AI models, systems, and methods for reducing manufacturing failures of car t drug products

Abstract

Systems and methods for predicting manufacturing failures in production of patient-specific CAR T drug products for a target patient are disclosed. An example method includes receiving quantitative data for a set of manufacturing failure parameters, generating an input feature vector comprising the quantitative data for the set of manufacturing failure parameters, and applying the input feature vector to a trained machine learning model to generate an output feature vector that predicts whether production of a patient-specific CAR T drug product will result in a manufacturing failure.

Description

Related applications

[0001] This application claims priority to U.S. Provisional Patent Application Serial Nos. 63 / 602,289, 63 / 602,359, 63 / 602,356, 63 / 602,356, 63 / 606,827, and 63 / 606,880, filed on November 22, 2023, the entire contents of which are incorporated herein by reference and are based on. Technical Field

[0002] This application relates to improved samples for the manufacture of CAR T drugs, methods for manufacturing CAR T drug products using the same, and methods for predicting and reducing manufacturing failures of CAR T drug products. Background Technology

[0003] Medical treatments via drug products containing CAR T cells (also referred to herein as "CAR T-drug therapy") utilize isolated T cells that have been genetically modified to enhance their specificity for specific tumor-associated antigens. These T cells are typically autologous T cells, isolated from a patient to receive CAR T-drug therapy. This isolation involves collecting the patient's blood and separating lymphocytes from the blood via apheresis. Genetic modification may involve the expression of chimeric antigen receptors (CARs) or exogenous T-cell receptors to provide novel antigen specificity to the T cells. T cells expressing chimeric antigen receptors (referred to herein as "CAR T cells" or "CAR+ T cells") can induce tumor immune reactivity. B-cell maturation antigen (BCMA) is a molecule expressed on the surface of mature B cells and malignant plasma cells and is a target molecule for treating cancers such as multiple myeloma. There is a need not only for better cancer therapies utilizing CAR T cells (specifically, CAR T cells specific to the BCMA tumor-associated antigen), but also for better methods to determine whether a particular apheresis product can be successfully manufactured into a CAR T-drug therapy suitable for treating a patient.

[0004] To produce CAR T-cell drug products for CAR T-drug therapy, single samples typically undergo a meticulous manufacturing process in which T cells in the single sample are activated, enriched, expanded, and transduced to express CAR+. Numerous controllable and uncontrollable variables govern the manufacturing process and ultimately influence the properties of CAR T-cell drug products. Because the production of CAR T-cell drug products is expensive and involves time and expertise, defective CAR T-cell drug products lead to wasted resources. Furthermore, delays in the production of CAR T-cell drug products required by patients can impact health outcomes. Therefore, there is an expectation and a need to better predict the properties of CAR T-cell drug products and optimize the manufacturing process.

[0005] For example, expectations and needs can better predict systems and methods for potential manufacturing failures during the production of CAR T drugs. Identifying aspects of the manufacturing process that may lead to the failure of CAR T drug products can help salvage the CAR T drug production process, recover production resources, and avoid delays in patient care.

[0006] Various embodiments of this disclosure address one or more of the disadvantages described above. Summary of the Invention

[0007] This disclosure describes various systems, methods, and improved manufacturing techniques for predicting and reducing manufacturing failures in CART drug products.

[0008] According to various implementation schemes, a method for predicting manufacturing failure of a patient-specific CAR T drug product for a target patient is also disclosed, the method comprising: receiving quantitative data of a set of manufacturing failure parameters, wherein the set of manufacturing failure parameters includes manufacturing failure parameters selected from Table 1, wherein each manufacturing failure parameter belongs to one of a plurality of parameter types as outlined in Table 1; generating an input feature vector including the quantitative data of the set of manufacturing failure parameters; and applying the input feature vector to a trained machine learning model to generate an output feature vector predicting whether the production of the patient-specific CAR T drug product will result in manufacturing failure.

[0009] In some aspects that can be combined with any other aspect of this disclosure, such as the manufacturing failure parameters outlined in Table 1, the manufacturing failure parameters are ordered in order of significance in predicting whether the production of a patient-specific CAR T drug product will result in manufacturing failure, wherein when a trained machine learning model is used to predict whether the production of a patient-specific CAR T drug product will result in manufacturing failure, manufacturing failure parameters with higher significance are assigned higher weights than other manufacturing failure parameters.

[0010] In some aspects that can be combined with any other aspect of this disclosure, the set of manufacturing failure parameters includes a set of screening parameters selected from Table 1A, wherein Table 1A consists of screening parameters from Table 1.

[0011] In some aspects that can be combined with any other aspect of this disclosure, the screening parameters in Table 1A are ordered in order of significance in predicting whether the production of a patient-specific CAR T drug product will lead to manufacturing failure, wherein screening parameters with higher significance are assigned higher weights than other screening parameters when using a trained machine learning model to predict whether the production of a patient-specific CAR T drug product will lead to manufacturing failure. In some aspects that can be combined with any other aspect of this disclosure, this set of manufacturing failure parameters includes a set of manufacturing stage parameters selected from Table 1B, wherein Table 1B consists of manufacturing stage parameters from Table 1.

[0012] In some aspects that can be combined with any other aspect of this disclosure, the manufacturing stage parameters in Table 1B are arranged in order of significance of whether the production of a patient-specific CAR T drug product will lead to manufacturing failure when using a trained machine learning model, wherein manufacturing stage parameters with higher significance are assigned higher weights than other manufacturing stage parameters when using a trained machine learning model to predict whether the production of a patient-specific CAR T drug product will lead to manufacturing failure.

[0013] In some aspects that can be combined with any other aspect of this disclosure, receiving quantitative data of the set of manufacturing failure parameters includes receiving unstructured data of the set of manufacturing failure parameters. In some embodiments, the method further includes vectorizing the unstructured target data into an input feature vector via a feature extraction module of a computing device.

[0014] In some aspects that can be combined with any other aspect of this disclosure, the trained machine learning model is trained using reference data from multiple reference CART drug products manufactured from multiple reference patients with known manufacturing failures.

[0015] In some aspects that can be combined with any other aspect of this disclosure, the method further includes: receiving reference data via a computing device, wherein the reference data includes a set of input feature parameters and known manufacturing failure results for each of a plurality of reference CAR T drug products manufactured from a plurality of reference patients, wherein for a given reference patient among the plurality of reference patients, the set of input feature parameters includes at least the set of manufacturing failure parameters; vectorizing the set of input feature parameters and known manufacturing failure results into reference input feature vectors and reference output feature vectors, respectively, for each of the plurality of reference CAR T drug products manufactured from the plurality of reference patients, via a feature extraction module of the computing device, thereby generating a plurality of reference input feature vectors and a plurality of reference output feature vectors; associating the plurality of reference input feature vectors with the plurality of reference output feature vectors in a machine learning model via a training module of the computing device; and training the machine learning model via the training module of the computing device by iteratively minimizing the error to within a predetermined threshold to generate a trained machine learning model, wherein the trained machine learning model includes a plurality of weights, each weight indicating the significance between the input feature parameters and the manufacturing failure results.

[0016] In some aspects that can be combined with any other aspect of this disclosure, the set of input characteristic parameters is outlined in Appendix A.

[0017] In some aspects that can be combined with any other aspect of this disclosure, for each of the multiple reference CAR-T drug products manufactured from corresponding multiple reference patients, the set of input characteristic parameters includes two or more of the following: the percentage of cells as CAR+ T cells in T cell culture samples harvested from intermediate to late stages of the manufacturing process of the reference CAR-T drug product; the concentration of lactate or glucose in T cell culture samples from late intermediate stages of the manufacturing process of the reference CAR-T drug product; the concentration of lactate or glucose in T cell culture samples from intermediate stages of the manufacturing process of the reference CAR-T drug product; the ratio of CD4+ T cells to CD8+ T cells in T cell culture samples from the initial stage of the manufacturing process of the reference CAR-T drug product; the volume of vector added to the T cell culture samples during early intermediate stages of the manufacturing process of the reference CAR-T drug product; the multiple of infection (MOI) of the vector added to the T cell culture samples during early intermediate stages of the manufacturing process of the reference CAR-T drug product; the mean percentage of live T cells in each population of T cell culture samples from early intermediate stages of the manufacturing process of the reference CAR-T drug product; and the percentage of cells as CAR+ T cells from the reference CAR-T drug product. The average concentration of live T cells in each population during the early intermediate stages of the manufacturing process of the T-drug product; the concentration of lymphocytes in a single sample prior to the manufacturing process of the reference CAR T-drug product; the percentage of cells as live T cells in a T-cell culture sample from the initial stage of the manufacturing process of the reference CAR T-drug product; whether the reference patient is refractory to pomalidomide treatment; the reference patient's sex; the reference patient's age; the reference patient's body mass index (BMI); or the reference patient's number of previous lines of treatment.

[0018] In some aspects that can be combined with any other aspect of this disclosure, the method further includes: determining that the production of a patient-specific CAR T drug product would result in a manufacturing failure; and adjusting one or more manufacturing process parameters for manufacturing the CAR T drug product for the target patient.

[0019] In some aspects that can be combined with any other aspect of this disclosure, the method further includes: determining that the production of a patient-specific CAR T drug product will not result in manufacturing failure; and initiating the production of a CAR T drug product for a target patient based on the set of manufacturing failure parameters.

[0020] In some aspects that can be combined with any other aspect of this disclosure, the group of manufacturing failure parameters includes the single-sample stage parameters in Table 1, which describe the concentration of lymphocytes in a single sample prior to the manufacturing process of a patient-specific CAR T drug product. In some aspects that can be combined with any other aspect of this disclosure, two or more manufacturing failure parameters include two or more of the following: the percentage of cells as CAR+ T cells in T cell culture samples harvested from intermediate to late stages of the manufacturing process of the CAR T drug product; the concentration of lactate or glucose in T cell culture samples from late intermediate stages of the manufacturing process of the CAR T drug product; the concentration of lactate or glucose in T cell culture samples from intermediate stages of the manufacturing process of the CAR T drug product; the ratio of CD4+ T cells to CD8+ T cells in T cell culture samples from the initial stage of the manufacturing process of the CAR T drug product; the volume of vector added to the T cell culture samples during early intermediate stages of the manufacturing process of the CAR T drug product; the multiplicity of infection (MOI) of the vector added to the T cell culture samples during early intermediate stages of the manufacturing process of the CAR T drug product; the average percentage of live T cells in each population of T cell culture samples from early intermediate stages of the manufacturing process of the CAR T drug product; the average concentration of live T cells in each population from early intermediate stages of the manufacturing process of the CAR T drug product; and the percentage of cells as CAR+ T cells in each population of the CAR T drug product. The concentration of lymphocytes in a single sample taken before the manufacturing process of the T-drug product; the percentage of cells as live T cells in a T-cell culture sample from the initial stage of the manufacturing process of the CAR T-drug product; whether the target patient is refractory to pomalidomide treatment; the sex of the target patient; the age of the target patient; the body mass index (BMI) of the target patient; or the number of previous lines of treatment for the target patient.

[0021] According to various implementation schemes, a method for treating cancer in a subject in need is also disclosed, which includes administering a CAR T drug product prepared by any of the methods described herein to the subject to treat the cancer.

[0022] According to another embodiment, a system for predicting manufacturing failures in the production of patient-specific CAR T drug products for a target patient is disclosed. The system includes: a processor; and a memory storing instructions that, when executed by the processor, cause the processor to perform the methods described in any aspect disclosed herein.

[0023] According to another embodiment, a non-transitory computer-readable medium is disclosed. Each non-transitory computer-readable medium has computer-readable instructions stored thereon that are executable to cause the operation of the methods described in any aspect disclosed herein to be performed.

[0024] According to some aspects, the methods described in this disclosure can be embedded as computer program code in a computer-readable medium, the computer program code including instructions that cause a processor to perform the steps of the method.

[0025] Other aspects, features, and specific embodiments will become apparent to those skilled in the art after reading the following description of specific exemplary aspects in conjunction with the accompanying drawings. While features may be discussed with respect to certain aspects and the drawings below, various aspects may include one or more of the advantageous features discussed herein. In other words, while one or more aspects may be discussed as having certain advantageous features, one or more of such features may also be used according to various aspects. Similarly, while exemplary aspects may be discussed below as aspects of an apparatus, system, or method, exemplary aspects may be implemented in various apparatuses, systems, and methods.

[0026] The foregoing has provided a fairly broad overview of the features and technical advantages of the examples according to this disclosure in order to better understand the detailed description that follows. Additional features and advantages will be described below. The disclosed concepts and specific examples can be readily used as the basis for modifying or designing other structures for performing the same purpose of this disclosure. Such equivalent constructions do not depart from the scope of the appended claims. The characteristics of the concepts disclosed herein (both their organization and manner of operation) and the associated advantages will be better understood from the following description when considered in conjunction with the accompanying drawings. Each drawing is provided for illustrative and descriptive purposes and not as a definition of limitation of the claims.

[0027] While aspects and implementations are described herein by way of example, those skilled in the art will understand that additional specific implementations and use cases may occur in many different arrangements and scenarios. The innovations described herein can be implemented across many different platform types, devices, systems, shapes, sizes, and packaging arrangements. While some examples may or may not be specific to a particular use case or application, a wide variety of applicability to the described innovations can occur. Attached Figure Description

[0028] Figures 1A to 1B This is a block diagram illustrating an example method 100 for manufacturing a CAR T drug product according to a non-limiting embodiment of the present disclosure.

[0029] Figure 2This is a block diagram illustrating the various stages of an example CAR T drug product manufacturing process 200, from which parameters are generated to predict whether the manufactured CAR T drug product will result in manufacturing failure (“manufacturing failure outcome”).

[0030] Figure 3 This is a block diagram illustrating an example computer network environment 300 for predicting manufacturing failure outcomes in the production of CAR T drugs and optimizing CAR T drug production based on manufacturing failure outcomes, according to a non-limiting embodiment of the present disclosure.

[0031] Figure 4 This is a block diagram illustrating an example process 400 for predicting manufacturing failure outcomes in the production of CAR T drugs and optimizing CAR T drug production based on manufacturing failure outcomes, according to a non-limiting embodiment of this disclosure.

[0032] Figure 5A This is a diagram illustrating an example decision tree modeling of parameter thresholds for predicting manufacturing failure outcomes of CART drug products according to a non-limiting embodiment of this disclosure.

[0033] Figure 5B This is a block diagram illustrating an example process for training a decision tree model according to a non-limiting embodiment of the present disclosure.

[0034] Figure 6A This is a block diagram illustrating an example method 600 for predicting whether a patient-specific CAR T drug product for a target patient will fail to be manufactured, according to a non-limiting embodiment of this disclosure. Furthermore, Figure 6B Figure 6C shows a table of example parameters described in this disclosure that are significant for the ability of a patient-specific CAR T drug product to predict whether it will fail to be manufactured for a target patient.

[0035] The same reference numerals and names in the various figures indicate the same elements. Detailed Implementation

[0036] As previously discussed, expectations and needs can better predict potential manufacturing failures during the production of CAR-T drugs. Identifying aspects of the manufacturing process that could lead to CAR-T drug product failure can help salvage the CAR-T drug production process, recover production resources, and avoid delays in patient care.

[0037] Therefore, this disclosure describes various systems, methods, and improved manufacturing techniques for predicting and reducing manufacturing failures of CAR T drug products. For example, this disclosure provides, at least in part, embodiments for determining manufacturing outcomes of CAR T drug therapies such as Chidamide Orenza DP. This disclosure relates, at least in part, to the finding that certain characteristics (such as screening characteristics, pre-inoculation characteristics, inoculation characteristics, and / or manufacturing characteristics) can be determinants of manufacturing outcomes. In some embodiments, patient factors can be associated with manufacturing failure.

[0038] Data on such characteristics (referred to herein as parameters) can be obtained at various stages of the CAR T drug manufacturing process, including but not limited to the patient screening stage, pre-inoculation stage, inoculation stage, and manufacturing stage. Furthermore, each stage can include or be divided into one or more sub-stages. For example, the manufacturing stage can include or be divided into an initial stage, an early intermediate stage, an intermediate stage, a late intermediate stage, and a late stage, wherein the aforementioned sub-stages can be distinguished from each other based on time, sequence, and / or associated events. The present invention has found that each set of parameters is significant for predicting the outcome of the CAR T drug product produced during the CAR T drug manufacturing process. In at least one embodiment, a set of parameters obtained during the CAR T drug product manufacturing process can be used to predict whether using the CAR T drug product produced by the manufacturing process will lead to manufacturing failure. Furthermore, the prediction can be used to optimize or improve the CAR T drug product manufacturing process, for example, by adjusting the parameters of the CAR T drug product manufacturing process. For example, a computing device can receive quantitative data of a set of input parameters from one or more stages of the manufacturing process. This set of input parameters can include the input parameters outlined in Appendix A. Each input parameter can belong to or be classified into one of the multiple parameter types outlined in Appendix A. The computing device can generate an input feature vector that includes quantitative data of the set of input parameters. This input feature vector can be applied to a trained machine learning model to generate an output feature vector that predicts whether the production of a CAR-T drug will experience manufacturing failure (referred to herein as a "manufacturing failure outcome"). The manufacturing failure outcome can be based on an assessment of one or more attributes of the CAR-T drug product, evaluating whether it meets the requirements or recommendations for one or more attributes in the CAR-T drug product's specifications.

[0039] As used herein, "significance of...for predictability," "significance of...prediction," or "significance of...prediction," such as when used to describe the significance of parameters used to predict manufacturing failure outcomes of CAR T drug products, can refer to a quantitative measure of how accurately the parameters predict manufacturing failure outcomes of CAR T drug products. In some embodiments, the significance of parameters in predicting manufacturing failure outcomes can be represented as mathematical weights, whereby parameters with higher weights will predict manufacturing failure outcomes better than parameters with lower weights. The weights of various parameters in relation to the significance of predicting manufacturing failure outcomes can be determined or learned by training a machine learning model. Furthermore, a set of parameters can be ranked (e.g., ranked) based on their significance in predicting manufacturing failure outcomes for a given CAR T drug production, where parameters with higher rankings predict manufacturing failure outcomes better than parameters with lower rankings.

[0040] As used herein, the “screening phase” can refer to a stage in the manufacturing process of CAR T drug products, in which patients are selected to obtain biological samples for use in the manufacturing process of CAR T drug products. During the screening phase, parameters related to patient characteristics (e.g., patient demographics, diagnosis of the patient’s disease, previous treatment history, treatment refractory status, etc.) can be obtained; these parameters may be referred to herein as “screening parameters.”

[0041] As used in this article, the "pre-apheresis phase" can refer to the stage in the manufacturing process of CAR-T drugs, following the screening stage, where laboratory tests are performed on biological samples from patients to determine additional information about them in preparation for apheresis. During the pre-apheresis phase, parameters related to these laboratory tests can be obtained.

