifugal force (RCF) at the wall of a chamber or other container. In some embodiments, rotation at such low speeds is performed using repeated intervals of rotation followed by rest periods, e.g., 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 seconds of rotation and / or rest, e.g., about 1 or 2 seconds of rotation followed by about 5, 6, 7, or 8 seconds of rest.

[0212] In some embodiments, such a process is carried out in a completely closed system with an integrated chamber. In some embodiments, this process (and in some aspects, one or more additional steps, such as a previous washing step for washing a cell-containing sample, such as a blood purification therapy sample) is carried out in an automated manner, so that cells, reagents, and other components are drawn into the chamber, pushed out of the chamber, and centrifuged at the appropriate time to complete the washing and binding steps in a single closed system using an automated program.

[0213] In some embodiments, after incubation and / or mixing of cells and one and / or multiple selection reagents, the incubated cells are subjected to separation to select cells based on the presence or absence of a specific reagent or reagents. In some embodiments, the separation is performed in the same closed system in which the cells were incubated with the selection reagent. In some embodiments, after incubation with the selection reagent, the incubated cells, including the cells bound by the selection reagent, are transferred to a system for immunoaffinity-based separation of cells. In some embodiments, the system for immunoaffinity-based separation is or contains a magnetic separation column.

[0214] Such separation steps can be based on positive selection, in which cells that bind to a reagent, such as an antibody or binding partner, are retained for further use, and / or negative selection, in which cells that do not bind to a reagent, such as an antibody or binding partner, are retained. In some instances, both fractions are retained for further use. In some aspects, negative selection can be particularly useful when antibodies that specifically identify cell types in a heterogeneous population are not available, so that separation is best performed based on markers expressed by cells other than the desired population.

[0215] In some embodiments, the process steps further comprise negative and / or positive selection of the incubated cells, such as by using a system or instrument capable of affinity-based selection. In some embodiments, isolation is achieved by enrichment of a particular cell population by positive selection, or depletion of a particular cell population by negative selection. In some embodiments, a specific gene (marker) expressed at or relatively high levels on positively or negatively selected cells, respectively, is isolated. high Positive or negative selection is achieved by incubating the cells with one or more antibodies or other binding agents that specifically bind to one or more surface markers expressed on the surface of the target cell (marker +). Multiple rounds of the same selection step, e.g., positive or negative selection steps, can be performed. In certain embodiments, the positively or negatively selected fraction is subjected to a selection process, such as by repeating the positive or negative selection step. In some embodiments, the selection is repeated two, three, four, five, six, seven, eight, nine, or more than nine times. In certain embodiments, the same selection is performed up to five times. In certain embodiments, the same selection step is performed three times.

[0216] Separation does not necessarily result in 100% enrichment or removal of a particular cell population or cell that expresses a particular marker.For example, positive selection or enrichment of a particular type of cell, such as a cell that expresses a marker, refers to increasing the number or proportion of such cells, but does not necessarily result in the complete absence of cells that do not express the marker.Similarly, negative selection, removal, or depletion of a particular type of cell, such as a cell that expresses a marker, refers to reducing the number or proportion of such cells, but does not necessarily result in the complete removal of all such cells.

[0217] In some examples, multiple separation steps are performed, in which the positively or negatively selected fraction from one step is subjected to another separation step, such as a subsequent positive or negative selection. In some examples, a single separation step can simultaneously deplete cells expressing multiple markers, such as by incubating cells with multiple antibodies or binding partners, each specific for a marker targeted for negative selection. Similarly, multiple cell types can be simultaneously positively selected by incubating cells with multiple antibodies or binding partners expressed on various cell types. In certain embodiments, one or more separation steps are repeated and / or performed more than once. In some embodiments, the positively or negatively selected fraction obtained from a separation step is subjected to the same separation step, such as by repeating the positive or negative selection step. In some embodiments, a single separation step is repeated and / or performed more than once, for example, to increase the yield of positively selected cells, increase the purity of negatively selected cells, and / or further remove positively selected cells from the negatively selected fraction. In certain embodiments, one or more separation steps are performed and / or repeated 2, 3, 4, 5, 6, 7, 8, 9, 10, or more than 10 times. In certain embodiments, one or more selection steps are performed and / or repeated 1 to 10, 1 to 5, or 3 to 5 times. In certain embodiments, one or more selection steps are repeated 3 times.

[0218] For example, in some aspects, specific subpopulations of T cells, such as cells expressing positive or high levels of one or more surface markers, e.g., CD28+, CD62L+, CCR7+, CD27+, CD127+, CD4+, CD8+, CD45RA+, and / or CD45RO+ T cells, are isolated by positive or negative selection techniques. In some embodiments, such cells are selected by incubation with one or more antibodies or binding partners that specifically bind to such markers. In some embodiments, the antibodies or binding partners can be conjugated, directly or indirectly, to a solid support or matrix for selection, such as magnetic or paramagnetic beads. For example, CD3+, CD28+ T cells can be positively selected using anti-CD3 / anti-CD28 conjugated magnetic beads (e.g., DYNABEADS® M-450 CD3 / CD28 T Cell Expander and / or ExpACT® beads).

[0219] In some embodiments, T cells are separated from PBMC samples by negative selection for markers expressed on non-T cells, such as B cells, monocytes, or other leukocytes, such as CD14. In some aspects, a CD4+ or CD8+ selection step is used to separate CD4+ helper T cells and CD8+ cytotoxic T cells. Such CD4+ and CD8+ populations can be further differentiated into subpopulations by positive or negative selection for markers expressed or relatively highly expressed on one or more naive, memory, and / or effector T cell subpopulations.

