Using machine learning to predict cell therapy characteristics
Disclosed are systems and methods for improving processes for developing cell therapies by applying machine learning to data including manufacturing process data and clinical measurements (e.g., patient response and treatment data) to determine parameters and settings for a manufacturing process for engineering cells for use in cell therapy. Parameters and settings for a manufacturing process for genetically engineered T-cells including, but not limited to, Chimeric Antigen Receptor (CAR) T cells can be determined. A method can include receiving a set of process parameters of a cell engineering process, predicting a clinical response associated with an output of the cell engineering process by applying a machine learning model on the received set of process parameters, where the machine learning model is trained on process parameter data and clinical response data, and generating a visualization for use in a graphical user interface of the predicted clinical response.
1 . A simulation-assisted method for predicting a clinical response for a cell engineering process, the method comprising:
receiving, by at least one processor, a set of measured or recorded process parameters of the cell engineering process;
training a machine learning bioprocess simulation model on process parameter data and clinical response data to computationally replicate, in silico, time-dependent cell-population dynamics and product characteristics of the cell-engineering, wherein the process parameter data comprises:
operator identification, initial volume, donor, mixing, dilution speed, input bag rinsing, optical cell detection, product filling speed, waste extraction speed, intermediate volume, pre-wash cycles, pre-wash g-force, pre-wash sedimentation time, switch washing solution, lactate concentration, oxygen concentration, CO2 concentration, hold time prior to freeze, cell freezing parameters, and thaw parameters;
executing by the at least one processor the machine learning bioprocess simulation model on the cell-engineering process;
generating, based on the simulation, a clinical response associated with the simulated biological outcome of the cell engineering process to produce a predicted clinical response;
generating, by the at least one processor, data usable to generate a visualization in a graphical user interface of the predicted clinical response; and
implementing control logic that sets one or more process parameters in accordance with the predicted clinical response to regulate operation of the cell-engineering system.
2 . The method of claim 1 , wherein the machine learning model comprises at least one of logistic regression, an elastic net, a k-nearest neighbor, a decision tree, a random forest, a support vector machine, a support vector, a light gradient boosting method, an extreme gradient boosting method, a neural network, or a multi-layer perceptron.
3 . The method of claim 1 , wherein the machine learning model is further trained on in vitro assay results of the cell engineering process, wherein the in vitro assay results comprises one or more of a cell number, percentage phenotype, cell recovery data, cell diameter, hold time, expansion properties of the engineered cells, persistence properties of the engineered cells, cytokine release patterns, or cytotoxicity levels in vitro.
4 . The method of claim 1 , wherein the cell engineering process comprises a process for generating Chimeric Antigen Receptor (CAR) T cells.
5 . The method of claim 1 , further comprising pre-processing the received set of process parameters by at least one of cleaning, deduplicating, standardizing, transforming, applying feature engineering, normalizing, scaling, encoding, integrating, or reducing the received set of process parameters.
6 . The method of claim 1 , further comprising:
adjusting one or more process parameters of the cell engineering process based on the predicted clinical response.
7 . The method of claim 1 , further comprising:
generating a set of cells based on the cell engineering process having adjusted process parameters.
8 . The method of claim 1 , wherein providing the predicted clinical response further comprises:
displaying in a graphical user interface the predicted clinical response and at least one of: one or more characteristics of the trained machine learning model, or the received set of process parameters.
9 . A method comprising:
obtaining a machine learning model which translates a set of process parameters of a cell engineering process into a simulated biological outcome;
training the machine learning model on process parameter data and clinical response data wherein the process parameter data comprises: operator identification, initial volume, donor, mixing, dilution speed, input bag rinsing, optical cell detection, product filling speed, waste extraction speed, intermediate volume, pre-wash cycles, pre-wash g-force, pre-wash sedimentation time, switch washing solution, lactate concentration, oxygen concentration, CO 2 concentration, hold time prior to freeze, cell freezing parameters, and thaw parameters;
obtaining the set of process parameters characterizing the cell engineering process;
converting, using the machine learning model, handwritten process parameters to a structured dataset for computational analysis with the machine learning model;
applying the machine learning model in silico to the dataset to simulate a biological outcome of the cell-engineering process;
predicting using the trained machine learning model, a clinical response associated with the simulated biological outcome; and
outputting or displaying, by the at least one processor, the predicted clinical response for use in adjusting or verifying process parameters of the cell-engineering process.