[0042] As used herein, “collection phase” can refer to a phase in the manufacturing process of a CAR T-drug product that follows the screening phase (and in some embodiments, follows the pre-collection phase) but precedes the manufacturing phase, in which a blood sample containing T cells is separated from a selected patient for use in a manufacturing facility to manufacture the CAR T-drug product. Parameters obtained from this separated blood sample (referred to herein as a single sample) can be referred to as “collection phase parameters” or “collection parameters”.

[0043] As used herein, "manufacturing stage" or "manufacturing process" can refer to a stage in the manufacturing process of a CAR T-drug product following the single-sample stage, in which the single sample is further processed to manufacture the CAR T-drug product. Processing includes genetically modifying the T cells of the single sample to produce a chimeric antigen receptor (CAR). As used herein, cell culture samples derived from the single sample and used in the various sub-stages of the manufacturing stage can be referred to as "T-cell culture samples." Parameters obtained from this T-cell culture sample during the manufacturing stage or process can be referred to as "manufacturing stage parameters."

[0044] Furthermore, it is conceivable that the manufacturing stage or process may include or may be divided into various sub-stages, including but not limited to the initial stage, early intermediate stage, intermediate stage, late intermediate stage and late stage of the manufacturing stage or process.

[0045] As used herein, the “initial phase” of a manufacturing process can refer to the first sub-phase of the manufacturing process following the single-sample phase and can be associated with the preparation of T cell cultures using single samples and the enrichment and activation of T cells in the T cell cultures. In some embodiments, the initial phase may include or may include day 0, day 1, day 2, or day 3 of the manufacturing process, a range or value defined by any two of the aforementioned days, such as days 0-2 of the manufacturing process, preferably days 0-1 of the manufacturing process. In some embodiments, the initial phase may be further subdivided into an early initial phase and a late initial phase. As used herein, the “early initial phase” of a manufacturing process can refer to the first sub-phase of the manufacturing process following the single-sample phase and can be associated with the preparation of T cell cultures using single samples. In some embodiments, the early initial phase may include day 0 of the manufacturing process. As used herein, the “late initial phase” of a manufacturing process can refer to the first sub-phase of the manufacturing process associated with the enrichment and activation of T cells in the T cell cultures. In some implementations, the later initial phase may include or may include day 1, day 2, or day 3 of the manufacturing process, or a range or value defined by any two of the aforementioned days, such as days 1-3 of the manufacturing process, preferably day 1 of the manufacturing process.

[0046] As used herein, an “early intermediate stage” of the manufacturing process can refer to a sub-stage of the manufacturing process associated with CAR stimulation and / or transduction of T cells in a T cell culture. In some embodiments, the early intermediate stage may include or may include day 2, day 3, day 4, or day 5 of the manufacturing process, or a range or value defined by any two of the aforementioned days, such as days 2-4 of the manufacturing process, preferably day 3 of the manufacturing process.

[0047] As used herein, an “intermediate stage” of the manufacturing process may refer to a sub-stage of the manufacturing process associated with the expansion and growth monitoring of T cell cultures. In some embodiments, an intermediate stage may include or may include day 5, day 6, or day 7 of the manufacturing process, or a range or value defined by any two of the aforementioned days, such as day 6 of the manufacturing process.

[0048] As used herein, a “late intermediate stage” of the manufacturing process can refer to a sub-stage of the manufacturing process that is associated with the continued expansion and growth monitoring of the T-cell culture after the intermediate stage. In some embodiments, the late intermediate stage may include or may include day 7, day 8, day 9, or day 10 of the manufacturing process, or a range or value defined by any two of the aforementioned days, such as days 7-9 of the manufacturing process, preferably day 8 of the manufacturing process.

[0049] As used herein, a “late stage” of the manufacturing process can refer to a sub-stage of the manufacturing process associated with harvesting and releasing the final product (e.g., a CAR+ T cell drug) from the T cell culture. In some embodiments, the late stage may include or may include days 9, 10, 11, 12, 13, or 14 of the manufacturing process, or a range or value defined by any two of the aforementioned days, such as days 9-12 of the manufacturing process, days 10-12 of the manufacturing process, preferably day 10 of the manufacturing process.

[0050] I. Example Techniques for Parameter Acquisition

[0051] In some embodiments, the parameter information used in the methods disclosed herein is collected from the patient. Parameter information may include screening-stage characteristics (including patient characteristics), pre-abortion characteristics, apheresis characteristics, and manufacturing characteristics. Patient characteristics may include patient demographics, diagnostic information of the patient's disease, the patient's prior treatment history, the patient's refractory status, or a combination thereof. Pre-abortion characteristics may include characteristics of biological samples (such as blood samples) obtained from the patient. Pre-abortion characteristics may include physical characteristics of the biological sample, measured protein levels, protein electrophoresis measurements (such as urine protein electrophoresis and / or serum protein electrophoresis), or a combination thereof. Apheresis characteristics may include measurement characteristics of the apheresis material obtained from the patient. Apheresis characteristics may include flow cytometry data obtained from the apheresis material. Apheresis characteristics may include gene expression data. Apheresis characteristics may include sequencing data, including RNA sequencing data. Manufacturing characteristics may include sites of any of the following: sites of manufacturing cell therapies (including any cell therapies disclosed herein), sites of sample collection, sites of sample cryopreservation, sites of clinical research, and / or sites of processing any material. Manufacturing characteristics may include processing characteristics, post-thawing characteristics, post-washing characteristics, vitality characteristics, or combinations thereof.

[0052] In some embodiments, the parameter information includes any parameters from the screening phase 610. In some embodiments, the parameter information includes any parameters from the pre-collection phase 620. In some embodiments, the parameter information includes any parameters from the collection phase 630. In some embodiments, the parameter information includes any parameters from the manufacturing phase 640.

[0053] II. Example Methods for Manufacturing CAR T Drug Products

[0054] A. Screening Phase

[0055] In various implementations, the manufacturing process of CAR T-drug products can begin with a screening phase, where patients can be screened for various parameters (referred to herein as “screening phase parameters”). Screening phase parameters can help select suitable patients for generating a biological sample (from which the CAR T-drug product is generated), as well as obtain additional patient information that can be used for the efficacy of the CAR T-drug product therapy. Therefore, data for the screening phase parameters can be obtained. Screening phase parameters may include patient characteristics such as patient demographics (referred to herein as patient demographic parameters) and patient medical history (referred to herein as patient medical history parameters). In some implementations, the data is provided by individuals (e.g., patients or healthcare providers) and / or extracted from networks of electronic health records (EHRs), insurance claims, and census data. In some aspects, EHRs can be used alone to access, identify, or implement categories of these networks. In some implementations, screening phase parameters may also be based on social determinants of health, exocomics, tumor registries, biological samples, genomic outcomes, natural language processing, or patient-generated data. In some implementations, data for the screening phase parameters may be obtained by clinicians and entered into computing devices. Parameters used as screening parameters for patient demographics may include, but are not limited to: age, sex, race, body mass index, ethnicity, and country of origin.

[0056] B. Biomaterials Data

[0057] In some implementations, biological samples are collected from patients (including any patients disclosed herein). The biological samples can be processed and measured. Processing and / or measurement can be used to obtain one or more parameters disclosed herein, such as pre-apheresis characteristics, apheresis characteristics, and / or manufacturing characteristics.

[0058] 1. Sample preparation

[0059] In some embodiments, the method involves obtaining a sample from a subject. The methods provided herein may include biopsy methods such as fine-needle aspiration, core needle biopsy, vacuum-assisted biopsy, excisional biopsy, excisional biopsy, drill biopsy, scraping biopsy, or skin biopsy. In some embodiments, the sample is a blood sample. In some embodiments, the sample is obtained from a biopsy. Alternatively, the sample may be obtained from any other source, including but not limited to urine, sweat, hair follicles, cheek tissue, tears, menstrual blood, feces, or saliva. In some embodiments of the method of the invention, any medical professional, such as a physician, nurse, or medical technician, can obtain the biological sample for testing. Furthermore, biological samples can be obtained without the assistance of a medical professional.

[0060] Samples may include, but are not limited to, tissues, cells, or biological materials derived from or originating from subject cells. Biological samples may be heterogeneous or homogeneous populations of cells or tissues. Biological samples may be obtained using any method known in the art that can provide samples suitable for the analytical and / or manufacturing methods described herein.

[0061] Samples can be prepared using methods known in the art. In some embodiments, samples are obtained by biopsy. In other embodiments, samples are obtained by venipuncture or any other method known in the art. In some cases, samples can be obtained, stored, or transported using components of a kit based on the methods of the present invention. In some cases, multiple samples, such as multiple blood samples, can be obtained for processing, determination, and / or manufacturing by the methods described herein.

[0062] In some embodiments, the biological sample may be obtained by a physician, nurse, or other medical professional, such as a medical technician, endocrinologist, cytologist, hemistry physician, radiologist, or pulmonologist. The medical professional may instruct on the appropriate tests or assays to be performed on the sample. In some embodiments, the molecular profiling company may provide advice on which assays or tests are most suitable. In another embodiment of the method of the invention, the patient or subject may obtain the biological sample for testing without the assistance of a medical professional, such as obtaining a whole blood sample.

[0063] In other cases, samples are obtained through invasive procedures, including but not limited to: biopsy, needle aspiration, endoscopy, or venous incision. Needle aspiration may also include fine-needle aspiration, core needle biopsy, vacuum-assisted biopsy, or large-needle biopsy. In some embodiments, multiple samples can be obtained using the methods described herein to ensure a sufficient quantity of biological material.

[0064] In some implementations of the methods described herein, healthcare professionals are not required to be involved in the initial diagnosis or sample collection. Individuals may alternatively obtain samples using over-the-counter (OTC) kits. OTC kits may contain devices for obtaining samples as described herein, devices for storing the samples for testing, and instructions for proper use of the kit. In some cases, molecular profiling services are included in the purchase price of the kit. In other cases, molecular profiling services are billed separately. Samples suitable for use by molecular profiling companies can be any material containing tissues, cells, nucleic acids, genes, gene fragments, expression products, gene expression products, or fragments of gene expression products of the individual to be tested. Methods for determining sample suitability and / or adequacy are provided.

[0065] In some implementations, the subject can refer to an expert, such as an oncologist, surgeon, or endocrinologist. The expert may also obtain the biological sample for testing or send the individual to a testing center or laboratory to submit the biological sample. In some cases, medical professionals may send the subject to a testing center or laboratory to submit the biological sample. In other cases, the subject may provide the sample. In some cases, molecular profiling companies may obtain the sample.

[0066] 2. Material characteristics

[0067] In some embodiments, the characteristics of biological materials, such as blood samples (which may contain pre-apheresis materials), apheresis materials, or manufactured cell therapy materials, are determined. In some embodiments, the characteristics include one or more of the pre-apheresis parameters, apheresis parameters, and / or manufacturing parameters disclosed herein. In some embodiments, the characteristics are determined using known protein assay methods such as protein electrophoresis. In some embodiments, the characteristics are determined by measuring specific proteins such as specific antibodies, specific light chains, specific heavy chains, specific immunoglobulins, and / or specific cell markers. In some embodiments, the characteristics are determined by measuring cell viability. Cell viability can be measured using known techniques, including flow cytometry.

[0068] 3. Flow cytometry

[0069] In some embodiments, flow cytometry data are collected. In some embodiments, flow cytometry data are collected on materials from the single-harvest stage. In some embodiments, flow cytometry data are collected on materials from the manufacturing stage. Flow cytometry can be performed using standard techniques.

[0070] In some embodiments, the biological material to be measured by flow cytometry (including apheresis-stage material and / or manufacturing-stage material) is prepared for flow cytometry, including by generating a single-cell suspension. The prepared material may be contacted with one or more proteins capable of binding selective cell markers. In some embodiments, selective cell markers indicating that the apheresis material will result in or may result in a non-compliant CAR T drug product. In some embodiments, selective cell markers include one or more markers disclosed herein, including any markers disclosed in the apheresis-stage and / or manufacturing-stage parameters. In some embodiments, selective cell markers include skeletal markers (including viability markers), lineage markers, activation markers, differentiation markers, depletion markers, or combinations thereof. In some implementations, selective cell markers include CD14, CD19, CD16, CD56, HLA-DR, CD25, CD57, CCR7, CD45RA, CD45RO, CD95, CD127, CD27, CD28, CD57, KLRG1, CD39, CD244, CD160, CX3CR1, CD85j, Tim-3, NKG2A, CD90, CD126, PD-1, LAG-3, TIGIT, OX-40, CD103, KLRG1, CD80, GPR56, CD158, CD123, CD38-HIT, CD244, CD45, CD3, CD4, CD8, anti-ID, or combinations thereof. In some aspects, one or more proteins capable of binding to the selective cell marker are labeled with a fluorophore.

[0071] 4. RNA sequencing

[0072] Input data used in the methods described herein (including those for the pre-absorption, absorption, and / or manufacturing stages) may include sequencing data, including but not limited to raw sequencing reads of RNA from a subject (e.g., a patient), including raw sequencing reads from a single cell. In some aspects, RNA can be analyzed by sequencing. RNA can be prepared for sequencing using any method known in the art, such as poly-A selection, cDNA synthesis, strand or non-strand library preparation, or combinations thereof. RNA can be prepared for any type of RNA sequencing technology, including strand-specific RNA sequencing. In some aspects, sequencing can be performed to generate approximately 10M, 15M, 20M, 25M, 30M, 35M, 40M, or more reads, including paired reads. Sequencing can be performed with read lengths of approximately 50bp, 55bp, 60bp, 65bp, 70bp, 75bp, 80bp, 85bp, 90bp, 95bp, 100bp, 105bp, 110bp, or longer. In some respects, raw sequencing data can be converted into estimated read counts (RSEM), fragments per million mapped reads per kilobase transcript (FPKM), and / or reads per million mapped reads per kilobase transcript (RPKM).

[0073] In some respects, RNA sequencing includes single-cell RNA sequencing (scRNA-Seq). In other respects, RNA sequencing includes known sequencing technologies, including but not limited to any of the following:

[0074] CITE-Seq

[0075] CITE-Seq (Cell Indexing of Transcriptome and Epitopes by Sequencing) is a method for performing RNA sequencing and obtaining quantitative and qualitative information about surface proteins at the single-cell level using available antibodies. It provides an additional layer of information to the same cell by combining proteomics and transcriptomics data. For phenotypic analysis, this method has been demonstrated by the team that developed it to be as accurate as flow cytometry (the gold standard).

[0076] Drop-Seq

[0077] Drop-Seq analyzes mRNA transcripts from single-cell droplets in a highly parallel manner. This single-cell sequencing method uses a microfluidic device to partition droplets containing single cells, lysis buffer, and microbeads coated with barcode primers. Each primer contains: 1) a 30 bp oligo(dT) sequence that binds to mRNA; 2) an 8 bp molecular index to uniquely identify each mRNA strand; 3) a 12 bp barcode specific to each cell; and 4) a universal sequence identical across all beads. After partitioning, the cells in the droplets lyse, and the released mRNA hybridizes to the oligo(dT) bundles of the primer beads. Next, all droplets are combined and broken to release the beads. After separating the beads, they are reverse transcribed using a template conversion. This generates a first cDNA strand with the PCR primer sequences replacing the universal sequence. The cDNA is PCR amplified, and sequencing adapters are added using the Nextera XT Library Preparation Kit. Barcode mRNA samples are prepared for sequencing. The method is further described in Macosko, Evan Z. et al., Cell, 2015, 161(5): 1202-1214, which is incorporated herein by reference.

[0078] inDrop

[0079] inDrop is used for high-throughput single-cell labeling. This method is similar to Drop-seq, but it uses hydrogel microspheres to introduce oligonucleotides. Single cells from a cell suspension are isolated into droplets containing lysis buffer. After cell lysis, the cell droplets are fused with hydrogel microspheres containing cell-specific barcodes and another droplet containing an enzyme for RT. Droplets from all wells are combined and subjected to an isothermal reaction for RT. The barcodes are annealed with poly(A)+ mRNA and act as primers for reverse transcriptase. Now that each mRNA strand has a cell-specific barcode, the droplets are combined and destroyed, and the cDNA is purified. The 3' end of the cDNA strand is ligated to an adapter, amplified, annealed to an index primer, and further amplified prior to sequencing. This method is further described in Klein, Allon M. et al., Cell, 2015, 161(5): 1187–1201, which is incorporated herein by reference.

[0080] CEL-seq

[0081] CEL-Seq overcomes the challenge of low input by using RNA barcoding and merging. In this method, each cell undergoes RT with uniquely barcoded primers in its separate tube. After second-strand synthesis, cDNA from all reaction tubes is merged and PCR amplified. Paired-end deep sequencing of the PCR products allows for accurate detection of sequence information derived from both strands. This method and the associated CEL-seq2 are further described in Hashimshony, T. et al., Cell Reports, 2012, 2(3): 666-673 and Hashimshony, T. et al., Genome Biology, 2016, 17(1): 77, which are incorporated herein by reference.

[0082] Quartz-Seq

[0083] The Quartz-Seq method optimizes whole-transcriptome amplification (WTA) for single cells. In this method, RT primers containing the T7 promoter and PCR target are first added to the extracted mRNA. First-strand cDNA is synthesized via RT, followed by digestion of the RT primers with exonuclease I. Next, a poly(A) tail is added to the 3' end of the first-strand cDNA along with a poly(dT) primer containing the PCR target. After second-strand generation, blocking primers are added to ensure sufficient PCR enrichment for sequencing. Deep sequencing allows for accurate, high-resolution representation of the whole transcriptome of a single cell.

[0084] MARS-Seq

[0085] MARS-Seq analyzes transcriptional dynamics of single cells at high resolution in an automated and massively parallel workflow. MARS-Seq can be used for in vivo samples containing a variety of different cell subpopulations. First, single cells are isolated into individual wells using FACS. Each cell is lysed, and the 3' end of the mRNA is annealed to a unique molecular identifier containing the T7 promoter. The mRNA is reverse transcribed to generate a first cDNA strand, and treated with exonuclease I to remove any remaining RT primers. Next, the cell lysates are combined and converted to double-stranded cDNA. The DNA strand is transcribed into RNA, and treated with DNase to remove any remaining DNA template from the mixture. The RNA strand is fragmented and annealed to sequencing adapters, followed by RT to generate a barcoded cDNA library ready for sequencing.

[0086] CytoSeq

[0087] CytoSeq enables gene expression profiling analysis of thousands of single cells. In this method, single cells are randomly deposited in wells. A combined library of beads containing specific capture probes is added to each well. After cell lysis, mRNA is hybridized to the beads, which are then combined for RT, amplification, and sequencing. Deep sequencing provides accurate, high-coverage gene expression profiles for several single cells.