[0220] In some embodiments, CD8+ T cells are further enriched or depleted for naive, central memory, effector memory, and / or central memory stem cells, such as by positive or negative selection based on surface antigens associated with each subpopulation. In some embodiments, enrichment of central memory T (TCM) cells is performed to enhance efficacy, e.g., to improve long-term survival, expansion, and / or engraftment after administration, which in some aspects is particularly robust in such subpopulations. See Terakura et al., (2012) Blood. 1: 72-82; Wang et al. (2012) J Immunother. 35(9): 689-701. In some embodiments, efficacy is further enhanced by combining TCM-enriched CD8+ T cells with CD4+ T cells.

[0221] In some embodiments, memory T cells are present in both the CD62L+ and CD62L- subsets of CD8+ peripheral blood lymphocytes. PBMCs can be enriched or depleted for the CD62L-CD8+ and / or CD62L+CD8+ fractions, such as by using anti-CD8 and anti-CD62L antibodies.

[0222] In some embodiments, enrichment of central memory T (TCM) cells is based on positive or high surface expression of CD45RO, CD62L, CCR7, CD28, CD3, and / or CD127, and in some aspects on negative selection for cells expressing or highly expressing CD45RA and / or granzyme B. In some aspects, isolation of a CD8+ population enriched in TCM cells is performed by depletion of cells expressing CD4, CD14, CD45RA, and positive selection or enrichment of cells expressing CD62L. In one aspect, enrichment of central memory T (TCM) cells is performed starting from a negative fraction of cells selected on the basis of CD4 expression, and subjected to negative selection on the basis of CD14 and CD45RA expression, and positive selection on the basis of CD62L.

[0223] In some aspects, such selection is performed simultaneously, while in other aspects, it is performed sequentially in either order. In some aspects, both the positive and negative fractions from CD4-based separation are retained, and optionally after one or more additional positive or negative selection steps, the same CD4 expression-based selection step used to prepare the CD8+ T cell population or subpopulation is also used to generate the CD4+ T cell population or subpopulation for use in subsequent steps of the method. In some embodiments, the selection of the CD4+ T cell population and the selection of the CD8+ T cell population are performed simultaneously. In some embodiments, the selection of the CD4+ T cell population and the CD8+ T cell population are performed sequentially in either order. In some embodiments, methods for selecting cells can include those described in published U.S. Patent Application Publication No. US20170037369. In some embodiments, the selected CD4+ T cell population and the selected CD8+ T cell population can be combined after selection. In some aspects, the selected CD4+ T cell population and the selected CD8+ T cell population can be combined in a container or bag, such as a bioreactor bag. In some embodiments, the selected CD4+ T cell population and the selected CD8+ T cell population are processed separately, the selected CD4+ T cell population is enriched in CD4+ T cells, incubated with a stimulatory reagent (e.g., anti-CD3 / anti-CD28 magnetic beads), transduced with a viral vector encoding a recombinant protein (e.g., a CAR), and cultured under conditions that expand T cells, and the selected CD8+ T cell population is enriched in CD8+ T cells, incubated with a stimulatory reagent (e.g., anti-CD3 / anti-CD28 magnetic beads), transduced with a viral vector encoding a recombinant protein (e.g., a CAR) that is the same recombinant protein as used to engineer CD4+ T cells from the same donor, and cultured under conditions that expand T cells, such as according to the methods provided.

[0224] In certain embodiments, a biological sample, such as a sample of PBMCs or other white blood cells, is subjected to the selection of CD4+ T cells, which retain both negative and positive fractions. In certain embodiments, CD8+ T cells are selected from the negative fraction. In some embodiments, a biological sample is subjected to the selection of CD8+ T cells, which retain both negative and positive fractions. In certain embodiments, CD4+ T cells are selected from the negative fraction.

[0225] In a particular example, a sample of PBMCs or other white blood cell sample is subjected to selection of CD4+ T cells, retaining both the negative and positive fractions. The negative fraction is then subjected to negative selection based on expression of CD14 and CD45RA or CD19, and positive selection based on markers characteristic of central memory T cells, such as CD62L or CCR7, where positive and negative selection can be performed in either order.

[0226] CD4+ T helper cells can be sorted into naive cells, central memory cells, and effector cells by identifying cell populations with cell surface antigens. CD4+ lymphocytes can be obtained by standard methods. In some embodiments, naive CD4+ T lymphocytes are CD45RO-, CD45RA+, CD62L+, or CD4+ T cells. In some embodiments, central memory CD4+ T cells are CD62L+ and CD45RO+. In some embodiments, effector CD4+ T cells are CD62L- and CD45RO-.

[0227] In one example, to enrich CD4+ T cells by negative selection, a monoclonal antibody cocktail typically includes antibodies against CD14, CD20, CD11b, CD16, HLA-DR, and CD8. In some embodiments, the antibody or binding partner is bound to a solid support or matrix, such as a magnetic bead or a paramagnetic bead, to allow for cell separation for positive and / or negative selection. For example, in some embodiments, cells and cell populations are separated or isolated using immunomagnetic (or affinity magnetic) separation techniques (reviewed in Methods in Molecular Medicine, vol. 58: Metastasis Research Protocols, Vol. 2: Cell Behavior In Vitro and In Vivo, p17-25 Edited by: SA Brooks and U. Schumacher (Copyright) Humana Press Inc., Totowa, NJ).

[0228] In some aspects, the incubated sample or composition of cells to be separated is incubated with a selection reagent containing a small, magnetizable or magnetically responsive material, such as magnetically responsive particles or microparticles, such as paramagnetic beads (e.g., Dynalbeads or MACS® beads). The magnetically responsive material, e.g., particles, are generally attached, directly or indirectly, to a binding partner, e.g., an antibody, that specifically binds to a molecule, e.g., a surface marker, present on a cell, cells, or population of cells that one wishes to separate, e.g., negatively or positively select.

[0229] In some embodiments, magnetic particles or beads comprise magnetically responsive materials bound to specific binding members, such as antibodies or other binding partners.Many well-known magnetically responsive materials are known for use in magnetic separation methods, such as those described in U.S. Patent No. 4,452,773 to Molday and European Patent No. EP 452342 B, which are incorporated herein by reference.Colloidal-sized particles can also be used, such as those described in U.S. Patent No. 4,795,698 to Owen and U.S. Patent No. 5,200,084 to Liberti et al.

[0230] Incubation is generally carried out under conditions in which the antibody or binding partner bound to the magnetic particle or bead, or a molecule that specifically binds to such an antibody or binding partner, e.g., a secondary antibody or other reagent, specifically binds to the cell surface molecule, if present on cells in the sample.

[0231] In certain embodiments, magnetically responsive particles are coated with a primary antibody or other binding partner, a secondary antibody, a lectin, an enzyme, or streptavidin. In certain embodiments, magnetic particles are bound to cells via coating with a primary antibody specific for one or more markers. In certain embodiments, cells, rather than beads, are labeled with a primary antibody or binding partner, and then magnetic particles coated with a cell-type-specific secondary antibody or other binding partner (e.g., streptavidin) are added. In certain embodiments, streptavidin-coated magnetic particles are used in combination with biotinylated primary or secondary antibodies.

[0232] In some aspects, separation is achieved in a procedure in which a sample is placed in a magnetic field, and cells to which magnetically responsive or magnetizable particles are bound are attracted to the magnet and separated from unlabeled cells. In the case of positive selection, cells attracted to the magnet are retained; in the case of negative selection, unattracted cells (unlabeled cells) are retained. In some aspects, a combination of positive and negative selection is performed in the same selection step, in which case the positive and negative fractions are retained and further processed or subjected to additional separation steps.

[0233] In some embodiments, affinity-based selection is via magnetically activated cell sorting (MACS) (Miltenyi Biotec, Auburn, CA). Magnetically activated cell sorting (MACS), such as the CliniMACS system, allows for high-purity selection of cells bound to magnetized particles. In certain embodiments, MACS operates in a manner in which non-target and target species are sequentially eluted after application of an external magnetic field. That is, cells bound to magnetized particles are held in place, while unbound species are eluted. Then, after this first elution step is completed, species captured by the magnetic field and not eluted are released in some manner so that they can be eluted and recovered. In certain embodiments, non-target cells are labeled and depleted from a heterogeneous population of cells.

[0234] In some embodiments, the magnetically responsive particles remain attached to the cells to be subsequently incubated, cultured, and / or manipulated, and in some aspects, the particles remain attached to the cells for administration to a patient. In some embodiments, the magnetizable particles or magnetically responsive particles are removed from the cells. Methods for removing magnetizable particles from cells are known, and include, for example, the use of competitive unlabeled antibodies, magnetizable particles, or antibodies conjugated to cleavable linkers. In some embodiments, the magnetizable particles are biodegradable.

[0235] B. Cell Activation and Stimulation In some embodiments, the one or more treatment steps include stimulating isolated cells, such as a selected cell population. The incubation can occur prior to or in conjunction with genetic engineering, such as genetic engineering resulting from embodiments of the transduction methods described above. In some embodiments, the stimulation results in activation and / or proliferation of the cells, e.g., prior to transduction.

[0236] In some embodiments, the treatment step includes incubation of cells, such as selected cells, and the incubation step can include culture, cultivation, stimulation, activation, and / or propagation of the cells. In some embodiments, the composition or cells are incubated under stimulatory conditions or in the presence of a stimulatory agent. Such conditions include conditions designed to induce proliferation, activation, and / or survival of cells in a population to mimic antigen exposure and / or to prime cells for genetic manipulation, e.g., introduction of a recombinant antigen receptor.

[0237] In some embodiments, conditions for stimulation and / or activation can include one or more of a particular medium, temperature, oxygen content, carbon dioxide content, time, agents such as nutrients, amino acids, antibiotics, ions, and / or stimulatory factors such as cytokines, chemokines, antigens, binding partners, fusion proteins, recombinant soluble receptors, and any other agents designed to activate cells.