[0088] Hi-SCL

[0089] Hi-SCL uses a custom microfluidic system to generate transcriptome profiles of thousands of single cells, similar to Drop-Seq and inDrop. Single cells from cell suspension are isolated into droplets containing lysis buffer. After cell lysis, the cell droplets are fused with a droplet containing cell-specific barcodes and another droplet containing an enzyme for reverse transcriptase. Droplets from all wells are combined and subjected to an isothermal reaction for RT. The barcodes are annealed with poly(A)+ mRNA and act as primers for reverse transcriptase. Now that each mRNA strand has a cell-specific barcode, the droplets are destroyed and cDNA is purified. The 3' end of the cDNA strand is ligated to an adapter, amplified, annealed to index primers, and further amplified before sequencing.

[0090] Seq-Well

[0091] Single-cell RNA-seq can accurately resolve cellular states, but applying this method to low-input samples is challenging. Here, the inventors present Seq-Well, a portable, low-cost platform for massively parallel single-cell RNA-seq. Barcode mRNA capture beads and single cells are sealed in a sub-nano-well array using a semi-permeable membrane, enabling efficient cell lysis and transcript capture. This method is further described in Gierahn et al., Nat Methods. 2017 Apr; 14(4):395-398, which is incorporated herein by reference. This method is further described in Gierahn, TM et al., Nature Methods, 2017.14: p. 395, which is incorporated herein by reference.

[0092] Microwell-seq

[0093] Microwell-seq confines single-cell and barcode-based (dT) mRNA-capturing beads in a sub-nanowell PDMS array. The pore size is designed to accommodate only one bead. Cells are loaded by gravity at a dual occupancy rate, which can be tuned by adjusting the cell number, and are loaded and visualized prior to processing. This method is further described in Han, X. et al., Cell, 2018, 172(5): 1091–1107.e17, which is incorporated herein by reference.

[0094] Nanogrid-seq

[0095] Nanogrid-seq is a nanogrid platform and microfluidic deposition system that enables the parallel imaging, selection, and sequencing of thousands of single cells or nuclei. This method is further described in Gao, R. et al., Nature Communications, 2017.8(1): p. 228, which is incorporated herein by reference.

[0096] sci-seq

[0097] Sci-seq refers to Single-Cell Composite Index Sequencing (SCI-seq), which can be used as a means to simultaneously generate thousands of low-pass single-cell libraries for somatic copy number variation detection. This is further described in Vitak, SA et al., Nature Methods, 2017.14: p. 302, which is incorporated herein by reference.

[0098] Direct labeling

[0099] An enzyme called a transposase randomly cuts DNA into short fragments (“tags”). Adapters (ligations) are added to either side of the cut site. Strains that failed to ligate adapters are washed away. Adapters may contain barcodes and / or primer binding sites for use in detecting and amplifying genomic sequences. This is further described in Zahn, H. et al., Nature Methods, 2017.14: p.167, which is incorporated herein by reference.

[0100] Sci-ATAC-seq

[0101] sci-ATAC-seq is a single-cell ATAC-seq protocol. This technique can be used to determine chromatin accessibility between and within single-cell populations. Single-cell ATAC-Seq relies on a combined cell index and therefore does not require physical separation of individual cells during library construction. This technique exhibits sublinear growth in time and cost and can analyze thousands of individual cells in a single experiment. This method is further described in Cusanovich, DA et al., Science, 2015, 348(6237): 910, which is incorporated herein by reference. A related method, nanopore scATAC-seq, is described in Mezger, A. et al., High-throughput chromatin accessibility profiling at single-cell resolution, bioRxiv, 2018, which is incorporated herein by reference.

[0102] Other methods include the 10x genomic RNA sequencing platform, described in Zheng, GXY et al., Nature Communications, 2017.8: p. 14049; SMART-seq, described in Ramsköld, D. et al., Nature Biotechnology, 2012.30: p. 777; and SMART-seq2, described in Picelli, S. et al., Nature Protocols, 2014.9: p. 171. All of these references are incorporated herein by reference in their entirety. It is conceivable that aspects of the published references may be incorporated into aspects described herein.

[0103] 5. Sequencing methods

[0104] Massive parallel signature sequencing (MPSS) .

[0105] The first next-generation sequencing technology, massively parallel signature sequencing (or MPSS), was developed by Lynx Therapeutics in the 1990s. MPSS is a bead-based approach that uses a complex method of adapter ligation followed by adapter decoding to read sequences in increments of four nucleotides. This method makes it susceptible to sequence-specific bias or the loss of specific sequences. Because the technology was so complex, Lynx Therapeutics performed MPSS only “internally,” and no DNA sequencers were sold to independent laboratories. Lynx Therapeutics merged with Solexa (later acquired by Illumina) in 2004, leading to the development of sequencing-by-synthesis, a simpler method derived from Manteia Predictive Medicine, which rendered MPSS obsolete. However, the fundamental characteristics of MPSS output are typical of later “next-generation” data types, including hundreds of thousands of short DNA sequences. In the case of MPSS, these are often used for sequencing cDNA to measure gene expression levels. In fact, the powerful Illumina HiSeq2000, HiSeq2500, and MiSeq systems are all based on MPSS.

[0106] Polony sequencing .

[0107] Developed in George M. Church's lab at Harvard, the Polony sequencing method was one of the first next-generation sequencing systems and was used to sequence the entire genome in 2005. It combines in vitro paired-tag libraries with emulsion PCR, automated microscopy, and ligation-based sequencing chemistry to sequence the *E. coli* genome with >99.9999% accuracy and approximately one-ninth the sequencing cost of Sanger. The technology was licensed to Agencourt Biosciences, subsequently spun off into Agencourt Personal Genomics, and eventually integrated into the Applied Biosystems SOLiD platform, now owned by Life Technologies.

[0108] 454 pyrosequencing .

[0109] Parallel versions of pyrosequencing were developed by 454 Life Sciences, which was subsequently acquired by Roche Diagnostics. This method amplifies DNA within water droplets in an oil solution (emulsion PCR), each droplet containing a single DNA template attached to a single primer-coated bead, which then forms clonal colonies. The sequencer contains numerous picoliter-volume wells, each containing a single bead and a sequencing enzyme. Pyrosequencing uses luciferase to generate light to detect single nucleotides added to the nascent DNA, and the combined data is used to generate sequence reads. Compared to Sanger sequencing at one end and Solexa and SOLiD at the other, this technology offers intermediate read lengths and cost-per-base.

[0110] Illumina (Solexa) sequencing .

[0111] Solexa, now part of Illumina, has developed sequencing methods based on reversible dye terminator technology, as well as its internally developed engineered polymerases. The chemistry of termination was developed within Solexa, and the concept for the Solexa system was conceived by Balasubramanian and Klennerman from the Department of Chemistry at the University of Cambridge. In 2004, Solexa acquired Manteia Predictive Medicine to gain access to massively parallel sequencing technology based on “DNA clusters,” which involves the clonal amplification of DNA on a surface. This cluster technology was acquired in conjunction with Lynx Therapeutics in California. Solexa Ltd. subsequently merged with Lynx to form Solexa Inc.

[0112] In this method, DNA molecules and primers are first attached to a glass slide and amplified with polymerase to form locally cloned DNA colonies, which then become “DNA clusters.” To determine the sequence, four types of reversible terminator bases (RT-bases) are added, and any unincorporated nucleotides are washed away. Images of the fluorescently labeled nucleotides are captured with a camera, and the dye is then chemically removed from the DNA along with a 3' end blocker, allowing the next cycle to begin. Unlike pyrosequencing, the DNA strand is extended one nucleotide at a time, and images can be acquired at delayed moments, allowing for the capture of very large arrays of DNA colonies from a single camera in a series of images. Separating the enzymatic reaction and image capture enables optimal throughput and theoretically unlimited sequencing capabilities. Therefore, under optimal configuration, the final achievable instrument throughput is determined solely by multiplying the camera’s analog-to-digital conversion rate by the number of cameras and dividing by the number of pixels per DNA colony required for optimal visualization (approximately 10 pixels per colony). In 2012, using cameras operating at A / D conversion rates exceeding 10 MHz and available optics, fluid dynamics, and enzymology, throughput could be multiples of 1 million nucleotides per second, roughly corresponding to one human genome equivalent per hour at 1x coverage per instrument, and one human genome resequencing per day per instrument (equipped with a single camera) at approximately 30x.

[0113] SOLiD sequencing .

[0114] Applied Biosystems' (now a Thermo Fisher Scientific brand) SOLiD technology employs ligation-while-sequencing. Here, all possible pools of oligonucleotides of fixed length are labeled according to the sequencing location. The oligonucleotides are annealed and ligated; preferential ligation using DNA ligase to match sequences generates a signal providing nucleotide information for that location. DNA is amplified by emulsion PCR prior to sequencing. The resulting beads, each containing a single copy of the same DNA molecule, are deposited on a glass slide. The result is a sequence of comparable quantity and length to Illumina sequencing. This ligation-while-sequencing method has reportedly some issues when sequencing palindromic sequences.

[0115] Ion-current semiconductor sequencing .

[0116] Ion Torrent Systems Inc. (now owned by Thermo Fisher Scientific) has developed a system based on standard sequencing chemistry but featuring a novel semiconductor-based detection system. Unlike the optical methods used in other sequencing systems, this approach is based on detecting hydrogen ions released during DNA polymerization. Micropores containing the template DNA strand to be sequenced are filled with a single type of nucleotide. If the introduced nucleotide is complementary to the leader template nucleotide, it is integrated into the growing complementary strand. This results in the release of hydrogen ions, which triggers an ultrasensitive ion sensor indicating that a reaction has occurred. If homopolymer repeats are present in the template sequence, multiple nucleotides are incorporated into a single cycle. This results in a correspondingly higher number of hydrogen ions released and a proportionally higher electronic signal.

[0117] DNA nanosphere sequencing .

[0118] DNA nanosphere sequencing is a high-throughput sequencing technology used to determine the entire genome sequence of an organism. Complete Genomics uses this technology to sequence samples submitted by independent researchers. The method uses rolling circle replication to amplify small fragments of genomic DNA into DNA nanospheres. Then, non-stranded ligation-while-sequencing is used to determine the nucleotide sequence. Compared to other next-generation sequencing platforms, this DNA sequencing method allows sequencing of a large number of DNA nanospheres per run at a low reagent cost. However, since only a short sequence of DNA is determined from each DNA nanosphere, mapping these short reads to a reference genome is difficult. This technology has been used in multi-genome sequencing projects.

[0119] Heliscope single-molecule sequencing .

[0120] Heliscope sequencing is a single-molecule sequencing method developed by Helicos Biosciences. It uses DNA fragments with added poly-A tail adapters, which are attached to the surface of a flow cell. Subsequent steps involve extension-based sequencing, where the flow cell is cyclically washed with fluorescently labeled nucleotides (one nucleotide type at a time, as in the Sanger method). Reads are executed by the Heliscope sequencer. Reads are initially short, up to 55 bases per run, but recent improvements allow for more accurate reads of extension regions of a single nucleotide type. The genome of M13 bacteriophage was sequenced using this sequencing method and instrument.

[0121] Single-molecule real-time (SMRT) sequencing .

[0122] SMRT sequencing is based on sequencing via synthesis. DNA is synthesized in a zero-mode waveguide (ZMW) – a small-well container where a trapping tool is located at the bottom of the well. Sequencing is performed using an unmodified polymerase (attached to the bottom of the ZMW) and fluorescently labeled nucleotides flowing freely in solution. The well is constructed in such a way that fluorescence occurring only at the bottom of the well is detected. The fluorescent label is then separated from the nucleotides incorporated into the DNA strand, leaving the unmodified DNA strand. According to Pacific Biosciences, the developer of SMRT technology, this method allows the detection of nucleotide modifications, such as cytosine methylation. This occurs by observing polymerase kinetics. This method allows reads of 20,000 nucleotides or longer, with an average read length of 5 kilobases.

[0123] C. Imaging

[0124] In some embodiments, imaging data are also collected from the patient and used in the methods described herein. In some embodiments, the imaging data may be used alone or in combination with the manufacturing results of the cell therapy described herein to form a deterministic conclusion. Specific examples of contrast mechanisms using different types of imaging modalities, image types, and features include, but are not limited to, magnetic resonance imaging (MRI), computed tomography (CT), positron emission tomography (PET), and photoacoustic tomography (PAT).

[0125] III. Example Methods for Manufacturing CAR-T Drug Products

[0126] Figure 1 is a flowchart illustrating an example method 100 for manufacturing a CAR T drug product according to an example embodiment of the present disclosure. As will be described herein, various embodiments of the present disclosure describe systems and methods for predicting and optimizing manufacturing failure outcomes in CAR T drug production based on parameters obtained at various stages of the CAR T drug product manufacturing process (such as, but not limited to, example process 100).

[0127] A. Screening Phase

[0128] In various implementations, the CAR T drug product manufacturing process can begin with a screening phase, where patients can be screened for various parameters (referred to herein as “screening phase parameters” or “screening parameters”) (box 110). Screening phase parameters can help select suitable patients 112 for generating a biological sample (from which the CAR T drug product is generated), and obtain additional patient information for efficacy of the CAR T drug therapy. Therefore, data for the screening phase parameters can be obtained (box 114). Screening phase parameters can include patient characteristics such as patient demographics (referred to herein as “patient demographic parameters”) and patient medical history (referred to herein as “patient medical history parameters” or “medical history parameters”). In some implementations, the data is provided by an individual (e.g., a patient or healthcare provider) and / or extracted from networks of electronic health records (EHRs), insurance claims, and census data. In some aspects, EHRs can be used alone to access, identify, or implement categories of these networks. In some implementations, screening phase parameters can also be based on social determinants of health, exocomics, tumor registries, biological samples, genomic outcomes, natural language processing, or patient-generated data. In some implementations, data on screening parameters can be obtained by clinicians and entered into a computing device. In some implementations, patients selected during the screening phase may be patients with or who have had a disease such as multiple myeloma (MM) or another cancer that requires or indicates CAR T drug therapy.

[0129] B. Pre-single-agent stage

[0130] In various embodiments, following patient screening, an example process 100 for the CAR T drug product manufacturing process may include a pre-abortion stage 120, in which one or more biological samples 122 are obtained from the screened patient 112 for laboratory testing. The characteristics of the biological samples 122 tested in the pre-abortion stage 120 (e.g., in a laboratory) (referred herein to as “pre-abortion stage parameters” or “pre-abortion parameters”) can provide additional insights about the patient from whom the CAR T drug product is to be derived. Example process 100 may involve receiving data on these pre-abortion parameters (box 124). In some embodiments, additional insights may be used to further screen patients prior to the ablation stage 130 and manufacturing stage 140 of the CAR T drug product manufacturing process. In some aspects, the patient 112 may have or may have had a disease requiring CAR T drug therapy, such as multiple myeloma (MM). The biological samples 122 may be obtained from the patient based on the techniques described herein.

[0131] In at least one embodiment (e.g., as shown in box 122), the biological sample may include a patient's blood sample. In the example procedure, laboratory testing in the pre-abortion phase may include analyzing proteins from the patient's blood sample. In some embodiments, the blood sample may be electrophoresed to identify and / or detect various proteins and / or their characteristics. For example, electrophoresis and other techniques may be used to detect or measure characteristics (e.g., percentage, volume, concentration, etc.) of albumin, α-1 globulin, α-2 globulin, β-globulin, γ-globulin, monoclonal spike protein 1, or monoclonal spike protein 2 in the blood sample (e.g., in the serum of the blood sample). In some embodiments, the blood sample is subjected to standard clinical laboratory testing to determine total protein levels or to determine clinically relevant protein information. In some embodiments, blood urea nitrogen in the blood sample is measured. In some embodiments, electrophoresis and other techniques may be used to determine total M protein or total serum volume in the serum. In some embodiments, blood samples may be further tested to determine the characteristics of light chains, such as the absolute difference between affected and unaffected free light chains (DFLC value), a measurement of the amount of λ free light chains in the blood sample, the ratio of free κ light chains to free λ light chains in the blood sample, or a measurement of the amount of κ free light chains in the blood sample. Alternatively or additionally, as shown in box 124, the biological sample may include a patient's urine sample. In the example procedure, laboratory testing in the pre-collection phase may include analyzing proteins from a patient's urine sample. In some embodiments, protein electrophoresis may be performed on the urine sample to identify and / or detect various proteins and / or their characteristics. For example, the amount of protein in urine over 24 hours may be determined (e.g., protein in a 24-hour aliquot of urine) (e.g., via multiple urine samples from the patient over 24 hours). Furthermore, electrophoresis and other techniques may be used to detect or measure the characteristics (e.g., percentage, volume, concentration, etc.) of albumin, α-1 globulin, α-2 globulin, β-globulin, γ-globulin, and / or monoclonal spike protein 1 in the urine sample. In some embodiments, proteinuria in the patient is assessed. In some implementations, blood urea nitrogen in a urine sample is measured. In some implementations, results from laboratory tests during the apheresis phase can be used to determine a disease or disease classification, such as multiple myeloma (MM) classification (e.g., MM classification 2).

[0132] After laboratory testing using biological samples (e.g., urine samples, blood samples, etc.) to further screen patients and / or collect patient information in the pre-apheresis phase and obtain data on pre-apheresis parameters via laboratory testing (box 124), example method 100 of the CAR T drug product manufacturing process can proceed to the apheresis phase 130. Alternatively, in some embodiments, the CAR T drug product manufacturing process can proceed to the apheresis phase 130 after the screening phase 110.

[0133] C. Single-agent stage

[0134] In various embodiments, after patient screening phase 110 (and in some embodiments, after pre-apheresis phase 120), but before manufacturing phase 140, example method 100 of the CAR T drug product manufacturing process may include apheresis phase 130. Example method of apheresis phase 130 may include performing an apheresis procedure on a selected patient. The apheresis procedure may involve separating a blood sample from the patient containing T cells from the selected patient for use in a manufacturing facility to manufacture the CAR T drug product. In at least one embodiment, apheresis may be performed by an apheresis device fluidly connected to the patient's blood circulation, thereby allowing blood from the patient to enter the apheresis device. The apheresis device may be configured to separate various components of the blood (e.g., plasma, red blood cells, white blood cells, and platelets). The apheresis device may be further configured to separate the component of interest carrying T cells (e.g., white blood cells) from the remainder of the patient's blood to form a single sample (box 132).

[0135] Example process 100 may include receiving data on various parameters (referred to herein as “single-collection stage parameters” or “single-collection parameters”) from isolated single samples during the single-collection stage (box 136). Data acquisition may rely on various techniques described herein, such as, but not limited to, flow cytometry, sequencing, electrophoresis, and imaging (techniques indicated via box 134). In some embodiments, single-collection stage parameters may include parameters describing the expression or non-expression of cell surface markers, including any cell surface markers described herein, such as cell surface proteins, cell surface receptors, cell surface macromolecules, etc., and such parameters may be designated herein as cell surface marker parameters. When such cell surface marker parameters are acquired during the single-collection stage, such parameters may be further designated as single-collection stage-cell surface marker parameters.

[0136] Example CAR T drug product manufacturing process 100 may involve obtaining data on cell surface marker parameters using flow cytometry, sequencing, and / or electrophoresis techniques 134 described herein. Specifically, such techniques can be used to detect the presence of cell surface markers or to determine one or more characteristics of cell surface markers described by cell surface marker parameters, such as, but not limited to, the percentage of cells in a single sample that express or do not express a cell surface marker, the concentration of cells in a single sample that express or do not express a cell surface marker, the ratio between cells in a single sample that express a cell surface marker and cells in a single sample that express another cell surface marker, the count or volume of cells in a single sample that express a cell surface marker, etc.

[0137] Examples of cell surface markers for which information is obtained during the single-sample phase may include, but are not limited to, the presence or absence of CD4, CD8, CD11b, CD14, CD16, CD33, CD62L, HLA, DRA1, DRA1, CD192, CAR, CD25, CD27, CD27, CD28, CD28, CD38, CD39, CD3, PD1, CD57, KLRG, CCR7, CD45RA, or combinations thereof. For example, the techniques described herein (e.g., flow cytometry) can detect the presence of cells having a combination of the aforementioned cell surface markers or can determine one or more characteristics of such cells, such as the percentage of cells as CD25+, CAR-, or CD4+ T cells in a single sample.

[0138] In some implementations, the single-collection phase parameters may include parameters describing various characteristics of the single sample obtained using the single-collection process. As used herein, parameters describing the characteristics of a sample (e.g., a single sample) obtained via a process (e.g., single-collection parameters) and excluding cell surface marker parameters may be referred to as “process parameters.” When such process parameters are obtained during the single-collection phase, they may be further specified herein as “single-collection phase-process parameters.” An example CAR T drug product manufacturing process 100 may involve obtaining data on these process parameters using the flow cytometry, sequencing, and / or electrophoresis techniques described herein. Specifically, such process parameters may include measurements (e.g., percentages or concentrations) of one or more contents of the single sample, such as, but not limited to, lymphocytes, leukocytes, natural killer cells, stem cell natural killer cells, natural killer T cells, stem cell natural killer T cells, regulatory T cells, stem cell regulatory T cells, monocytes, neutrophils, memory T cells, and / or stem cell memory T cells in the single sample. In some implementations, process parameters may include a measurement of one type of content in a single sample relative to another type of content in a single sample (e.g., the percentage of white blood cells as monocytes in a single sample).

[0139] D. Manufacturing Stage

[0140] In various embodiments, following the single-sample stage 130, an example method 100 of the CAR T-drug product manufacturing process may include a manufacturing stage 140. In manufacturing stage 140, the single sample from the single-sample stage 130 may be further processed to manufacture the CAR T-drug product. Processing may include one or more of the following: activating and enriching T cells, genetically modifying T cells to produce a chimeric antigen receptor (CAR), and expanding and monitoring the growth of CAR+ T cells. Furthermore, manufacturing stage 140 may involve obtaining data on various parameters (referred to herein as “T-cell culture sample”) obtained from a cell culture sample derived from the single sample and used in various sub-stages of the manufacturing process. Such parameters obtained during any of the sub-stages of manufacturing stage 140 may herein be referred to as “manufacturing stage parameters” or “manufacturing process parameters”. Figure 1B As shown, manufacturing stage 140 may include, or may be divided into, an initial stage 141, an early intermediate stage 152, an intermediate stage 158, a late intermediate stage 164, and a late stage 170 of the manufacturing process. In some embodiments, at various sub-stages of the manufacturing process, T cell culture samples may be incubated, and various aspects of the incubation of the T cell culture (e.g., incubation time, incubation temperature, CO2 saturation, etc.) may be measured. In some embodiments, at various sub-stages of the manufacturing process, measurements (e.g., count, volume, etc.) of the T cells seeded in the T cell culture sample may be determined.

[0141] Manufacturing process 140 may begin with an initial stage 141, which may be further subdivided into an early initial stage 142 and a late initial stage 148. In the example embodiment shown in FIG1, the initial stage may include day 0 and day 1 of the manufacturing process (e.g., the early initial stage may occur on day 0 of the manufacturing process, while the late initial stage may occur on day 1 of the manufacturing process). However, in some embodiments, the initial stage may include or may include day 0, day 1, day 2, or day 3 of the manufacturing process, a range or value defined by any two of the aforementioned days (e.g., day 0 and day 1 of the manufacturing process).

[0142] In at least one embodiment, in an early initial stage 142, single samples can be prepared for the manufacturing process by thawing (e.g., from a frozen state after apheresis) and then washing the thawed sample (box 144). Furthermore, various characteristics of T cells in the single sample (e.g., viability, cell diameter, expression or non-expression of various cell surface markers, etc.) can be assessed after thawing and can be reassessed after washing. For example, flow cytometry can be used to detect or measure the characteristics of cells in T cell culture samples that express or do not express cell surface markers CD4, CD8, CD3, CD16, CD56, CD19, and / or CD14. In some embodiments, compounds can be added to prepare T cell culture samples for the manufacturing process, such as anticoagulants (e.g., ACD-A) and / or DNases (e.g., Pulmozyme). In some embodiments, in-line filtration can be performed to prepare single samples for the manufacturing process. In some embodiments, T cells from a T cell culture sample may be arranged or distributed in a bag (e.g., a CultiLife bag) that provides a sterile, breathable, closed system for growing and transducing the T cell culture sample. In such embodiments, the number of cells in each bag and the number of bags may be monitored at one or more sub-stages of the manufacturing process. Example method 100 may also include adding a cell culture medium, such as a GMP medium specifically for culturing T cells (e.g., TexMACS), to form or maintain a T cell culture (box 146).

[0143] In at least one embodiment, in a later initial stage 148, T cells in a T cell culture sample can be activated and enriched for use in the remainder of the manufacturing process (box 150). In some embodiments, activation can be performed by adding activation beads (e.g., T Cell TransAct beads) configured to activate and amplify an enriched population of T cells or resting T cells from a single sample. In some embodiments, the example methods may also involve detecting or evaluating the characteristics of clumps (e.g., the presence of clumps, the number of clumps, the size of one or more clumps, and the effect of mixing T cell culture samples on clumps) before and after T cell activation within the T cell culture sample (box 151).

[0144] Following the initial phase 141, the manufacturing process may proceed to an early intermediate phase 152, which may be associated with stimulating and / or transducing T cells in the T cell culture sample with a chimeric antigen receptor (CAR). In the example method shown in Figure 1, the early intermediate phase may include day 3 of the manufacturing process. However, in some embodiments, the early intermediate phase may include or may include day 2, day 3, day 4, or day 5 of the manufacturing process, or a range or value defined by any two of the aforementioned days, such as days 2-4 of the manufacturing process, preferably day 3 of the manufacturing process. In some embodiments, the example method 100 of the early intermediate phase 152 may involve assessing T cell viability (e.g., percentage of live T cells, live T cell count, live T cell concentration, T cell volume) in the T cell culture sample or within one or more bags containing the T cell culture sample (box 153). The example method may also involve mixing T cell culture samples (box 154). In some embodiments, example methods may also involve detecting or evaluating the characteristics of clumps within the T cell culture sample (e.g., the presence, size, number, and effects of mixing) before and after mixing the T cell culture sample. Furthermore, various characteristics of the T cells in the mixed T cell culture sample (e.g., pooled viability, pooled cell diameter, initial volume, etc.) can be evaluated. In some embodiments, example methods for early intermediate stages may involve sampling, seeding, and rapidly expanding (e.g., via gas-permeable rapid expansion (G-Rex)) T cells in the T cell culture sample, and measuring viability before and after these processes.

[0145] Example methods for the early intermediate stage 152 may also include transducing a T cell culture sample to enable T cells to express CAR (box 156). In some embodiments, a vector carrying the CAR expression gene (e.g., a lentiviral vector) may be added to the T cell culture sample or to one or more bags containing the T cell culture sample to enable T cells to express CAR. Various parameters associated with the transduction process in the early intermediate stage may be measured (e.g., vector batch number, vector lot number, syringe batch number used during transduction, vector type, vector titer (IU / mL), multiplicity of infection (MOI) of the target vector added to the T cell culture sample, number of vector vials used during transduction, vector holding time (min), environmental holding time of the syringe during transduction, and / or volume of vector added to the T cell culture sample).

[0146] Following the early intermediate stage 152, the manufacturing process 140 can proceed to intermediate stage 158, which can be correlated with the expansion and growth monitoring of T cell cultures after transduction with CAR. Figure 1BIn the illustrated example method, intermediate stage 158 may include day 6 of the manufacturing process. However, in some embodiments, intermediate stage 158 may include or may include day 5, day 6, or day 7 of the manufacturing process, or a range or value defined by any two of the aforementioned days, such as day 6 of manufacturing process 140. In intermediate stage 158, example method 100 may involve expanding a T cell culture containing CAR-expressing T cells (referred to herein as CAR+ T cells or CAR T cells) (box 160) and monitoring the growth of the T cell culture (box 162). In some embodiments, expansion can be promoted by adding interleukin-2 (IL-2) to drive T cell expansion and differentiation. In some embodiments, growth can be monitored by detecting the presence of glucose or lactate in the T cell culture sample and measuring its properties (e.g., concentration).

[0147] Manufacturing process 140 can further proceed to a late intermediate stage 164, which is associated with the continued expansion and growth monitoring of the T cell culture after the intermediate stage. For example, the T cell culture sample can be further expanded using reagents such as IL-2 (box 166), and growth can be further monitored by measuring the concentration of glucose or lactate in the T cell culture sample (box 168). As shown in Figure 1, late intermediate stage 164 may include day 8 of the manufacturing process. However, in some embodiments, late intermediate stage 164 may include or may include day 7, day 8, day 9, or day 10 of the manufacturing process, or a range or value defined by any two of the aforementioned days, such as days 7-9 of the manufacturing process, preferably day 8 of the manufacturing process.

[0148] Following the expansion and growth of CAR T cells in the intermediate and late intermediate stages, the manufacturing process 140 can proceed to the late stage 170, during which the final product (e.g., CAR+ T cell drug) is harvested and released from the T cell culture sample. Figure 1BAs shown, late stage 170 may include or begin from day 10 of manufacturing process 140. However, in some embodiments, late stage 170 may include or may include day 9, day 10, day 11, day 12, day 13, or day 14 of manufacturing process 140, or a range or value defined by any two of the aforementioned days, such as days 9-12 of manufacturing process 140, days 10-12 of manufacturing process 140, preferably day 10 of manufacturing process 140. In late stage 170, example method 100 may include harvesting T cells from a T cell culture sample and washing the harvested T cells (box 172). In some embodiments, various aspects of the harvested T cells (e.g., viable cell concentration, total cell concentration, viable cell percentage, volume, total cell count and viable cell count, CAR+ expression, etc.) may be measured before and after washing using the techniques described herein (e.g., flow cytometry) (box 174). In some embodiments, T cell growth may be monitored by measuring the concentration of glucose or lactose in the harvested sample. Furthermore, the harvested T cells can be further examined (e.g., for clumps or granules) (box 176). CAR T cells from the harvested T cell sample can be released and formulated into the final product (box 178).

[0149] Throughout the manufacturing process 140, data on various parameters can be obtained, for example, through evaluation, measurement, step completion, reagent addition, etc. (box 180). As will be discussed herein, the production of CAR T drugs may or may not result in manufacturing failure, and this outcome can be predicted based on data on parameters obtained throughout the CAR T drug product manufacturing process.

[0150] IV. Example parameters obtained from each stage of the example CAR-T drug manufacturing process

[0151] Figure 2This is a block diagram illustrating the various stages of an example CAR T drug product manufacturing process, by which parameters are generated for predicting manufacturing failure outcomes of the CAR T drug product (e.g., whether the production of the CAR T drug will lead to manufacturing failure). As previously discussed in the foregoing description, CAR T drug product manufacturing may include, for example, a screening stage 210, a pre-treatment stage 220, a treatment stage 230, and a manufacturing stage 240. Each stage of the CAR T drug production process can be characterized by various parameters that affect or otherwise predict the manufacturing failure outcome 252 of the CAR T manufacturing process. In some respects, manufacturing failure can be indicated by or based on the following: the percentage of T cells as CAR+ T cells in the final product (FP); the percentage of surviving T cells in the final product; the processing time of the T cell culture sample incubated during an intermediate stage of the manufacturing process; the step yield of live CD3+ T cells in the T cell culture sample run on a cell processing platform (e.g., Prodigy); the step yield of live T cells in the T cell culture sample run on a cell processing platform (e.g., Prodigy); the proviral vector copy number; the number of live T cells per G-Rex in the harvested sample of the T cell culture sample during a late stage of the manufacturing process; and the dose of CAR+ T cells per unit mass.

[0152] As previously discussed, in screening phase 210, patients may be screened and / or selected from whom biological samples can be obtained to manufacture patient-specific CAR T drug products. However, patient demographics and medical history can influence the manufacturing failure outcome of the CAR T drug product to be produced. Therefore, the screening phase can be characterized by various parameters (referred to herein as “screening parameters” 212), including parameters related to patient demographics (referred to herein as “patient demographic parameters” 214) and patient medical history (referred to herein as “patient medical history parameter type”). Specifically, this disclosure describes various screening parameters 212 that have been found to be predictive of manufacturing failure outcomes.

[0153] Examples of patient demographic parameters 214 from the screening phase that were found to be predictive include, but are not limited to: age, sex, race, body mass index, ethnicity, and country of origin.

[0154] Examples of patient medical record parameters 216 from the screening phase that were found to be predictive include, but are not limited to: time since initial diagnosis (e.g., the disease the CAR T drug product therapy is intended to treat), measurable disease type (e.g., the disease the CAR T drug product therapy is intended to treat), tumor performance status score at baseline (e.g., Eastern Cooperative Oncology Group (ECOG) performance status at baseline), left ventricular ejection fraction (%), baseline tumor burden category, baseline number of extramedullary plasmacytomas, baseline presence of evaluable bone marrow assessment, baseline International Staging System (ISS) stage, baseline myeloma type, baseline percentage of plasma cells in bone marrow aspiration, baseline percentage of plasma cells in bone marrow, category of percentage of plasma cells in bone marrow aspiration, category of percentage of plasma cells in bone marrow aspiration, previously used alkylating agents (e.g., in the patient's previous treatment), the patient's previous allogeneic transplantation, the patient's previous anthracycline use, the patient's previous number of autologous transplants, the patient's previous autologous transplantation, the patient's previous bortezomib use, the patient's previous cancer-related surgery / procedure, the patient's previous carfilzomib use, and the patient's previous anti-CD40. 38. Antibody use, patient's previous daratumumab use, patient's previous dexamethasone use, patient's previous erlotuzumab use, patient's previous immunomodulatory drug (IMiD) use, patient's previous esatuximab use, patient's previous ixazomib use, patient's previous lenalidomide use, number of previous lines of treatment experienced by the patient, patient's previous opzomib use, patient's previous pabisostat use, patient's previous primary immunodeficiency (PI), patient's previous pomalidomide use, patient's previous prednisone use, patient's previous radiation therapy use, patient's previous corticosteroid use, patient's previous mizotumumab (e.g., TAK-079) use, patient's previous thalidomide use, previous transplantation performed on the patient, patient's refractory status, whether the patient is refractory to penta-based therapy, whether the patient is refractory to alkylating agent-based therapy. Treatment refractory status, whether the patient is refractory to bortezomib-based therapy, whether the patient is refractory to carfilzomib-based therapy, whether the patient is refractory to anti-CD38 antibody-based therapy alone, whether the patient is refractory to daratumumab-based therapy, whether the patient is refractory to erlotuzumab-based therapy, whether the patient is refractory to IMiD-based therapy alone, whether the patient is refractory to exatuximab-based therapy, whether the patient is refractory to ixazomib-based therapy, whether the patient is refractory to lenalidomide-based therapy, whether the patient is refractory to last-line therapy, whether the patient is refractory to pabisostat-based therapy, whether the patient is refractory to pomalidomide-based therapy, whether the patient is refractory to any prior therapy, whether the patient is refractory to mizetuzumab-based therapy (e.g.,Does the patient exhibit refractory behavior to TAK-079 treatment? Is the patient refractory to thalidomide-based treatments? Is the patient refractory to any anti-CD38 antibody-based treatment? Is the patient refractory to any IMiD-based treatment? Is the patient refractory to any prior primary immunodeficiency (PI) treatment?

[0155] Furthermore, in the pre-intubation stage 220 of the CAR T drug manufacturing process, various characteristic laboratory tests can be performed on the patient's biological samples (e.g., urine samples, blood samples, etc.) to determine additional information about the patient and provide additional screening. However, laboratory tests can be used to measure or otherwise provide data on various parameters of the biological samples (referred to herein as "pre-intubation parameters" 222). Specifically, this disclosure describes various parameters in the pre-intubation stage that have been found to be predictive of manufacturing failure outcomes. Examples of such pre-abortion parameters 222 may include, but are not limited to: the total volume of the obtained biological sample, urine protein electrophoresis collection criteria, serum protein electrophoresis collection criteria, whether a urine protein electrophoresis sample was received, the absolute difference (DFLC value) between affected and unaffected serum free light chains in the biological sample, the subject's multiple myeloma (MM) classification (e.g., MM-2 classification), the date and / or time elapsed associated with the pre-abortion laboratory test, the measurement of total protein (e.g., in the subject's urine or blood sample), 24-hour aliquots of urine protein, 24-hour aliquots of urine protein indicating myeloma, detection or measurement of λ free light chains in the biological sample, and the presence of free light chains in the biological sample. The ratio of κ light chains to free λ light chains, measurements of albumin in biological samples (e.g., volume percentage), measurements of α-1 globulin in biological samples (e.g., volume percentage), measurements of α2 globulin in biological samples (e.g., volume percentage), measurements of β globulin in biological samples (e.g., volume percentage), measurements of γ globulin in biological samples (e.g., volume percentage), measurements of monoclonal spike protein 1 in biological samples (e.g., volume percentage), measurements of monoclonal spike protein 2 in biological samples (e.g., volume percentage), measurements based on myeloma immunofixation blot assays using serum protein electrophoresis, total or accumulated amount of M protein in biological samples, and serum volume of biological samples.

[0156] Furthermore, in the single-collection stage 230 of the CAR T drug manufacturing process, flow cytometry and other equipment can be used to test various parameters of the collected sample. These parameters can be associated with: the presence, absence, and / or measurement of various components within the single sample produced by the single-collection process (referred herein to as “Single-Collection Stage - Process Parameters” 236); and the presence, absence, and / or measurement of cell surface markers (referred herein to as “Cell Surface Marker Parameters” 634). Specifically, this disclosure describes how various parameters in the single-collection stage have been found to be predictive of manufacturing failure outcomes.