[0238] In some embodiments, the stimulatory conditions or agents include one or more agents, e.g., ligands, capable of stimulating or activating the intracellular signaling domain of the TCR complex. In some aspects, the agents activate or initiate the TCR / CD3 intracellular signaling cascade in T cells. Such agents may include agents suitable for delivering a primary signal, e.g., initiating activation of an ITAM-induced signal, such as an antibody specific for the TCR, e.g., an antibody such as anti-CD3. In some embodiments, the stimulatory conditions include one or more agents, e.g., ligands, capable of stimulating a costimulatory receptor, e.g., anti-CD28 or anti-4-1BB. In some embodiments, such agents and / or ligands may be bound to a solid support, such as beads, and / or one or more cytokines. Among the stimulatory agents are anti-CD3 / anti-CD28 beads (e.g., DYNABEADS® M-450 CD3 / CD28 T Cell Expander and / or ExpACT® beads). Optionally, the expansion method can further include adding anti-CD3 and / or anti-CD28 antibodies to the culture medium (e.g., at a concentration of at least about 0.5 ng / ml). In some embodiments, the stimulatory agent includes IL-2, IL-7, and / or IL-15, e.g., an IL-2 concentration of at least about 10 units / ml, at least about 50 units / ml, at least about 100 units / ml, or at least about 200 units / ml.

[0239] Conditions can include one or more of a particular medium, temperature, oxygen content, carbon dioxide content, time, agents such as nutrients, amino acids, antibiotics, ions, and / or stimulatory factors such as cytokines, chemokines, antigens, binding partners, fusion proteins, recombinant soluble receptors, and any other agents designed to activate cells.

[0240] In some aspects, the incubation is performed according to techniques such as those described in U.S. Pat. No. 6,040,177 to Riddell et al., Klebanoff et al. (2012) J Immunother. 35(9): 651-660, Terakura et al. (2012) Blood. 1: 72-82, and / or Wang et al. (2012) J Immunother. 35(9): 689-701.

[0241] In some embodiments, at least a portion of the incubation in the presence of one or more stimulating conditions or stimulating agents is performed in the internal cavity of the centrifuge chamber under centrifugal rotation, for example, as described in International Publication No. 2016 / 073602. In some embodiments, at least a portion of the incubation performed in the centrifuge chamber includes mixing with one or more reagents to induce stimulation and / or activation. In some embodiments, cells, such as selected cells, are mixed with the stimulating conditions or stimulating agents in the centrifuge chamber. In some aspects of such processes, a volume of cells is mixed with one or more stimulating conditions or stimulating agents in an amount that is much smaller than the amount typically used when performing similar stimulation in a cell culture plate or other system.

[0242] In some embodiments, the stimulus is added to the cells in the cavity of the chamber in a substantially lesser amount (e.g., 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, or 80% or less) than would normally be used or required to achieve about the same or similar efficiency of selection of the same number of cells or the same volume of cells when selection is performed in a centrifuge chamber, e.g., in a tube without mixing in a tube or bag, with periodic shaking or rotation. In some embodiments, for example, about 10 mL to about 200 mL, or about 20 mL to about 125 mL, e.g., at least 10 mL, 20 mL, 30 mL, 40 mL, 50 mL, 60 mL, 70 mL, 80 mL, 90 mL, 100 mL, 105 mL, 110 mL, 115 mL, 120 mL, 125 mL, 130 mL, 135 mL, 140 mL, 145 mL, 150 mL, 160 mL, 170 mL, 180 mL, 190 mL, or 200 mL, or at least about 10 mL, 20 mL, 30 mL, 40 mL, 50 mL, 60 mL, 70 mL, 80 mL, 90 mL, 100 mL, 105 mL, 110 mL, 115 mL, 120 mL, 125 mL, 130 mL, 135 mL, 140 mL, 145 mL, 150 mL, 160 mL, 170 mL, 180 mL, 190 mL, or 200 mL. The incubation buffer is added to the cells and stimuli to achieve a target incubation volume of 5 mL, 130 mL, 135 mL, 140 mL, 145 mL, 150 mL, 160 mL, 170 mL, 180 mL, 190 mL, or 200 mL, or about 10 mL, 20 mL, 30 mL, 40 mL, 50 mL, 60 mL, 70 mL, 80 mL, 90 mL, 100 mL, 105 mL, 110 mL, 115 mL, 120 mL, 125 mL, 130 mL, 135 mL, 140 mL, 145 mL, 150 mL, 160 mL, 170 mL, 180 mL, 190 mL, or 200 mL of reagent, and incubation is performed. In some embodiments, the incubation buffer and stimuli are premixed before being added to the cells. In some embodiments, the incubation buffer and stimuli are added separately to the cells.In some embodiments, the stimulation incubation is performed under periodic gentle mixing conditions that help promote energetically favorable interactions, thereby allowing for less overall stimulation while still achieving cell stimulation and activation.

[0243] In some embodiments, the force or speed is generally relatively low, e.g., a speed lower than that used to pellet the cells, e.g., 600 rpm to 1700 rpm, or about 600 rpm to about 1700 rpm (e.g., 600 rpm, 1000 rpm, or 1500 rpm, or 1700 rpm, or about 600 rpm, 1000 rpm, or 1500 rpm, or 1700 rpm, or at least 600 rpm, 1000 rpm, or Incubation is generally performed under mixing conditions, such as in the presence of rotation at a speed of 1500 rpm, 1700 rpm, or 1800 rpm, e.g., at or about 80 g to 100 g (e.g., 80 g, 85 g, 90 g, 95 g, or 100 g or about 80 g, 85 g, 90 g, 95 g, or 100 g or at least 80 g, 85 g, 90 g, 95 g, or 100 g) of sample or at a relative centrifugal force (RCF) at the wall of a chamber or other container. In some embodiments, rotation at such low speeds is performed using repeated intervals of rotation followed by rest periods, e.g., 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 seconds of rotation and / or rest, e.g., about 1 or 2 seconds of rotation followed by about 5, 6, 7, or 8 seconds of rest.