[0157] Examples of such parameters from the single-collection phase 230 (referred to herein as “single-collection phase parameters” 232, and which may include, but are not limited to, cell surface marker parameters 234, and / or process parameters 236 found to predict manufacturing failure outcomes) include, but are not limited to: the ratio of CD4+ T cells to CD8+ T cells in the single sample; the percentage of CAR-, CD4+ T cells as CAR-, CD4+ terminally differentiated effector memory T cells (TEMRA) in the single sample; the percentage of lymphocytes as CAR- natural killer (NK) T cells in the single sample; the percentage of lymphocytes as CAR- NK cells in the single sample; the concentration of CAR- regulatory T cells in the single sample; the percentage of lymphocytes as CAR- T cells in the single sample; the percentage of regulatory T (Treg) cells as CAR- Treg cells in the single sample; the percentage of CAR-, CD4+ T cells as CAR-, CD4+ Treg cells in the single sample; and the percentage of CAR- Treg cells as CAR- Treg cells in the single sample. T cell percentage; percentage of white blood cells as CAR-monocytes in a single sample; percentage of CAR- and CD4+ T cells as CAR- and naive CD4+ T cells in a single sample; percentage of white blood cells as CAR-neutrophils in a single sample; percentage of CAR- and CD4+ T cells as CAR- and CD4 stem cell memory T cells in a single sample; percentage of CAR- and CD4+ T cells as CAR- and CD4+ T central memory cells in a single sample; percentage of CAR- and CD4+ T cells as CAR- and CD4+ effector memory T cells in a single sample; percentage of CAR- T cells as CAR- and CD4+ T cells in a single sample; percentage of CAR- and CD8+ T cells as CAR- and CD8+ effector memory T cells in a single sample; percentage of CAR- and CD8+ T cells as CAR- and CD8+ stem cell memory T cells in a single sample; percentage of CAR- and CD8+ T cells as CAR- and CD8+ T cells in a single sample. Percentage of T cells; percentage of CAR- and CD8+ T cells as CAR- and CD8+ TEMRA in a single sample; percentage of CAR- and CD8+ T cells as CAR- and CD8+ naive T cells in a single sample; percentage of CAR- T cells as CAR- double-negative T cells in a single sample; percentage of CAR- T cells as CAR- double-positive T cells in a single sample; percentage of CAR- and CD4+ T cells as CD25+, CAR-, and CD4+ T cells in a single sample; percentage of CAR-CD8+ T cells as CD25+, CAR-, and CD8+ T cells in a single sample;Percentage of CAR- and CD4+ T cells as CD27+, CAR-, naive, and CD4+ T cells in a single sample; percentage of CAR- and CD4+ T cells as CD27+, CAR-, and CD4+ central memory T cells in a single sample; percentage of CAR- and CD4+ T cells as CD27+, CAR-, and CD4+ effector memory T cells in a single sample; percentage of CAR- and CD4+ T cells as CD27+, CAR-, and CD4+ stem cell memory T cells in a single sample; percentage of CAR- and CD8+ T cells as CD27+, CAR-, and CD8+ central memory T cells in a single sample; percentage of CAR- and CD8+ T cells as CD27+, CAR-, and CD8+ effector memory T cells in a single sample; percentage of CAR- and CD8+ T cells as CD27+, CAR-, and CD8+ stem cell memory T cells in a single sample; percentage of CAR- and CD8+ T cells as CD27+, CAR-, and CD8+ naive T cells in a single sample. Percentage of T cells; percentage of CAR- and CD4+ T cells as CD27-, CAR-, and CD4+ effector memory T cells in a single sample; percentage of CAR- and CD8+ T cells as CD27-, CAR-, and CD8+ effector memory T cells in a single sample; percentage of CAR- and CD8+ T cells as CD27-, CAR-, and CD8+ TEMRA in a single sample; percentage of CAR- and CD4+ T cells as CD28+, CAR-, and CD4+ T cells in a single sample; percentage of CAR- and CD8+ T cells as CD28+, CAR-, and CD8+ T cells in a single sample; concentration of CD3+, CAR-, CD4+, and CD8- T cells in a single sample; concentration of CD3+, CAR-, CD4-, and CD8+ T cells in a single sample; percentage of CD3+ T cells as CD3+ and CAR- T cells in a single sample; concentration of CD3+ and CAR- T cells in a single sample; CAR- T cells as CD38+ and CAR- Treg cells in a single sample. Percentage of Treg cells; percentage of CAR- and CD4+ T cells as CD38+, CAR-, and CD4+ T cells in a single sample; percentage of CAR- and CD8+ T cells as CD38+, CAR-, and CD8+ T cells in a single sample; percentage of CAR- Treg cells as CD38+, CD39+, and CAR- Treg cells in a single sample; percentage of CAR- Treg cells as CD38-, CD39-, and CAR- Treg cells in a single sample; percentage of CAR- Treg cells as CD39+ CAR- Treg cells in a single sample; concentration of lymphocytes in a single sample;Percentage of white blood cells as monocytes in a single sample; percentage of lymphocytes as NK T cells in a single sample; percentage of lymphocytes as NK cells in a single sample; percentage of white blood cells as neutrophils in a single sample; percentage of CAR- and CD4+ T cells as PD1+, CAR-, and CD4+ T cells in a single sample; percentage of CAR- and CD8+ T cells as PD1+, CAR-, and CD8+ T cells in a single sample; percentage of lymphocytes as T cells in a single sample; percentage of CD4+ T cells as CD4+ Treg cells in a single sample; percentage of T cells as Treg cells in a single sample.

[0158] This disclosure also describes sites in the single-collection phase where tests 30 have been shown to predict manufacturing failure outcomes, including but not limited to manufacturing sites, clinical sites, cryopreservation sites, clinical studies, and procedures.

[0159] Furthermore, as previously discussed, in manufacturing stage 240 of the CAR T drug manufacturing process, selected post-harvest samples can be prepared to become CAR T drug products. CAR T drug products can be provided in a final container. In some embodiments, manufacturing stage 240 can be divided into multiple sub-stages. For example, sub-stages may include an early initial stage, a late initial stage, an early intermediate stage, an intermediate stage, a late intermediate stage, and a late stage closer to and / or including the release of the final product. Manufacturing stage parameters can be obtained from the evaluation of cell culture samples at any of these sub-stages. In some embodiments, parameters may indicate the detection of cell surface markers or the measurement of cells expressing such cell surface markers in one or more sub-stages of the manufacturing process; such parameters may be referred to herein as manufacturing stage cell surface marker parameter 644. In some embodiments, manufacturing stage parameters may describe the characteristics of the process performed in one or more sub-stages of the manufacturing process and / or may describe the contents of cell culture samples in one or more sub-stages of the manufacturing process; such parameters may be referred to herein as manufacturing stage process parameter 246.

[0160] In some implementations, the manufacturing stages and sub-stages can influence whether CAR T drug production will experience manufacturing failure, based on various parameters measured at various points in the manufacturing stages and / or sub-stages on the working product and / or sample (e.g., T cell culture sample, T cell culture population, etc.).

[0161] This disclosure notes that various manufacturing stage parameters 242 have been found to predict manufacturing failure outcomes. Specifically, this disclosure notes that manufacturing stage parameters 242 found in sub-stages closer to the final product are more predictive of manufacturing failure outcomes (e.g., parameters obtained from early intermediate stages, intermediate stages, late intermediate stages, or late stages are more predictive than manufacturing stage parameters obtained during early or late initial stages of the manufacturing process).

[0162] Examples of such manufacturing stage parameters 242 that have been found to be predictive in the early initial stages of the manufacturing process, obtained from T cell culture samples, include, but are not limited to: thawing duration; concentration of live T cells (e.g., after thawing and / or washing, in T cell culture samples undergoing positive or negative selection); percentage of live T cells (e.g., after thawing and / or washing, round or non-round, in T cell culture samples undergoing positive or negative selection); T cell diameter (e.g., after thawing and / or washing, in T cell culture samples undergoing positive or negative selection); single-absorption volume; and volume of anticoagulant (e.g., ACD-A) added to the T cell culture sample (e.g., after thawing). ; Live T cell count before sampling (e.g., after thawing and / or washing, in T cell culture samples that have undergone positive or negative selection); Sample volume (e.g., after thawing and / or washing, in T cell culture samples that have undergone positive or negative selection); Live T cell count after sampling (e.g., after thawing and / or washing, in T cell culture samples that have undergone positive or negative selection); Whether a DNase (e.g., Pulmozyme) was added to the T cell culture sample (e.g., after thawing); Whether an anticoagulant (e.g., ACD-A) was added to the T cell culture sample (e.g., after thawing); Whether online filtration of the T cell culture sample occurred (e.g., after thawing); As CD4+ The percentage of T cells (e.g., after thawing and / or washing, in positively selected T cell culture samples); the percentage of CD8+ T cells (e.g., after thawing and / or washing, in positively selected T cell culture samples); the ratio of CD4+ T cells to CD8+ T cells in T cell culture samples (e.g., after thawing and / or washing, in positively selected T cell culture samples); the percentage of live CD3+ T cells (e.g., after thawing and / or washing, in positively selected T cell culture samples); the percentage of CD3+ T cells (e.g., after thawing and / or washing, in positively selected T cell culture samples); the percentage of CD16+ and / or CD56+ T cells (e.g., after thawing and / or washing, in positively selected T cell culture samples); the percentage of CD19+ T cells (e.g., after thawing and / or washing, in positively selected T cell culture samples); and the percentage of CD14+ T cells. The percentage of T cells (e.g., after thawing and / or washing, in T cell culture samples that have undergone positive selection); the volume of the T cell culture sample (e.g., after washing); the volume of clumps removed from the T cell culture sample (e.g., after washing); the total incubation time for labeled CD4+ T cells and CD8+ T cells.Total enrichment time of labeled CD4+ T cells and CD8+ T cells; number of cycles of running the T cell culture sample through a cell processing platform (e.g., Prodigy); whether CD4+ beads and / or CD8+ beads were manually removed; time spent between thawing the T cell culture sample and running it through a cell processing platform (e.g., Prodigy); time spent running the T cell culture sample through a cell processing platform (e.g., Prodigy); concentration of positively selected live T cells (e.g., after washing); percentage of positively selected live T cells (e.g., after washing); number of live T cells per bag; number of T cell culture bags; number of live T cells used for recovery; volume of T cells added to the bag for culture; volume of culture medium (e.g., TexMACS) added to the bag for culture; volume of T cells seeded in the bag; density of live T cells in the bag; actual number of live T cells seeded; and post-selection hold time.

[0163] Examples of manufacturing process parameters 242 that may be found to be predictive, which can be obtained from the evaluation of T cell culture samples in the early stages of the manufacturing process, may include, but are not limited to: the incubation time of the T cell culture samples in the initial stage; the presence of any pre-activation clumps in the T cell culture samples; the number of pre-activation clumps (e.g., before massage); the alleviating effect of massage on pre-activation clumps; the volume of activation beads (e.g., Transact beads) added to the T cell culture sample bag; the volume of T cell culture samples seeded in the bag; the presence of post-activation clumps; the number of post-activation clumps before massage; the size of post-activation clumps before massage; and the alleviating effect of massage on post-activation clumps.

[0164] Examples of manufacturing process parameters 242 that have been found to be predictive and can be obtained from the evaluation of T cell culture samples at an early intermediate stage of the manufacturing process may include, but are not limited to: the incubation time of the T cell culture samples at an early intermediate stage (e.g., days 1-3 of the manufacturing process); the concentration of live T cells in the T cell culture sample bag; the percentage of cells in the T cell culture sample bag that are live T cells; the volume of T cells in the T cell culture sample bag; the number of cells in the T cell culture sample bag that are live T cells; the presence of any clumps in the T cell culture sample (e.g., before or after mixing); the number of clumps in the T cell culture sample (e.g., before or after mixing); and the size of one or more clumps. (e.g., before or after mixing); effectiveness of mixing clumps in T cell culture samples; mean concentration of live T cells in each population of T cell culture samples; mean percentage of live T cells in each population of T cell culture samples; mean cell diameter of live T cells in each population of T cell culture samples; initial volume of T cell culture samples; number of live T cells before sampling; sample volume of T cell culture; number of live T cells after sampling; number of live T cells available for seeding in T cell culture; number of gas-permeable rapid expansion (G-Rex) seeded, number of live T cells per G-Rex; volume of T cell culture sample transferred from bag to G-Rex; seeded in G-Rex The volume of T cells in A; the carrier batch number of the carrier used to transduce the T cell culture sample; the syringe batch number of the syringe used to transduce the T cell culture sample; the carrier type of the carrier used to transduce the T cell culture sample; the carrier titer of the carrier used to transduce the T cell culture sample; the target multiplicity of infection (MOI) of the carrier used to transduce the T cell culture sample; the target number of carrier vials used to transduce the T cell culture sample; the number of carrier vials used to transduce the T cell culture sample. The vector holding time associated with the transduced T cell culture sample; the syringe environment holding time of the syringe used to transduce the T cell culture sample; the volume of vector added to G-Rex; the incubation time of G-Rex; the number of target live T cells for vesicular stomatitis virus glycoprotein (VSV-g) sampling; the volume of VSV-g sampling; the concentration of live T cells in the VSV-g sample; the actual number of live T cells in the VSV-g sample; the number of particles generated; the percentage of T cell loss; and the number of live T cells used for expansion after VSV-g sampling.

[0165] Examples of manufacturing process parameters 242 that may be found to be predictive and obtainable from the evaluation of T cell culture samples at intermediate stages of the manufacturing process may include, but are not limited to: incubation temperature (e.g., inside or outside an incubator); incubation CO2 saturation (e.g., inside or outside an incubator); total incubation time (e.g., from day 3 to day 6 of the manufacturing process); whether a batch of IL-2 is added to the T cell culture sample; the protein content of IL-2 added to the T cell culture sample; the activity of IL-2; the volume of IL-2 added to the G-Rex of the T cell culture sample; the concentration of lactate or glucose in the T cell culture sample; and the concentration of lactate or glucose in the G-Rex of the T cell culture sample.

[0166] Examples of manufacturing process parameters 242 that may be found to be predictive, which can be obtained from the evaluation of T cell culture samples in the later, intermediate stages of the manufacturing process, may include, but are not limited to: incubation temperature (e.g., inside or outside an incubator); incubation CO2 saturation (e.g., inside or outside an incubator); total incubation time (e.g., from day 6 to day 8 of the manufacturing process); whether a batch of IL-2 is added to the T cell culture sample; the protein content of IL-2 added to the T cell culture sample; the activity of IL-2; the volume of IL-2 added to the G-Rex of the T cell culture sample; the concentration of lactate or glucose in the T cell culture sample; and the concentration of lactate or glucose in the G-Rex of the T cell culture sample.

[0167] Examples of manufacturing process parameters 242 that may be found to be predictive and obtainable from the evaluation of T cell culture samples at late stages of the manufacturing process may include, but are not limited to: incubation temperature (e.g., inside or outside an incubator); incubation CO2 saturation (e.g., inside or outside an incubator); total incubation time (e.g., from day 8 to 10 of the manufacturing process); total expansion or incubation time of the T cell culture sample (e.g., from day 3 to 10 of the manufacturing process); concentration of live CAR+ T cells in the harvested T cell culture sample (e.g., before and after washing); concentration of T cells in the harvested T cell culture sample (e.g., before and after washing); percentage of cells as live CAR+ T cells in the harvested T cell culture sample (e.g., before and after washing); volume of the harvested T cell culture sample (e.g., before and after washing); live CAR+ T cells in the harvested T cell culture sample. The number of T cells (e.g., before and / or after washing, before and after sampling); the number of T cells in the harvested sample of the T cell culture (e.g., before and / or after washing, before and / or after sampling); the concentration of glucose or lactate in the harvested sample of the T cell culture; the incubation temperature for flow cytometry (e.g., inside and outside the incubator); the CO2 saturation for incubation for flow cytometry (e.g., inside and outside the incubator); the harvest and sampling processing time; the time to complete flow cytometry on the harvested sample; the percentage of cells as CAR+ T cells in the harvested sample of the T cell culture (e.g., before and / or after washing); the post-wash dose of CAR+ T cells in the harvested sample of the T cell culture; the number of live target CAR+ T cells per dose; the number of live CAR+ T cells per dose; the concentration of the target formulation for live CAR+ T cells; the number of bags of harvested samples for the T cell culture; the volume of CAR+ T cells per bag of harvested samples of the T cell culture; the CAR+ T cells used to formulate the final product. T cell volume; CS5 volume used to formulate the final product; results of particle inspection of harvested sample bags; presence of clumps in harvested sample bags; time taken to formulate the final product; CS5 contact time; concentration of CS5 live CAR+ T cells in the final formulation; percentage of CS5 live CAR+ T cells in the final formulation; mean concentration of CS5 live CAR+ T cells in the final formulation; mean percentage of CS5 live CAR+ T cells in the final formulation; percentage of dosing accuracy associated with the final formulation; volume of each bag of the final product; color appearance of the final product; appearance of the master container in the final product; BacT / Alert rapid sterility (or other measurement of sterility of the final product); endotoxin concentration in the final product; mycoplasma in the final product; replicating lentivirus (RCL) in the final product; results of VSV-g sampling in early intermediate stages of the manufacturing process;Results of VSV-g sampling in late stages of the manufacturing process; proviral vector copy number (e.g., copies / transduced cells); percentage of cells in the final product as live CAR+ T cells (e.g., after thawing); percentage of cells in the final product as live CD3+ T cells; proviral transduction efficiency (vector copies / cells); percentage of cells in the final product as CD19+ T cells; percentage of cells in the final product as NK CD3-, CD16+, CD56+ T cells; percentage of cells in the final product as CD3+ T cells; CAR of T cells in the final product; concentration of live T cells in the final product; count of live T cells in the final product; dose based on the number of live CAR+ T cells per mass; dose based on the number of live CAR+ T cells (cells); percentage of cells in the final product as live CAR+ T cells. The percentage of T cells; the presence or measurement of interferon (IFN) γ in the final product; the processing time associated with one or more sub-stages of the manufacturing process; the time from flow completion to PFB removal; the processing time associated with CRF; the step yield (%) of total viable T cells based on a cell processing platform (e.g., Prodigy); the viable CD3+ content of T cell culture samples based on a cell processing platform (e.g., Prodigy). T cell step yield (%); culture bag step recovery (%) (i.e., the percentage of T cells recovered from activated T cell culture samples in the early intermediate stage of the manufacturing process relative to T cells in a single sample in the early initial stage of the manufacturing process); LOVO step yield (%); percentage of pre-formulated bulk used to formulate the final product; percentage of recovered dose (e.g., from formulation to thaw); percentage of recovered total viable concentration (target to thaw); percentage of recovered dose (e.g., from target to thaw); population doubling time (PDT) of T cell culture samples measured between the early intermediate and late stages of the manufacturing process; cumulative population doubling level (cPDL) of T cell culture samples measured between the early intermediate and late stages of the manufacturing process; number of final bags used for the final product; actual dose of CAR+ T cells per unit mass of the final product; CAR+ T cells per unit mass of the final product. Calculated T-cell dosage; presence of clumps in the late initial or early intermediate stages of the manufacturing process; completion of the manufacturing process; completion of manufacturing and release testing; whether the final product is an OOS (Out of Stock); type of nonconformity of the final product to specifications; OOS type; OOS / termination comments; whether the final product is controllable (e.g., OOS or termination of the final product due to known causes) or uncontrollable (e.g., OOS or termination of the final product due to unknown causes); whether a batch of the final product was distributed to a patient; whether the distribution of a batch of the final product to a patient was an exceptional distribution (e.g., when the final product was found to be an OOS but it was still considered safe to distribute the batch to a patient);Whether a batch of final product was infused into a patient; whether the batch was terminated during manufacturing; the shift or time category for completing the final product (e.g., first time category, second time category, etc.); the total number of data points in the batch associated with the final product (e.g., the number of features per column); and the visual inspection results of the final product.