[0244] In some embodiments, for example, the total period of incubation with the stimulatory agent is between 1 hour and 96 hours, between 1 hour and 72 hours, between 1 hour and 48 hours, between 4 hours and 36 hours, between 8 hours and 30 hours, between 18 hours and 30 hours, or between 12 hours and 24 hours, or between about 1 hour and 96 hours, between 1 hour and 72 hours, between 1 hour and 48 hours, between 4 hours and 36 hours, between 8 hours and 30 hours, between 18 hours and 30 hours, or between about 12 hours and 24 hours, e.g., at least 6 hours, 12 hours, 18 hours, 24 hours, 36 hours, or 72 hours, or at least about 6 hours, 12 hours, 18 hours, 24 hours, 36 hours, or 72 hours, or about 6 hours, 12 hours, 18 hours, 24 hours, 36 hours, or 72 hours. In some embodiments, the further incubation is for a period of 1 hour to 48 hours, 4 hours to 36 hours, 8 hours to 30 hours, or 12 hours to 24 hours, or for a period of about 1 hour to 48 hours, about 4 hours to 36 hours, about 8 hours to 30 hours, or about 12 hours to 24 hours, inclusive.

[0245] C. Genetic Engineering In some embodiments, the treatment step involves the introduction of a nucleic acid molecule encoding a recombinant protein. Among such recombinant proteins are recombinant receptors, such as any of those described in Section III. Introduction of a nucleic acid molecule encoding a recombinant protein, such as a recombinant receptor, into cells can be carried out using any of several known vectors. Such vectors include viral and non-viral systems, including lentiviral and gammaretroviral systems, as well as transposon-based systems such as PiggyBac or Sleeping Beauty-based gene transfer systems. Exemplary methods include those for the introduction of a receptor-encoding nucleic acid, including by virus, e.g., retrovirus or lentivirus, transduction, transposon, and electroporation.

[0246] In certain embodiments, when culturing and / or expansion is included as a manufacturing step, the composition of cells is manipulated, e.g., transduced or transfected, prior to culturing the cells, e.g., under conditions that promote proliferation and / or expansion. In some embodiments, genetic manipulation of the cells, such as by transformation or transduction, occurs prior to incubating the cells, e.g., as described in Section II-C-3. In certain embodiments, the composition of cells is manipulated after the composition has been stimulated, activated, and / or incubated under stimulatory conditions. In certain embodiments, the composition is a stimulated composition. In certain embodiments, the stimulated composition has previously been cryopreserved and stored and is thawed prior to manipulation.

[0247] In some embodiments, gene transfer is achieved by first stimulating the cells, such as by combining them with a stimulus that induces a response such as proliferation, survival, and / or activation, as measured by, for example, expression of cytokines or activation markers, and then transducing the activated cells and expanding them in culture to numbers sufficient for clinical use.

[0248] In some embodiments, recombinant nucleic acids are introduced into cells using recombinant infectious viral particles, such as vectors derived from simian virus 40 (SV40), adenovirus, adeno-associated virus (AAV), and human immunodeficiency virus (HIV).

[0249] In some embodiments, the recombinant nucleic acid is introduced into the T cell via electroporation (see, e.g., Chicaybam et al. (2013) PLoS ONE 8(3): e60298 and Van Tedeloo et al. (2000) Gene Therapy 7(16): 1431-1437). In some embodiments, the recombinant nucleic acid is introduced into the T cell via transposition (see, e.g., Manuri et al. (2010) Hum Gene Ther 21(4): 427-437; Sharma et al. (2013) Molec Ther Nucl Acids 2, e74; and Huang et al. (2009) Methods Mol Biol 506: 115-126). Other methods for introducing and expressing genetic material into immune cells include calcium phosphate transfection (e.g., as described in Current Protocols in Molecular Biology, John Wiley & Sons, New York, NY), protoplast fusion, cationic liposome-mediated transfection; tungsten particle-promoted biolistics (Johnston, Nature, 346: 776-777 (1990)); and strontium phosphate DNA co-precipitation (Brash et al., Mol. Cell Biol., 7: 2031-2034 (1987)).

[0250] Other approaches and vectors for introducing nucleic acids encoding recombinant products are described, for example, in WO2014055668 and U.S. Pat. No. 7,446,190.

[0251] In some embodiments, cells, e.g., T cells, can be transfected with, e.g., a T cell receptor (TCR) or a chimeric antigen receptor (CAR) either during or after expansion. This transfection for the introduction of the gene of the desired receptor can be carried out, for example, using any suitable retroviral vector. The genetically modified cell population can then be released from the first stimulus (e.g., CD3 / CD28 stimulation) and subsequently stimulated with a second type of stimulus (e.g., via the newly introduced receptor). This second type of stimulus can include antigenic stimulation in the form of a peptide / MHC molecule, a cognate (cross-linking) ligand of the genetically introduced receptor (e.g., the natural ligand of the CAR), or any ligand (such as an antibody) that directly binds to the framework of the new receptor (e.g., by recognizing the constant region within the receptor). See, for example, Cheadle et al., "Chimeric antigen receptors for T-cell based therapy," Methods Mol Biol. 2012; 907: 645-66, or Barrett et al., Chimeric Antigen Receptor Therapy for Cancer Annual Review of Medicine Vol. 65: 333-347 (2014).