[0168] In some implementations, manufacturing stage parameter 242 may include, but is not limited to: the percentage of cells in the final product that are CAR+ T cells; the percentage of live T cells in the final product (e.g., after thawing); the weight of the subject (e.g., the patient); determining whether the VCN is OOS (“VCN OOS”); and determining whether the CAR is OOS (“CAR OOS”).

[0169] It has been found that the aforementioned parameters (e.g., the aforementioned examples of screening parameter 212, pre-single-harvest parameter 222, single-harvest parameter 230, and manufacturing process parameter 242) predict manufacturing failure results based on training a machine learning model, such as regarding... Figure 4 As described. Thus, at least a subset of the data from the aforementioned parameter examples can be obtained at the corresponding stage and sent to one or more computing devices, such as computing device 310, which will be discussed further below. At the computing device, the data can be structured (e.g., vectorized) and applied as one or more input feature vectors 282 to one or more trained machine learning models 280. The trained machine learning model 280 can then output an output feature vector 284 that can indicate a manufacturing failure result 252. Furthermore, as will be discussed regarding... Figure 4 As discussed, if the manufacturing failure outcome is known, data on the aforementioned parameter examples and the known manufacturing failure outcome can be obtained from various stages of the CART drug product manufacturing process. Such data, referred to as reference data or training data, can be used to form input and output feature vectors, respectively, for training a machine learning model. The input and output feature vectors thus formed can be referred to herein as “reference input feature vector” and “reference output feature vector” to indicate their formation from the reference data. The trained machine learning model 280 can then be used to predict unknown manufacturing failure outcomes based on parameters obtained from various stages of the CART drug product manufacturing process. In some embodiments, such as regarding… Figure 4 As described, predicted manufacturing failure outcomes can be used to adjust one or more manufacturing process parameters, for example, to optimize or otherwise correct defects that are associated with or may cause manufacturing failure outcomes of CART drug products.

[0170] For simplicity, parameters obtained for the purpose of predicting manufacturing failure outcomes in the production of CAR T drugs can be referred to as "manufacturing failure parameters".

[0171] V. Example System and Network Environment

[0172] Figure 3 This is a block diagram illustrating an example computer network environment 300 for predicting and optimizing CART drug manufacturing failure outcomes according to a non-limiting embodiment of the present disclosure.

[0173] The computer network environment 300 may include one or more computing devices 310, one or more clinical data systems (clinical data systems 340) storing records of CAR T drug therapy, one or more analysis systems 350, a bioreactor system 370, and one or more electronic health record (EHR) systems 330. Each system of the network environment 300 may communicate with one or more of the remaining systems via a communication network 780.

[0174] One or more computing devices 310 can be used to train and apply machine learning models to predict CAR-T drug manufacturing failure outcomes. The one or more computing devices may include general-purpose computing devices or specialized computing devices (e.g., hardware configured to facilitate numerous iterative processes involving large datasets). For simplicity, as used herein, computing device 310 may refer to any one or a subset of the one or more computing devices 310. In some aspects, while one or a set of computing devices 310 may be configured to train a machine learning model, another or another set of computing devices 310 may be configured to apply the machine learning model to patient-specific data from a target patient. In other aspects, training and application may be performed by the same computing device or the same set of computing devices. In some embodiments, the one or more computing devices may include computing devices that, after applying patient-specific data from a target patient to a machine learning model, optimize manufacturing process parameters for producing CAR-T drug products based on the output of the applied machine learning model.

[0175] In some embodiments, one or more computing devices 310 may include one or more components of the components shown for one or more computing devices 310, such as one or more processors 312, memory 314, link engine 316, network interface 324, feature extraction module 318, training module 320, application module 322, user interface 326, or optimization module 328. One or more processors 312 may include any one or more types of digital circuitry configured to perform operations on data streams, including the functions described in this disclosure. In some aspects, one or more processors 312 may include dedicated processors, such as natural language processors, image processors, etc. Alternatively or additionally, one or more processors 312 may include high-performance processors having functionality (e.g., processor speed, core count, etc.) configured to read large datasets (e.g., millions of gene sequence base pairs) and perform operations on them. Alternatively or additionally, one or more processors 312 may include general-purpose processors. Memory 314 may include any type of long-term memory, short-term memory, volatile memory, non-volatile memory, or other memory, and is not limited to any particular type of memory or any number of memories, or any type of medium on which memory is stored. Memory 314 may store instructions that, when executed by processor 312, may cause one or more computing devices 302 to perform one or more methods discussed herein. Network interface 324 (e.g., wired interface (e.g., electrical interface, RF interface (via coaxial cable), optical interface (via fiber optic)), wireless interface, modem, etc.) may allow computing device 310 to communicate with other systems via communication network 380.

[0176] Linking engine 316 may include software, programs, modules, and / or plugins that, when executed by a processor (such as, but not limited to, one or more processors 312), link or otherwise associate data received from different sources (e.g., electronic health record system 330, clinical data system 340, sample analysis system 350, bioreactor system 370). The data may include reference data for training machine learning models (e.g., multiple reference patients from which CAR-T drug products with known manufacturing failure outcomes are generated) and target data (e.g., target patients from which CAR-T drug products are generated and who are expected to predict manufacturing failure outcomes for CAR-T drug products). Links or associations may be based, for example, on data relating to patients (e.g., reference patients or target patients) or the specific CAR-T drug product manufacturing process. In some aspects, linking engine 316 may rely on metadata within the received data to form links or associations.

[0177] Feature extraction module 318 may include software, programs, modules, and / or plugins that, when executed by a processor (such as, but not limited to, one or more processors 312), cause the processor to generate features from the raw data arranged in feature vectors in a format supported by a machine learning model. Features may include structured and / or quantifiable data representing properties. The raw dataset may include, but is not limited to, natural language text, image data, RNA sequences, DNA sequences, or proteomic sequences. In some embodiments, feature extraction module 318 may be used to vectorize (e.g., generate in quantized data) unstructured data from the dataset into feature vectors (e.g., input feature vectors, output feature vectors, reference input feature vectors, reference output feature vectors, etc.). In some aspects, feature extraction module 318 may rely on a dedicated processor (e.g., a natural language processor, an image processor, a high-performance processor for gene sequencing, etc.) to extract features from the raw data.

[0178] Training module 320 may include software, programs, modules, and / or plugins that, when executed by a processor (such as, but not limited to, one or more processors 312), enable the processor to train a machine learning model using, for example, a training dataset (e.g., for supervised learning). In some aspects, training module 320 may be used to correlate input feature vectors (e.g., reference input feature vectors) with output feature vectors (e.g., reference output feature vectors). The input and output feature vectors may be generated by feature extraction module 318, or may be formed based on features extracted by feature extraction module 318 from the original dataset. As used herein, reference input feature vector or reference output feature vector may refer, respectively, to the input and output feature vectors specifically generated from the training dataset for the purpose of training the machine learning model. The training dataset may include raw data about multiple patients (referred to herein as reference patients), manufacturing failure results of CAR T drug products generated from reference patients, and process parameters associated with the production of such CAR T drug products. Furthermore, training module 320 may correlate reference input feature vectors with reference output feature vectors along the machine learning model. For example, in a neural network, the training module may take a reference input feature vector along the input layer and a reference output feature vector along the output layer, where the input and output layers are separated by a predetermined number of hidden layers. Training a machine learning model may involve performing an iterative process to determine the relationship between the input and output feature vectors. This relationship can be represented as a set of weights applied to the parameters represented by the input feature vectors, indicating the ability of a particular parameter to predict the output feature vector.

[0179] Application module 322 may include software, programs, modules, and / or plugins that, when executed by a processor (such as, but not limited to, one or more processors 312), cause the processor to apply an input feature vector to a trained machine learning model to generate an output feature vector. For example, application module 322 may be used to apply a trained machine learning model to generate an output feature vector that predicts the values ​​of a set of output parameters corresponding to one or more manufacturing qualities of the CAR-T drug intended to be manufactured. The input feature vector may correspond to quantitative data of parameters for a patient (referred to herein as the “target patient”) who is expected to know or predict the outcome of CAR-T drug manufacturing failure.

[0180] User interface 326 may include, for example, a graphical user interface, an input / output module, a keyboard or keypad, a mouse, a display, and other functionalities that allow data input and output.

[0181] The optimization module 328 may include software, programs, modules, and / or plugins that, when executed by a processor (such as, but not limited to, one or more processors 312), enable the processor to identify or recommend optimizations for one or more manufacturing process parameters of the CAR-T drug product based on predictions of manufacturing failure outcomes. In some aspects, based on the identification of optimizations, the computing device 310 may implement the optimization, for example, by transmitting commands to appropriate devices controlling the manufacturing process parameters via a communication network 380 (e.g., a bioreactor system 370).

[0182] In some implementations, network environment 300 may include one or more electronic health record systems 330 that facilitate the import of patient-specific data and the storage of patient-specific electronic health records (EHRs) in a database (e.g., patient health record database 332). In some aspects, electronic health record system 330 may also include an encryption module 336. Encryption unit 339 may include applications, programs, software, code, or plugins to implement methods for encrypting and decrypting electronically protected health information. The encryption and decryption protocols implemented through encryption unit 339 may comply with regulations (e.g., HIPAA). For example, computing device 310 may establish communication with the electronic health record system (e.g., via a communication network and a network interface). Electronic health record system 330 may also include a network interface 334, similar to network interface 334, allowing electronic health record system 330 to communicate with or receive communication from other systems and devices in network environment 300 via communication network 380. For example, computing device 310 can then receive patient-specific health data from the patient health record database 332 of electronic health record system 330, and extract various feature parameters from the patient-specific health data, such as parameters belonging to patient demographic parameter types and patient medical history parameter types. These parameters can be integrated into a feature vector for training and applying machine learning models to predict and optimize CAR T drug product manufacturing failure outcomes.

[0183] Network environment 300 may include a clinical data system 340 that stores records of CAR-T drug therapies (e.g., known manufacturing failures of manufactured or incompletely manufactured CAR-T drug products, and parameters at various stages of CAR-T drug product production). Clinical data system 340 may include an electronic data management system for storing and accessing clinical data, for example, as it relates to parameters affecting the manufacture of CAR-T drugs and the manufacturing failures of the corresponding CAR-T drug products. Clinical data may comply with applicable regulatory requirements. Such clinical data regarding the production and manufacturing failures of CAR-T drug products, and the parameters affecting said CAR-T drug products, may be stored in a database (e.g., CAR-T database 342). In some aspects, clinical data system 340 may include a query engine 348, which may include software, programs, modules, and / or plugins that allow (e.g., computing device 310) a user to search for clinical data from stored patient-specific EDC data and receive query results (e.g., answers to questions, search results, location of specific clinical data or files, etc.). Furthermore, patient-specific information and other sensitive information related to clinical data can be de-identified and / or encrypted (e.g., via encryption module 346). In some aspects, encryption module 346 can be used to decrypt or otherwise link various parameters related to the manufacture of CAR T drug products stored in the CAR T database to patient-specific parameters (e.g., patient demographic parameter types or patient medical history parameter types) that can be retrieved from the patient health record database 332. A network interface 344, similar to network interface 324, can allow the clinical data system 340 to communicate with or receive communication from other systems and devices in the network environment 300 via communication network 380. For example, computing device 310 can facilitate the linking of parameters obtained from the CAR T database 342 with parameters from the patient health record database 332 via network interfaces 324, 344, and 334.

[0184] The network environment 300 may also include one or more sample analysis systems 350. Sample analysis systems 350 may include or refer to systems, devices, and instruments for receiving data related to the manufacture of CAR-T drugs based on the analysis of biological samples (e.g., single samples) from a target patient. For example, a sample of single-sample starting material from a target patient may be obtained and then analyzed under one or more sample analysis systems 350 for cell characterization (e.g., single-cell RNA sequencing [scRNA-seq], cell indexing of transcriptome and epitope sequencing [CITE-seq], and flow cytometry). In some embodiments, sample analysis systems 350 may generate (e.g., after measurements are obtained from biological samples from a target patient via various modalities) data from which parameters related to CD markers, transcriptomics markers, patient laboratory results, and cellular components can be obtained. For example, sample analysis system 350 may include, but is not limited to, flow cytometry system 354 (e.g., for acquiring data related to CD markers and cellular components), single-cell sequencing system 3756 (e.g., for acquiring transcriptomics markers), and laboratory instruments (e.g., for acquiring patient laboratory results, apheresis markers, etc.). In some embodiments, data to be obtained from such analysis system 350 can be requested, viewed, filtered, and / or associated via one or more user interfaces 352. In some embodiments, sample analysis system 350 may also include one or more network interfaces 358, similar to network interface 324, allowing sample analysis system 350 to communicate with or receive communication from other systems and devices in network environment 300 via communication network 380. For example, computing device 310 can receive data related to the manufacture of CAR-T drugs based on the analysis of biological samples from target patients, and extract various feature parameters from this data, such as parameters belonging to cell surface marker parameters 234 in the single-absorption stage, pre-absorption parameters 622, single-absorption process parameters 236, manufacturing stage cell surface marker parameters 244, and manufacturing stage process parameters 246. These parameters can be integrated into a feature vector for training and applying machine learning models to predict and optimize CAR-T drug manufacturing failure outcomes.

[0185] In some embodiments, network environment 300 may include bioreactor system 370. Bioreactor system 370 may include devices (e.g., containers) or systems supporting an environment for manufacturing CAR T drug products, having the functionality to adjust various manufacturing process parameters 374 (e.g., manufacturing process parameter 242) via user interface 372. For example, bioreactor system 370 may include an active biological environment for culturing CAR T cell samples having desired parameters from stages of the CAR T drug production process prior to the manufacturing stage (e.g., screening stage, single-cell harvesting stage, etc.); however, as the selected CAR T cell samples are cultured and undergo other manufacturing processes, various parameters (manufacturing process parameters) can be adjusted during the manufacturing stage. Examples of such manufacturing process parameters include, but are not limited to, those shown in Appendix A. In some embodiments, bioreactor system 370 may also include network interface 376, similar to network interface 324, allowing bioreactor system 370 to communicate with or receive communication from other systems and devices in network environment 300 via communication network 380. For example, computing device 310 can receive current manufacturing process parameters used to manufacture a batch or group of CAR T drug products to predict manufacturing failure outcomes. Furthermore, based on the prediction, computing device 310 can transmit signals to bioreactor system 370 to change or adjust manufacturing process parameters, thereby achieving better results in controlling CAR T drug product manufacturing failures.

[0186] VI. Training and applying machine learning models to predict manufacturing failures of CAR-T drug products

[0187] Figure 4 This is a block diagram illustrating an example process 400 for predicting and optimizing CAR T drug manufacturing failure outcomes according to a non-limiting embodiment of the present disclosure. As shown, process 400 includes a plurality of enumerated steps, but aspects of process 400 may include additional steps before, after, and between the enumerated steps. In some embodiments, one or more of the enumerated steps may be omitted or performed in a different order. Process 400, which may include a training phase 400A and an application phase 400B, may be executed by one or more computing devices (e.g., such as, but not limited to, computing device 310). For example, process 400 may be executed by one or more processors (e.g., but not limited to, one or more processors 312) based on computer-executable or machine-readable instructions stored in the memory (e.g., but not limited to, memory 314) of one or more computing devices. In some aspects, training phase 400A may be executed by a separate or different computing device from the computing device executing application phase 400B, for example, to save computing resources and / or bandwidth.

[0188] In various implementations, training phase 400A may involve receiving reference data (box 402) from a reference CAR-T drug manufactured from a reference patient. The reference data may correspond to at least a subset of the aforementioned parameters from various stages of the CAR-T drug product manufacturing process (e.g., screening parameter 212 from screening phase 210, pre-abortion parameter 222 from pre-abortion phase 220, ablation parameter 2632 from ablation phase 230, and manufacturing phase parameter 242 from manufacturing phase 240), as well as known manufacturing failure results of the CAR-T drug product resulting from the CAR-T drug product manufacturing process. As previously discussed, manufacturing failure results of the CAR-T drug product can be a prediction of whether the production of the CAR-T drug product will lead to manufacturing failure.

[0189] Machine learning models trained in the training phase 400A can be specifically trained to predict manufacturing failure outcomes in CAR T drug production. In some respects, the reference data can be unstructured, and processors (e.g., natural language processors, image processors, dedicated gene sequencing processors, etc.) can process, translate, decrypt, decode, and / or quantize unstructured data into a vectorizable format.

[0190] The machine learning model trained in training phase 400A (which can then be applied in application phase 400B) may itself include any number of machine learning models and / or algorithms. For example, the machine learning model may include, but is not limited to, decision trees, deep learning models, neural networks, linear discriminant analysis models, quadratic discriminant analysis models, support vector machines, random forest algorithms, nearest neighbor algorithms (e.g., k-nearest neighbor algorithms), combined discriminant analysis models, k-means clustering algorithms, unsupervised models, multivariate regression models, penalized multivariate regression models, or at least one of another type of model. In various implementations, the machine learning model may include any number or combination of the models or algorithms described above.

[0191] In some implementations, reference data may be received from various sources, such as other computing systems, like electronic health record system 330, clinical data management system 340, sample analysis system 350, or bioreactor system 370 in a network environment, or databases and / or repositories, such as patient health record database 332 or clinical databases for CAR T drug therapy (“CAR T database” 342). In some aspects, the received reference data may be appropriately linked together, for example, corresponding to a reference patient or a reference CAR T drug product manufactured by a reference patient (e.g., using a biological sample from a reference patient). Links may be formed using, for example, a link engine 316 of computing device 310.