[0252] In some cases, vectors can be used that do not require cells, such as T cells, to be activated. In some such cases, cells can be selected and / or transduced prior to activation. Thus, cells can be manipulated prior to or after cell culture, and in some cases simultaneously with or during at least part of the culture.

[0253] In some aspects, cells are further engineered to promote the expression of cytokines or other factors.Additional nucleic acids, for example, genes for introduction, include those for improving therapeutic efficacy, for example, by promoting the survival and / or function of introduced cells; for example, genes for providing genetic markers for cell selection and / or evaluation, for evaluating survival or localization in vivo; for example, genes for improving safety, by making cells susceptible to negative selection in vivo, as described by Lupton SD et al., Mol. and Cell Biol., 11:6 (1991); and Riddell et al., Human Gene Therapy 3: 319-338 (1992); see also the publications PCT / US91 / 08442 and PCT / US94 / 05601 by Lupton et al., which describe the use of bifunctional selectable fusion genes obtained by fusing dominant positive selectable marker with negative selectable marker. See, for example, columns 14-17 of US Pat. No. 6,040,177 to Riddell et al.

[0254] In some embodiments, the introducing is performed by contacting one or more cells of the composition with a nucleic acid molecule encoding a recombinant protein, such as a recombinant receptor. In some embodiments, the contacting can be achieved using centrifugation, such as spinoculation (e.g., centrifugal seeding). Such methods include any of those described in International Publication No. 2016 / 073602. Exemplary centrifuge chambers include those manufactured and sold by Biosafe SA, such as those for use with the Sepax® and Sepax® 2 systems, including the A-200 / F and A-200 centrifuge chambers and various kits for use with such systems. Exemplary chambers, systems, and processing equipment and cabinets are described, for example, in U.S. Patent No. 6,123,655, U.S. Patent No. 6,733,433, U.S. Patent Application Publication No. 2008 / 0171951, and WO 00 / 38762, the entire contents of each of which are incorporated herein by reference. Exemplary kits for use with such systems include, but are not limited to, disposable kits sold by BioSafe SA under the product names CS-430.1, CS-490.1, CS-600.1, or CS-900.2.

[0255] In some embodiments, the system is included with and / or configured in association with other equipment, including equipment for operating, automating, controlling, and / or minimizing aspects of the transduction process performed within the system and one or more various other process steps, such as one or more process steps that can be performed with or in association with a centrifuge chamber system such as those described herein or in WO 2016 / 073602. In some embodiments, the equipment is housed within a cabinet. In some embodiments, the equipment comprises a cabinet containing a housing housing control circuitry, a centrifuge, a cover, a motor, a pump, sensors, a display, and a user interface. Exemplary devices are described in U.S. Patent No. 6,123,655, U.S. Patent No. 6,733,433, and U.S. Patent Application Publication No. 2008 / 0171951.

[0256] In some embodiments, the system comprises a series of containers, e.g., bags, tubing, stoppers, clamps, connectors, and a centrifuge chamber. In some embodiments, the containers, such as bags, include one or more containers, such as bags, containing cells to be transduced and viral vector particles in the same or separate containers, such as the same bag or separate bags. In some embodiments, the system further includes one or more containers, such as bags, containing media, such as diluents and / or wash solutions, that are drawn into the chamber, and / or other components for diluting, resuspending, and / or washing components and / or compositions during the method. The containers can be connected to one or more locations within the system, such as at locations corresponding to input lines, diluent lines, wash lines, waste lines, and / or output lines.

[0257] In some embodiments, the chamber is associated with a centrifuge that can rotate the chamber, such as around its rotation axis. Rotation can occur before, during, and / or after incubation in connection with cell transduction, and / or during one or more of the other processing steps. Thus, in some embodiments, one or more of the various processing steps are performed under rotation, for example, with a specific force. The chamber can typically rotate vertically or approximately vertically so that...

Claims

1. receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, the image data comprising one or more of phase image data, intensity image data, and overlay image data from each of the plurality of T cells, the image data being obtained using differential digital holographic microscopy (DDHM) without labeling the cells using a marker or dye; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and applying the image data from each cell of the plurality of T cells as input to a process for classifying each cell of the plurality of T cells as belonging to a first group or a second group, the process including a convolutional neural network trained on image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the image data for training the convolutional neural network including one or more of phase image data, intensity image data, and overlay image data from each cell of the T cells known to belong to the first group and T cells known to belong to the second group, wherein the first group is CD4+ T cells and the second group is CD8+ T cells. A method for classifying T cells, comprising:

2. receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, the image data comprising one or more of phase image data, intensity image data, and overlay image data from each of the plurality of T cells, the image data being obtained using differential digital holographic microscopy (DDHM) without labeling the cells using a marker or dye; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; determining the classification of each cell of the plurality of T cells as belonging to a first group or a second group, wherein the first group is CD4+ T cells and the second group is CD8+ T cells; and training a convolutional neural network based on the image data and the determined classification of each cell of the plurality of T cells. A method comprising:

3. receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, wherein the image data is obtained using differential digital holographic microscopy (DDHM) without labeling the cells using a marker or dye; determining the classification of each cell of said plurality of T cells as belonging to a first group or a second group, said first group being CD4+ T cells and said second group being CD8+ T cells; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; generating a feature map having one or more dimensions, each dimension of the one or more dimensions being associated with the morphological feature, the optical feature, the intensity feature, the topological feature, the system feature, or any combination thereof; and training a convolutional neural network based on the feature map and the determined classification of each cell of the plurality of T cells. A method comprising:

4. receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, wherein the image data is obtained using differential digital holographic microscopy (DDHM) without labeling the cells using a marker or dye; determining the classification of each cell of said plurality of T cells as belonging to a first group or a second group, said first group being CD4+ T cells and said second group being CD8+ T cells; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and training a neural network based on the one or more input features and the determined classification of each cell of the plurality of T cells. A method comprising:

5. receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, wherein the image data is obtained using differential digital holographic microscopy (DDHM) without labeling the cells using a marker or dye; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and applying the one or more input features for each cell of the plurality of T cells as input to a process to classify each cell of the plurality of T cells as belonging to a first group or a second group, the process comprising a neural network trained on one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof, wherein the first group is CD4+ T cells and the second group is CD8+ T cells. A method for classifying T cells, comprising:

6. The neural network (i) the one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group; and (ii) classifying each of the known T cells as belonging to the first group or the second group.

6. The method of claim 5, wherein the method is trained based on

7. receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, wherein the image data is obtained using differential digital holographic microscopy (DDHM) without labeling the cells using a marker or dye; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and applying the one or more input features for each cell of the plurality of T cells as input to a process to classify each cell of the plurality of T cells as belonging to a first group or a second group, the process comprising a support vector machine trained on one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof, wherein the first group is CD4+ T cells and the second group is CD8+ T cells. A method for classifying T cells, comprising:

8. receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, wherein the image data is obtained using differential digital holographic microscopy (DDHM) without labeling the cells using a marker or dye; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; determining the classification of each cell of the plurality of T cells as belonging to a first group or a second group, wherein the first group is CD4+ T cells and the second group is CD8+ T cells; and determining a random forest for a classification process, the random forest comprising one or more decision trees, the classification process relating one or more input features determined from the image data associated with each cell of the plurality of T cells to the classification of each cell of the plurality of T cells as belonging to the first group or as belonging to the second group; Including, a decision tree of the one or more decision trees is determined based on one or more input features for each cell of the plurality of T cells, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; method.

9. receiving image data associated with each of a plurality of T cells of a population of cells comprising T cells, wherein the image data is obtained using differential digital holographic microscopy (DDHM) without labeling the cells using a marker or dye; determining from the image data one or more input features for each cell of the plurality of T cells, wherein the one or more input features comprise morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof; and applying the one or more input features for each cell of the plurality of T cells as input to a process to classify each cell of the plurality of T cells as belonging to a first group or a second group, the process comprising a random forest classification process on one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group, the one or more input features comprising morphological features of the image data, optical features of the image data, intensity features of the image data, phase features of the image data, system features of the image data, or any combination thereof, wherein the first group is CD4+ T cells and the second group is CD8+ T cells. A method for classifying T cells, comprising:

10. The random forest classification process (i) the one or more input features determined from image data associated with T cells known to belong to the first group and T cells known to belong to the second group; and (ii) classifying each of the known T cells as belonging to the first group or the second group.

10. The method of claim 9, wherein the method is trained based on

11. 11. The method of claim 1, wherein the image data comprises phase image data and intensity image data for each cell of the plurality of T cells.

12. 12. The method of any one of claims 1 to 11, wherein said plurality of T cells comprises about, or at least 70% of, the T cells in said population of cells comprising T cells.

13. The one or more input features are an aspect ratio of the cell, a depth of the cell, an area of ​​the cell, a cell descriptor, a cell identifier, an image identifier, an object identifier, a centroid along the X axis, a centroid along the Y axis, a circularity of the cell, a compactness of the cell, a normalized aspect ratio, an elongation of the cell, a diameter of the cell, a peak diameter, a hu moment invariant 1, a hu moment invariant 2, a hu moment invariant 3, a hu moment invariant 4, a hu moment invariant 5, a hu moment invariant 6, a hu moment invariant 7, an average intensity contrast, an average entropy, an average Intensity, mean intensity uniformity, intensity contrast of the image of the cell, intensity correlation of the image of the cell, intensity entropy of the image of the cell, intensity homogeneity of the image of the cell, maximum intensity of the cell, mean intensity of the cell, minimum intensity of the cell, intensity skewness of the image of the cell, intensity smoothness of the image of the cell, intensity variance of the image of the cell, intensity uniformity of the image of the cell, plane at which the intensity of the cell is maximum, indication that a cell is located along a boundary of the field of view, indication that a refractive peak is located along a boundary of the field of view, mass eccentricity of the cell, maximum optical height of the cell in radians, maximum optical height of the cell in microns, average optical height of the cell in radians, average optical height of the cell in microns, normalized optical height of the cell, minimum optical height of the cell in radians, minimum optical height of the cell in microns, optical height variance of the phase of the image of the cell in radians, optical height variance of the phase of the image of the cell in microns, optical volume of the cell, area of ​​a refractive peak of the cell, area of ​​a refractive peak normalized by the area of ​​the cell, number of refractive peaks of the cell, intensity of a refractive peak of the cell 13. The method of any one of claims 3 to 12, wherein the phase image of the cells comprises one or more of: a normalized refractive peak height of the cells, a perimeter, a mean intensity contrast of a phase image of the cells, a mean entropy of a phase image of the cells, a mean phase of the cells, a mean phase uniformity of the cells, a phase intensity contrast of the cells, a phase correlation of the cells, a phase entropy feature of the cells, a phase homogeneity of the cells, a phase skewness of the cells, a phase smoothness of the cells, a phase uniformity of the cells, a mean radius of the cells, a variance of the radius of the cells, and a normalized radius variance of the cells.