[0192] At box 404, the computing device may vectorize the reference data to generate a reference input feature vector and a reference output feature vector. In some embodiments, each reference input feature vector may be associated with a corresponding reference patient from whom a CAR T drug product is manufactured (e.g., using a biological sample from a reference patient), and each reference output feature vector may be associated with a manufacturing failure reference from the corresponding reference patient in the production of the reference CAR T drug. Thus, each reference input feature vector may be paired with a corresponding reference output feature vector. In some aspects, vectorization may involve the feature extraction module 318 of the computing device 310 compressing the unstructured data received in box 402 such that different inputs to a given parameter can be aggregated into a composite input to that parameter. Vectorization may yield a reference input feature vector comprising a composite data input of each of a plurality of input parameters. In some aspects, the plurality of input parameters may include at least a subset of the input parameters shown in Appendix A. In some embodiments, redundant or unnecessary parameters may be removed, for example, through dimensionality reduction of the reference input feature vector. Dimensionality reduction can improve the speed of the trained machine learning model or can be used to overcome the problem of overfitting.

[0193] At box 406, the computing device can associate reference input feature vectors with reference output feature vectors on a machine learning model. For example, for each pair of reference input feature vectors (representing input parameters from the corresponding CAR-T drug manufacturing process from the corresponding reference patient) and reference output feature vectors (representing manufacturing failures of the resulting corresponding CAR-T drug product), input feature vectors can be input within the machine learning model, where each input parameter represented by the reference input feature vectors has randomized or initialized weights and / or biases. The machine learning model can be constructed to allow iterative adjustment of the weights through an error minimization process when determining the relationship between the reference input feature vectors and the corresponding reference output feature vectors. For example, for a neural network, the input feature vectors can be aligned along the input layer of the neural network, while the output feature vectors can be aligned along the output layer, which is separated from the input layer by one or more hidden layers. Each layer can include one or more nodes, which can involve activation functions. The aforementioned weights can be assigned to the respective nodes of the input layer.

[0194] At box 408, the computing device can train a machine learning model to iteratively minimize errors within a predetermined threshold. For example, the training module 320 of computing device 310 can train the machine learning model by iteratively minimizing the error in determining the relationship between parameters represented by a reference input feature vector and a reference output feature vector. This relationship can be represented by a set of weights assigned to the parameters represented by the input feature vector. An initial set of weights for the parameters of the input feature vector can be tested to determine how correctly the weights indicate the significance of the various parameters in their ability to predict manufacturing failure results represented by the reference output feature vector. Each prediction can be quantitative and / or binary data compared to known data of one or more manufacturing failure results. If the difference is not below a predetermined threshold or tolerance, an iterative process involving a new set of weights for the parameters occurs. Training involves determining the correct set of weights for the input parameters of the input feature vector. Each weight can indicate the significance of the parameter associated with the weight in the parameter's ability to predict the manufacturing failure result indicated by the output feature vector.

[0195] At box 410, the computing device can output a trained machine learning model containing a final weight set indicating the relationship between the input parameters and the manufacturing failure outcome of the reference CAR-T drug. For example, the trained machine learning model can be stored in memory (e.g., memory 314 of computing device 310), or can otherwise be accessed by the computing device performing the training or another computing device. Alternatively, the trained machine learning model can be stored on a local or remote server, accessible by the computing device performing application phase 400B.

[0196] In various implementations, application phase 400B may involve a computing device with a processor (e.g., computing device 310 with memory 314) receiving unstructured target data (box 412) of a target patient intended to generate a CAR-T drug from there. The target patient can be distinguished from a reference patient because the target patient is the intended recipient of the CAR-T drug product, which is optimized, or whose unknown manufacturing failure outcome is otherwise predicted using the systems and methods presented herein. As used herein, a reference patient may refer to a patient who may already know the manufacturing failure outcome of a CAR-T drug product obtained using a reference patient. Thus, the reference patient, the manufacturing process of the CAR-T drug product generated from the reference patient, and the manufacturing failure outcome of the CAR-T drug product are applicable to training phase 400A, while the target patient, the manufacturing process of the CAR-T drug product generated from the target patient, and the manufacturing failure outcome of the CAR-T drug production to be predicted are applicable to application phase 400B.

[0197] In some embodiments, the target data may correspond to at least a subset of the aforementioned parameters from various stages of the manufacturing process of the CAR T drug product (e.g., screening parameter 212 from screening stage 210, pre-abortion parameter 222 from pre-abortion stage 220, flow cytometry and site testing parameter 232 from flow cytometry and site testing stage 230, and manufacturing parameter 242 from manufacturing stage 244), for which manufacturing failure results are unknown or expected to be predicted. In some embodiments, the subset of the aforementioned parameters may include those parameters described herein as having significant predictive value for the expected manufacturing failure results. For example, as will be discussed herein, for manufacturing failure results, this disclosure describes key parameters that have been found to have significant predictive value for determining the manufacturing failure results of a given CAR T drug product.

[0198] As used in this paper, the data received for predicting manufacturing failure outcomes of CAR T drug products in target patients can be referred to as target data, to distinguish it from the data received for reference patients, used to train the machine learning model. The latter data received for training can be referred to as reference data in this paper.

[0199] In some respects, the target data can be unstructured, and processors (e.g., natural language processors, image processors, dedicated gene sequencing processors, etc.) can process, translate, decrypt, decode, and / or quantize unstructured data into a format that can be vectorized.

[0200] In some implementations, the target data may be received from various sources, such as other computing systems, like electronic health record system 330, clinical data management system 340, sample analysis system 350, or bioreactor system 370 in a network environment, or databases and / or repositories, such as patient health record database 732 or clinical databases for CAR T cell therapy (“CAR T database” 342). In some aspects, the received target data may be appropriately linked together, for example, corresponding to a target patient, or corresponding to various stages of the manufacturing process of a CAR T drug product produced using the target patient (e.g., using a biological sample from the target patient). Links may be formed using, for example, a link engine 316 of computing device 310.

[0201] At box 414, the computing device may vectorize the target data to generate an input feature vector. In some aspects, vectorization may involve the feature extraction module 318 of the computing device 310 compressing the unstructured data received in box 412, such that different inputs to a given parameter can be aggregated into a composite input to that parameter. Vectorization can yield an input feature vector comprising a composite data input of each of a plurality of input parameters. In some aspects, the plurality of input parameters may include at least a subset of the input parameters shown in Appendix A. For example, the subset may include parameters that this disclosure has found specifically to predict the manufacturing failure outcome of a desired CAR T drug product, as will be discussed with respect to the following figures. In some embodiments, redundant or unnecessary parameters may be removed, such as dimensionality reduction for reference input feature vectors. Dimensionality reduction can improve the speed of the trained machine learning model or can be used to overcome the problem of overfitting.

[0202] At box 416, the computing device may apply the input feature vector to a trained machine learning model (e.g., from box 410) to generate an output feature vector of data predicting one or more manufacturing failure outcomes for a CAR-T drug. As previously discussed, the trained machine learning model may have a stored set of weights that instruct each of a plurality of parameters on its ability to predict whether a CAR-T drug product will experience manufacturing failure. The plurality of parameters may include, contain, and / or correspond to the parameters represented by the input feature vector. Thus, the input feature vector can be associated with the set of weights in the trained machine learning model to generate an output feature vector predicting whether a CAR-T drug product will experience manufacturing failure.

[0203] At box 418, the computing device can assess whether the predicted manufacturing failure outcome indicates a manufacturing failure in the production of a CAR T drug. As discussed herein, regulatory agencies have established specifications or quality standards for quality control of manufactured drugs, such as CAR T drug products. A manufactured drug or its batch may meet such quality control standards, i.e., the drug may be “compliant,” or it may not meet such drug product standards, i.e., the drug may be “non-compliant (OOS).” Furthermore, such standards may include individual standards or subsets of standards for ensuring that CAR T drug production is not a manufacturing failure. Specifications for which assessments can be performed can be stored in memory 314 of the computing device 310 and can be updated periodically (e.g., based on updates to the specifications).

[0204] At box 420, if the predicted manufacturing failure outcome for the CAR-T drug production leads to manufacturing failure, the computing device can adjust or change one or more manufacturing process parameters associated with the production of the CAR-T drug product. For example, the computing device can output (e.g., via user interface 326) an indication that the CAR-T drug production will lead to manufacturing failure (e.g., based on specification standards) and can prompt a user (e.g., the manufacturer of the CAR-T drug product, the target patient, a healthcare professional associated with the target patient, etc.) to change or adjust one or more manufacturing process parameters. Examples of manufacturing process parameters include those described under “Manufacturing Process Parameters” in Appendix A. Alternatively or additionally, the computing device can automatically cause the equipment or apparatus performing the manufacturing to adjust the manufacturing process parameters. For example, the computing device 310 can transmit signals to the bioreactor system 370 via communication network 380 to change or adjust one or more manufacturing process parameters. In some aspects, the process of changing or adjusting manufacturing process parameters can be performed using programs, software, or logic stored in optimization module 328 of the computing device 310. For example, defects leading to the predicted manufacturing failure outcome can be identified, and optimization module 328 can be made to search for manufacturing process parameters that will mitigate the defects. In some implementations, after changing one or more manufacturing process parameters or generating recommendations for the changes, the computing device can use a modified input feature vector based on one or more of the changed manufacturing process parameters to repeat one or more steps of stage 400B. Furthermore, the application stage can be repeated until the predicted manufacturing failure outcome is that there will be no manufacturing failure.

[0205] At box 422, if the predicted manufacturing failure outcome is that there is no manufacturing failure in CAR T drug production, the computing device can induce the manufacture of the CAR T drug product based on the current set of manufacturing process parameters. For example, computing device 310 can display (e.g., via user interface 326) a prediction that the CAR T drug product being produced will not lead to manufacturing failure or will have a sufficiently low probability of leading to manufacturing failure. Alternatively or additionally, the computing device can transmit a signal that causes the equipment configured to manufacture the CAR T drug product (e.g., bioreactor system 370) to continue manufacturing.

[0206] VII. Exemplary machine learning models that may be used in the implementation schemes described herein :

[0207] In some implementations, example machine learning models trained (e.g., based on a reference dataset) and applied to predict manufacturing failure outcomes can include decision trees. For example, decision trees (such as classification decision trees) can be used to predict manufacturing failure outcomes for CART drug products characterized by binary outcomes (e.g., whether a manufacturing failure exists). As another example, decision trees (such as regression decision trees) can be used to predict manufacturing failure outcomes for CART drug products characterized by continuous values ​​(e.g., the probability of achieving a manufacturing failure, etc.). Figures 5A to 5B The references to decision trees and their accompanying descriptions are for illustrative purposes only, illustrating example machine learning models used in the implementation scheme, and are not intended to limit the machine learning models used in the implementation scheme to decision trees in any way. For example, other machine learning models may be implemented alternatively or otherwise in the implementation schemes described herein. Such machine learning models may include, but are not limited to, parametric models, nonparametric models, deep learning models, neural networks, linear discriminant analysis models, quadratic discriminant analysis models, support vector machines, random forest algorithms, nearest neighbor algorithms, ensemble discriminant analysis models, k-means clustering algorithms, supervised models, unsupervised models, logistic regression models, multivariate regression models, penalized multivariate regression models, or other types of models.

[0208] Figure 5A This is a diagram illustrating an example decision tree modeling of parameter thresholds for predicting manufacturing failure outcomes of CAR T drug products according to a non-limiting embodiment of this disclosure. In this example, each data point (indicated as either a circle or a star) is based on the values ​​of two input parameters. As previously discussed, such input parameters can be two parameters selected from any of the foregoing examples of screening parameter 212, pre-abortion parameter 222, ablation parameter 230, and manufacturing process parameter 242. Therefore, Figure 5A The two parameters shown as A and B can include, for example, the percentage of thawed live CAR+ T cells in the final product and the concentration of CAR+ T cells in the final product, respectively. A given data point can be represented as an input feature vector including the corresponding values ​​of parameters A and B (representing the two input features, respectively). A given data point can also be associated with an output feature vector, which can include manufacturing failure results for the CAR T drug product, i.e., result 1 (shown as a circle) or result 2 (shown as a star). The decision tree model can be used to determine a threshold 532 for the value of parameter A, for which data points satisfying the threshold 532 may be associated with a given result. For example, as... Figure 5AAs shown, data points with parameter A values ​​below the threshold 532 tend to be associated with manufacturing failure outcome 2, while data points with parameter A values ​​above the threshold 532 tend to be associated with manufacturing failure outcome 1. The decision tree model can also be used to determine a threshold 534 for parameter B, for which data points satisfying the threshold 534 may be associated with a given outcome. For example, as... Figure 5A As shown, data points where parameter B's value is below threshold 534 tend to be associated with manufacturing failure result 2, while data points where parameter B's value is above threshold 534 tend to be associated with manufacturing failure result 1. In some implementations, the thresholds can be adjusted to improve accuracy. For example, thresholds 532 and 534 can be adjusted to the threshold range of each of the values ​​of parameter A and parameter B, respectively. Therefore, the combined threshold range is shown in box 536. Figure 5A As shown, the data points in box 836 more accurately predict a specific result, namely manufacturing failure result 2, based on the values ​​of the input features (i.e., the values ​​of parameters A and B) of the data points falling within the threshold range specified in box 536.

[0209] although Figure 5A Two input parameters and a manufacturing failure result including two discrete outcomes are shown, but it is conceivable (based on the implementation described herein) that there may be a large number of input parameters used to train machine learning models, such as decision tree models, to predict manufacturing failure results. Although Figure 5A The example shown uses two input parameters for illustrative purposes, but it is conceivable that the use of a large number of input feature parameters (such as those in the implementations described herein) may not be possible via methods such as Figure 5A The diagram is used to depict this. In some aspects, the training and application of models based on a large number of input parameters can rely on computing devices equipped with processors to process large datasets characterized by a large number of dimensions of the corresponding input parameters. Furthermore, it is conceivable (based on the implementation described herein) that manufacturing failure outcomes of CAR T drug products may not necessarily be characterized by two outcomes. For example, manufacturing failure outcomes of CAR T drug products can be characterized by continuous or semi-continuous outcomes (e.g., to indicate the probability of manufacturing failure).

[0210] Figure 5B This illustrates the methods used to train decision tree models (such as, but not limited to, those for training decision tree models). Figure 5A The example shown is a block diagram of an example process for predicting manufacturing failure outcomes of CAR T drug products. In at least one embodiment, training of the decision tree model can be performed by a computing device having a processor configured to perform one or more of the following steps (e.g., computing device 310 such as, but not limited to, having processor 312). Training can involve a dataset comprising multiple data points (e.g., such as, but not limited to, in...). Figure 4The reference data received in box 402 of training phase 400A). For example, each data point can be a set of values ​​for various input feature parameters obtained in the development of a CAR T drug product, where one or more values ​​indicate known drug manufacturing failures of the CAR T drug product. For each input feature parameter being used (e.g., Figure 5A Given parameters A and B, the computing device can determine an initial candidate threshold to split the data points (box 540). In some aspects, the candidate threshold can be a randomized value. In some aspects, the candidate threshold can be based on the statistical characteristics of each data point (e.g., the maximum, minimum, or average of the input feature parameters). The candidate threshold can then be evaluated to determine its accuracy in splitting the data points based on known outcomes of the data points (box 542). For example, for an output characterized by two discrete outcomes, the number of data points belonging to a certain outcome (e.g., outcome 1) can be calculated for each side of the threshold. A measure of the performance of the candidate threshold can be based on maximizing the number of data points associated with a given outcome on one side of the threshold and minimizing the number of data points associated with a given outcome on the other side of the threshold.

[0211] The aforementioned steps of iteratively identifying candidate thresholds for a given input parameter and evaluating their accuracy in splitting data points based on known results can be repeated until convergence is achieved, i.e., a candidate threshold is found to optimally split the data points based on its results (e.g., compared to other candidate thresholds) (box 544). This convergence can be determined via an error minimization method, where the ability of a given candidate threshold to split data points based on its results is evaluated, and the error in doing so is measured. Convergence is achieved when the error is minimized to a preset tolerance level. Alternatively or additionally, convergence is achieved when it is found that a candidate threshold splits data points to a significantly better extent than previously tested candidate thresholds based on the results of the data points. Thus, in some embodiments, optimizing a candidate threshold may involve determining whether the distribution of data points on either side of a candidate threshold is better than that of a previously best candidate. Once an optimal candidate threshold is found, it can be identified or designated as a threshold for the input parameter (box 546). The aforementioned process can be repeated for other input feature parameters until thresholds for all input parameters are determined (box 548). Furthermore, the determined threshold for each of the multiple input feature parameters can thus serve as a weight or relation for predicting manufacturing failure outcomes of CAR T drug products. Therefore, the determined threshold can be stored as part of the output of a trained decision learning model for predicting manufacturing failure outcomes of CAR T drug products (box 550).

[0212] VIII. Predicting whether the production of patient-specific CAR-T drug products targeting specific patients will lead to manufacturing failure.

[0213] Figure 6AThis is a block diagram illustrating an example method 600 for predicting whether the production of a patient-specific CAR T drug product for a target patient (e.g., using a set of parameters (manufacturing failure parameters) obtained from various stages) will lead to manufacturing failure, according to a non-limiting embodiment of this disclosure. Furthermore, Figure 6B Figure 6D shows a table of example parameters described in this disclosure that are significant for its ability to predict whether the production of a patient-specific CAR T drug product for a target patient will result in manufacturing failure. Method 600 can be performed by one or more computing devices (e.g., such as, but not limited to, one or more computing devices 310). For example, method 600 can be performed by one or more processors (such as, but not limited to, one or more processors 312) based on computer-executable or machine-readable instructions stored in the memory (such as, but not limited to, memory 314) of one or more computing devices. As shown, method 600 includes a plurality of enumerated steps, but aspects of method 600 may include additional steps before, after, and between the enumerated steps. In some embodiments, one or more of the enumerated steps may be omitted or performed in a different order.

[0214] In various embodiments, method 600 may include receiving quantitative data of a set of manufacturing failure parameters (box 602). This set of manufacturing failure parameters may include manufacturing failure parameters selected from Table 1, such as... Figure 6B As shown. Each manufacturing failure parameter belongs to Table 1 (e.g. Figure 6B One of the parameter types outlined in (shown).

[0215] In some implementations, the manufacturing failure parameters, as outlined in Table 1, can be ordered according to the significance of predicting whether the production of a patient-specific CAR T drug product for a target patient will result in manufacturing failure (i.e., starting with the highest significance at the top and ending with the lowest significance at the bottom). When using a trained machine learning model to predict whether the production of a patient-specific CAR T drug product for a target patient will result in manufacturing failure, manufacturing failure parameters with higher significance can be assigned a higher weight than other manufacturing failure parameters.

[0216] In some embodiments, manufacturing failure can be characterized by one or more attributes, including but not limited to: whether the cause of failure of the first manufacturing attempt is considered controllable or uncontrollable; the percentage of cells as CAR+ T cells in the final product; the percentage of cells as live T cells in the thawed final product; the processing time between incubation initiation and incubation completion during intermediate stages of the manufacturing process; the step yield (%) of live CD3+ T cells in the final product (e.g., as determined via a cell processing platform (e.g., Prodigy); the step yield (%) of total live T cells in the final product (e.g., as determined via a cell processing platform (e.g., Prodigy); the proviral vector copy number (copy / transduced cells) of the final product; the number of washed live T cells harvested per G-Rex during late stages of the manufacturing process; and the actual dose of CAR+ T cells per unit mass (e.g., CAR+ T cells / kg). In some embodiments, attributes of manufacturing failure may include attributes indicating whether a patient-specific CAR T drug product is OOS (Out of Service).

[0217] In some implementations, the set of manufacturing failure parameters may include a subset of the parameters listed in Appendix A. For example, in at least one embodiment, the set of manufacturing failure parameters may include one or more of the following parameters: the percentage of pre-washed CAR+ T cells harvested from intermediate to late stages of the manufacturing process; the concentration of lactate or glucose in T cell culture samples from late intermediate stages of the manufacturing process; the concentration of lactate or glucose in T cell culture samples from intermediate stages of the manufacturing process; the ratio of CD4+ T cells to CD8+ T cells in T cell culture samples from the initial stage of the manufacturing process; the volume of vector added to T cell culture samples during early intermediate stages of the manufacturing process; the multiplicity of infection (MOI) of the vector added to T cell culture samples during early intermediate stages of the manufacturing process; the average percentage of live T cells from each population from early intermediate stages of the manufacturing process; the average concentration of live T cells from each population from early intermediate stages of the manufacturing process; the concentration of lymphocytes in a single sample prior to the manufacturing process; the percentage of cells as live T cells in T cell culture samples from the initial stage of the manufacturing process; whether the patient is refractory to pomalidomide treatment; the patient's sex; the patient's age; the patient's body mass index (BMI); and the number of previous lines of treatment. In one implementation, the aforementioned list of parameters is arranged in order of significance of whether the production of a patient-specific CAR T drug product for a target patient would result in manufacturing failure, as predicted using a trained machine learning model (i.e., starting with the highest significance at the top and ending with the lowest significance at the bottom). Therefore, when using a trained machine learning model to predict whether the production of a patient-specific CAR T drug product for a target patient would result in manufacturing failure, parameters with higher significance can be assigned higher weights than other manufacturing failure parameters (e.g., as discussed in subsequent steps).

[0218] In some implementations, this set of manufacturing failure parameters includes a set of screening parameters selected from Table 1A (as shown in Figure 6C). Furthermore, the screening parameters in Table 1A (as shown in Figure 6C) can be ordered according to the significance of predicting whether the production of a patient-specific CAR T drug product will lead to manufacturing failure (i.e., starting with the highest significance at the top and ending with the lowest significance at the bottom). When using a trained machine learning model to predict whether the production of a patient-specific CAR T drug product for a target patient will lead to manufacturing failure, screening parameters with higher significance can be assigned a higher weight than other screening parameters.

[0219] In some implementations, this set of manufacturing failure parameters includes a set of manufacturing stage parameters as outlined in Table 1B (as shown in Figure 6C). Furthermore, the manufacturing stage parameters in Table 1B can be ordered according to the significance of their prediction, using a trained machine learning model, of whether the production of a patient-specific CAR-T drug product for a target patient would lead to manufacturing failure (i.e., starting with the highest significance at the top and ending with the lowest significance at the bottom). For example, when using a trained machine learning model to predict whether the production of a patient-specific CAR-T drug product for a target patient would lead to manufacturing failure, manufacturing stage parameters with higher significance can be assigned a higher weight than other manufacturing stage parameters.

[0220] In various implementations, the method 600 for predicting whether the production of a patient-specific CAR T drug product for a target patient will result in manufacturing failure may further include generating an input feature vector (box 604) that includes quantitative data of the set of manufacturing failure parameters.

[0221] In various implementations, method 600 may further include applying the input feature vector to a trained machine learning model to generate an output feature vector (box 606) that predicts whether the production of a patient-specific CAR T drug product for a target patient will result in manufacturing failure.

[0222] In some implementations, receiving quantitative data for the set of manufacturing failure parameters includes receiving unstructured data for the set of manufacturing failure parameters. Method 600 may also include vectorizing the unstructured target data (e.g., via feature extraction module 318 of computing device 310) into an input feature vector.

[0223] In some implementations, the trained machine learning model can be trained using reference data from multiple reference CAR-T drug products manufactured from multiple reference patients, wherein the multiple reference CAR-T drug products may have known manufacturing failure results (e.g., whether the manufactured CAR-T drug product was a manufacturing failure or a success). Furthermore, method 600 may also include receiving (e.g., via computing device 310) reference data, which may include a set of input feature parameters for each of the multiple reference CAR-T drug products manufactured from multiple reference patients and known manufacturing failure results for the patient-specific CAR-T drug product. Additionally, for a given reference patient among the multiple reference patients, the set of input feature parameters may include at least the set of manufacturing failure parameters. In some aspects, method 600 may also include vectorizing (e.g., via feature extraction module 318 of computing device 310) the set of input feature parameters and the known manufacturing failure results for the patient-specific CAR-T drug product for each of the multiple reference CAR-T drug products manufactured from multiple reference patients into reference input feature vectors and reference output feature vectors, thereby generating multiple reference input feature vectors and multiple reference output feature vectors. In some aspects, method 600 may further include associating multiple reference input feature vectors with multiple reference output feature vectors in a machine learning model (e.g., via training module 320 of computing device 310). Furthermore, method 600 may also include training (e.g., via training module 320 of computing device 310) the machine learning model by iteratively minimizing errors to within predetermined thresholds to generate a trained machine learning model. As previously discussed, the trained machine learning model includes multiple weights. Each weight can indicate the significance of the input feature parameters in predicting whether the production of a patient-specific CAR T drug product for a target patient will result in manufacturing failure. In some embodiments, this set of input feature parameters is extracted from those parameters outlined in Appendix A.

[0224] In various embodiments, method 600 may further include determining whether the production of a patient-specific CAR T-drug product for a target patient will result in manufacturing failure (box 608). If manufacturing failure is predicted, method 600 may further include changing or adjusting one or more manufacturing process parameters used to manufacture the CAR T-drug product for the target patient (box 610). For example, one or more adjusted manufacturing process parameters may be output (e.g., as a recommendation via user interface 326 of computing device 310). Alternatively or additionally, the adjusted one or more manufacturing process parameters may be implemented during the production of the CAR T-drug product. In some embodiments, for example, if the production of a patient-specific CAR T-drug product for a target patient is predicted not to result in manufacturing failure (e.g., the production of the patient-specific CAR T-drug product would result in manufacturing success), method 600 may further include inducing the production of the CAR T-drug product (box 612). In some aspects, production may be based on the current set of manufacturing process parameters. In some embodiments, inducing production may involve the computing device displaying (e.g., via a user interface, such as user interface 326) a prediction that the CAR T-drug product being produced will not result in any manufacturing failure. Alternatively or concurrently, the computing device may transmit a signal that enables the equipment configured to manufacture CART drug products (e.g., bioreactor system 370) to continue manufacturing.

[0225] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in connection with the disclosure herein can be implemented as electronic hardware, computer software, or a combination of both. Those skilled in the art will also readily recognize that the order or combination of components, methods, or interactions described herein is merely illustrative, and that components, methods, or interactions of various aspects of this disclosure may be combined or performed in ways other than those illustrated and described herein.

[0226] The operation of the methods or algorithms disclosed herein can be implemented in a processor-executable software module that may reside on a computer-readable medium and be commercially available as software as a computer program product. Computer-readable media includes both computer storage media and communication media, including any medium that can be enabled to transfer a computer program from one place to another. Storage media can be any available medium accessible to a computer. By way of example, and not limitation, such computer-readable media may include random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), CD-ROM or other optical disc storage devices, magnetic disk storage devices or other magnetic storage devices, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer. Additionally, any connection may be properly referred to as a computer-readable medium. As used herein, disks and optical discs include compact discs (CDs), laser discs, optical discs, digital versatile optical discs (DVDs), floppy disks, and Blu-ray discs, wherein disks typically reproduce data magnetically and optical discs reproduce data optically using lasers. The combination of the above should also be included within the scope of computer-readable media.

[0227] Various modifications to the specific embodiments described in this disclosure will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other specific embodiments without departing from the spirit or scope of this disclosure. Therefore, the claims are not intended to be limited to the specific embodiments shown herein, but are to be given the broadest scope consistent with the disclosure, principles, and novel features disclosed herein.

[0228] Certain features described in this specification in the context of a single embodiment may also be implemented in a combined form in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments. Furthermore, while features may be described above as functioning in certain combinations and even initially so protected by the claims, one or more features from a claimed combination may be removed from that combination in some cases, and a claimed combination may involve sub-combinations or variations thereof.

[0229] Similarly, although operations are depicted in a specific order in the accompanying drawings, this should not be construed as requiring such operations to be performed in the specific order shown or sequentially, or requiring the execution of all illustrated operations to achieve the desired result. Furthermore, the drawings may schematically depict one or more example processes in the form of flowcharts. However, other operations not depicted may be incorporated into the schematically illustrated example processes. For example, one or more additional operations may be performed before, after, simultaneously with, or between any illustrated operation. In some cases, multitasking and parallel processing may be advantageous. Moreover, the separation of various system components in the embodiments described above should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products. Additionally, some other embodiments fall within the scope of the following claims. In some cases, the actions recited in the claims may be performed in a different order and still achieve the desired result.

[0230] As used herein (including the claims), the term “or” when used in a list of two or more items means that any one of the listed items may be used alone, or any combination of two or more of the listed items may be used. For example, if a composition is described as containing component A, B, or C, the composition may contain A alone; B alone; C alone; a combination of A and B; a combination of A and C; a combination of B and C; or a combination of A, B, and C. Furthermore, as used herein (including the claims), “or” as in a list of items beginning with “…at least one of…” indicates a retraction list, such that a list such as “at least one of A, B, or C” means, for example, A or B or C or AB or AC or BC or ABC (i.e., A and B and C) or any combination thereof.

[0231] The foregoing description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but is to be accorded the broadest scope consistent with the principles and novel features disclosed herein.

[0232] Appendix A

[0233]

[0234]

[0235]

[0236] --

[0237]

[0238]

[0239]

[0240]

[0241]

[0242]

[0243]

[0244]

[0245]

[0246]

[0247]

[0248]

[0249]

[0250]

[0251]

[0252]

[0253]

[0254]

[0255]

[0256]

[0257]

[0258]

[0259]

[0260]

[0261]

[0262]

Claims

1. A method for predicting manufacturing failures in the production of a patient-specific CAR T drug product for a target patient, the method comprising: Receive quantitative data of a set of manufacturing failure parameters, wherein the set of manufacturing failure parameters includes two or more manufacturing failure parameters selected from Table 1, wherein each manufacturing failure parameter belongs to one of a plurality of parameter types as outlined in Table 1; Generate an input feature vector that includes the quantitative data comprising the set of manufacturing failure parameters; as well as The input feature vector is applied to a trained machine learning model to generate an output feature vector that predicts whether the production of the patient-specific CAR T drug product will lead to manufacturing failure.

2. The method of claim 1, wherein the manufacturing failure parameters as outlined in Table 1 are ordered in order of significance in predicting whether the production of the patient-specific CAR T drug product will lead to the manufacturing failure, wherein when the trained machine learning model is used to predict whether the production of the patient-specific CAR T drug product will lead to the manufacturing failure, manufacturing failure parameters with higher significance are assigned higher weights than other manufacturing failure parameters.

3. The method according to claim 1 or 2, wherein the set of manufacturing failure parameters includes a set of screening parameters selected from Table 1A, wherein Table 1A consists of screening parameters from Table 1.

4. The method of claim 3, wherein the screening parameters in Table 1A are arranged in order of significance in predicting whether the production of the patient-specific CAR T drug product will lead to the manufacturing failure, wherein when the trained machine learning model is used to predict whether the production of the patient-specific CAR T drug product will lead to the manufacturing failure, screening parameters with higher significance are assigned higher weights than other screening parameters.

5. The method according to any one of the preceding claims, wherein the set of manufacturing failure parameters includes a set of manufacturing stage parameters selected from Table 1B, wherein Table 1B consists of manufacturing stage parameters from Table 1.

6. The method of claim 5, wherein the manufacturing stage parameters in Table 1B are arranged in order of significance of whether the production of the patient-specific CAR T drug product will lead to manufacturing failure when using the trained machine learning model, wherein manufacturing stage parameters with higher significance are assigned higher weights than other manufacturing stage parameters when using the trained machine learning model to predict whether the production of the patient-specific CAR T drug product will lead to manufacturing failure.

7. The method according to any one of the preceding claims, wherein receiving quantitative data of the set of manufacturing failure parameters includes receiving unstructured data of the set of manufacturing failure parameters, the method further comprising: The unstructured target data is vectorized into the input feature vector by the feature extraction module of the computing device.

8. The method according to any one of the preceding claims, wherein the trained machine learning model is trained using reference data from multiple reference CAR T drug products manufactured from multiple reference patients, the multiple reference CAR T drug products having known manufacturing failure results.

9. The method of claim 13, further comprising: The reference data is received via the computing device, wherein the reference data includes a set of input characteristic parameters for each of the plurality of reference CAR T drug products manufactured from the plurality of reference patients and the known manufacturing failure results. Wherein, for a given reference patient among the plurality of reference patients, the set of input feature parameters includes at least the set of manufacturing failure parameters; Through the feature extraction module of the computing device, for each of the multiple reference CAR T drug products manufactured from the multiple reference patients, the set of input feature parameters and the known manufacturing failure results are vectorized into reference input feature vectors and reference output feature vectors, respectively, thereby generating multiple reference input feature vectors and multiple reference output feature vectors; The training module of the computing device associates the plurality of reference input feature vectors with the plurality of reference output feature vectors in the machine learning model; and The training module of the computing device trains the machine learning model by iteratively minimizing the error to within a predetermined threshold to generate the trained machine learning model, wherein the trained machine learning model includes multiple weights, each weight indicating the significance between the input feature parameters and the manufacturing failure result.

10. The method of claim 14, wherein the set of input feature parameters is summarized in Appendix A.

11. The method according to claim 14 or 15, wherein, For each of the plurality of reference CAR T drug products manufactured from the respective plurality of reference patients, the set of input characteristic parameters includes two or more of the following: The percentage of CAR+ T cells in T cell culture samples harvested from intermediate to late stages of the manufacturing process of the reference CAR T drug product; the concentration of lactate or glucose in T cell culture samples from the late intermediate stage of the manufacturing process of the reference CAR T drug product; The concentration of lactate or glucose in the T cell culture sample from the intermediate stage of the manufacturing process of the reference CAR T drug product; The ratio of CD4+ T cells to CD8+ T cells in the T cell culture sample from the initial stage of the manufacturing process of the reference CAR T drug product; The volume of the carrier added to the T cell culture sample during an early intermediate stage of the manufacturing process of the reference CAR T drug product; The multiple of infection (MOI) of the vector added to the T cell culture sample during the early intermediate stage of the manufacturing process of the reference CAR T drug product. The average percentage of live T cells in each population of the T cell culture samples from the early intermediate stage of the manufacturing process of the reference CAR T drug product; The average concentration of live T cells in each population from the early intermediate stages of the manufacturing process of the reference CAR T drug product; The concentration of lymphocytes in a single sample prior to the manufacturing process of the reference CAR T drug product; The percentage of T cell cultures as live T cells in the T cell culture sample from the initial stage of the manufacturing process of the reference CAR T drug product; Whether the reference patients were refractory to pomalidomide treatment; The gender of the reference patient; The age of the reference patient; The reference patient's body mass index (BMI); or The reference patient's previous lines of treatment.

12. The method according to any one of the preceding claims further comprises: It has been determined that the production of the patient-specific CAR T drug product will result in manufacturing failure. as well as Adjust one or more manufacturing process parameters used to manufacture the CAR T drug product for the target patient.

13. The method according to any one of the preceding claims further comprises: It has been determined that the production of the patient-specific CAR T drug product will not result in manufacturing failure; as well as Based on the aforementioned set of manufacturing failure parameters, the manufacturing of the CAR T drug product for the target patient is initiated.

14. The method according to any one of the preceding claims, wherein the set of manufacturing failure parameters includes the single-sample stage parameters of Table 1, the single-sample stage parameters describing the concentration of lymphocytes in a single sample prior to the manufacturing process of the patient-specific CAR T drug product.

15. The method according to any one of the preceding claims, wherein the two or more manufacturing failure parameters include two or more of the following: The percentage of CAR+ T cells in T cell culture samples harvested from intermediate to late stages of the manufacturing process of the CAR T drug product; The concentration of lactate or glucose in T cell culture samples from a later intermediate stage of the manufacturing process of the CAR T drug product; The concentration of lactate or glucose in the T cell culture sample from the intermediate stage of the manufacturing process of the CAR T drug product; The ratio of CD4+ T cells to CD8+ T cells in the T cell culture sample from the initial stage of the manufacturing process of the CAR T drug product; The volume of the carrier added to the T cell culture sample during an early intermediate stage of the manufacturing process of the CAR T drug product; The multiple of infection (MOI) of the vector added to the T cell culture sample during the early intermediate stage of the manufacturing process of the CAR T drug product. The average percentage of live T cells in each population of the T cell culture samples from the early intermediate stage of the manufacturing process of the CAR T drug product; The average concentration of live T cells in each population from the early intermediate stages of the manufacturing process of the CAR T drug product; The concentration of lymphocytes in a single sample prior to the manufacturing process of the CAR T drug product; The percentage of T cell culture samples as live T cells from the initial stage of the manufacturing process of the CAR T drug product; Does the target patient have treatment resistance to pomalidomide? The gender of the target patient; The age of the target patient; The target patient's body mass index (BMI); or The number of previous lines of treatment for the target patient.

16. A system for predicting manufacturing failures in the production of a patient-specific CAR T-drug product for a target patient, the system comprising: Memory, the memory storing processor-readable code; as well as One or more processors coupled to the memory, the one or more processors being configured to execute processor-readable code to cause the one or more processors to perform the method according to any one of the preceding claims.

17. A non-transitory computer-readable medium storing computer instructions for predicting manufacturing failures in the production of a patient-specific CAR T drug product for a target patient, the computer instructions comprising the method according to any one of claims 1 to 20.

Images