14. 14. The method of any one of claims 3 to 13, wherein the one or more input features comprise morphological features of the image data selected from one or more of: aspect ratio of the cells, area of ​​the cells, circularity of the cells, compactness of the cells, normalized aspect ratio, elongation of the cells, diameter of the cells, Hu moment invariant 1, Hu moment invariant 2, Hu moment invariant 3, Hu moment invariant 4, Hu moment invariant 5, Hu moment invariant 6, Hu moment invariant 7, perimeter, mean radius of the cells, variance of radius of the cells, and normalized radius variance.

15. 15. The method of any one of claims 3 to 14, wherein the one or more input features comprise optical features of the image data selected from one or more of: a diameter of the cell, a maximum intensity of the cell, an average intensity of the cell, a minimum intensity of the cell, a mass eccentricity of the cell, a maximum optical height of the cell in radians, a maximum optical height of the cell in microns, an average optical height of the cell in radians, an average optical height of the cell in microns, a normalized optical height of the cell, a minimum optical height of the cell in radians, a minimum optical height of the cell in microns, an optical volume of the cell, an area of ​​a refractive peak of the cell, an area of ​​a refractive peak normalized by an area of ​​the cell, a number of refractive peaks of the cell, an intensity of a refractive peak of the cell, and a normalized refractive peak height of the cell.

16. 16. The method of any one of claims 3 to 15, wherein the one or more of the input features comprise intensity features of the image data selected from one or more of mean intensity contrast, mean entropy, mean intensity, mean intensity uniformity, intensity contrast of the image of the cell, intensity correlation of the image of the cell, intensity entropy of the image of the cell, intensity homogeneity of the image of the cell, intensity skewness of the image of the cell, intensity smoothness of the image of the cell, intensity variance of the image of the cell, plane at which the intensity of the cell is maximum, and intensity uniformity of the image of the cell.

17. 17. The method of any one of claims 3 to 16, wherein the one or more input features comprise phase features of the image data selected from one or more of: a variance of the phase optical height of the image of the cells in radians, a variance of the phase optical height of the image of the cells in microns, a mean intensity contrast of the phase image of the cells, a mean entropy of the phase image of the cells, a mean phase of the cells, a mean phase uniformity of the cells, a phase intensity contrast of the cells, a phase correlation of the cells, a phase entropy feature of the cells, a phase homogeneity of the cells, a phase skewness of the cells, a phase smoothness of the cells, and a phase uniformity of the cells.

18. 18. The method of any one of claims 3 to 17, wherein the one or more inputs comprise system features of the image data selected from one or more of: a cell depth, an identified cell, an image identifier, a cell descriptor, a centroid along the X axis, a centroid along the Y axis, an object identifier, an indication that a cell is located along a boundary of the field of view, and an indication that a refractive peak is located along a boundary of the field of view.

19. 19. The method of any one of claims 3-18, wherein the one or more input features comprise an average phase uniformity of the cells, the peak diameter, a normalized refractive peak height of the cells, an average phase of the cells, a refractive peak area normalized by an area of ​​the cells, a minimum optical height of the cells in microns, a compactness of the cells, a circularity of the cells, a phase smoothness of the cells, an intensity homogeneity of the image of the cells, a plane at which the intensity of the cells is maximum, an area of ​​the refractive peak of the cells, a phase correlation of the cells, a depth of the cells, an intensity contrast of the image of the cells, an intensity uniformity of the image of the cells, a normalized radial variance, an intensity smoothness of the image of the cells, a phase intensity contrast of the cells, a maximum optical height of the cells, an average intensity contrast of the cells, an average intensity uniformity of the cells, an intensity skewness of the image of the cells, a hu moment invariant 1, an intensity variance of the image of the cells, an average entropy, and an intensity correlation of the image of the cells.

20. 20. The method of any one of claims 3-19, wherein the one or more input features comprise depth of a cell, intensity contrast of the image of the cell, intensity uniformity of the image of the cell, normalized radial variance, intensity smoothness of the image of the cell, phase intensity contrast of the cell, maximum optical height of the cell in microns, mean intensity contrast of the cell, mean intensity uniformity of the cell, intensity skewness of the image of the cell, hu moment invariant 1, intensity variance of the image of the cell, mean entropy, and intensity correlation of the image of the cell.

21. 21. The method of any one of claims 3 to 20, wherein the one or more input features comprise a cell depth.

22. 20. The method of any one of claims 3 to 19, wherein the one or more input features comprise an average phase uniformity of the cell, a peak diameter, a normalized refractive peak height of the cell, an average phase of the cell, a refractive peak area normalized by the area of ​​the cell, a minimum optical height in microns of the cell, a compactness of the cell, a circularity of the cell, a phase smoothness of the cell, an intensity homogeneity of the image of the cell, a plane at which the intensity of the cell is maximum, an area of ​​the refractive peak of the cell, and a phase correlation of the cell.

23. 23. The method of any one of claims 1 to 22, wherein sorting T cells of said population of cells comprising T cells is performed in a closed system that is sterile and / or automated.

Images