Using machine learning and/or neural networks to validate stem cells and their derivatives (2-D cells and 3-D tissues) for use in cell therapy and tissue engineered products
Inventors
Bharti, Kapil • Hotaling, Nathan A. • SCHAUB, Nicholas J. • SIMON, Carl G.
Assignees
US Department of Health and Human Services
Publication Number
US-12020494-B2
Publication Date
2024-06-25
Expiration Date
2039-03-15
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Abstract
A method is provided for non-invasively predicting characteristics of one or more cells and cell derivatives. The method includes training a machine learning model using at least one of a plurality of training cell images representing a plurality of cells and data identifying characteristics for the plurality of cells. The method further includes receiving at least one test cell image representing at least one test cell being evaluated, the at least one test cell image being acquired noninvasively and based on absorbance as an absolute measure of light, and providing the at least one test cell image to the trained machine learning model. Using machine learning based on the trained machine learning model, characteristics of the at least one test cell are predicted. The method further includes generating, by the trained machine learning model, release criteria for clinical preparations of cells based on the predicted characteristics of the at least one test cell.
Core Innovation
The invention provides a non-invasive method for predicting characteristics of cells and their derivatives using machine learning models trained on images acquired noninvasively, specifically using absorbance as an absolute measure of light. The method includes training a machine learning model with training cell images and corresponding characteristic data, receiving a test cell image, and providing it to the trained model to predict cell characteristics. Importantly, the method generates release criteria for clinical preparations of cells based on the predicted characteristics.
The disclosed method overcomes limitations of existing assays for cell health, function, and maturity, which are often destructive, non-quantitative, variable, expensive, or require opening cell culture dishes, thus disturbing the cells. Current methods lack high-content information and are not compatible with longitudinal or high-throughput assessment. The invention allows non-invasive, automated, reproducible, and rapid evaluation of cell therapies, drug discovery, and toxicity testing using bright-field microscopy and machine learning, eliminating human bias and enabling high throughput assessment without disturbing the cells.
The invention utilizes novel computational frameworks combining quantitative bright-field absorbance microscopy (QBAM) and advanced machine learning including deep neural networks. QBAM converts pixel intensity images to absolute absorbance measurements ensuring reproducibility across microscopes. Deep learning models segment cells, extract features, and predict physiological and biochemical functions including transepithelial resistance (TER), donor identity, cell maturity, and function. The models can classify and predict drug effects, disease states, and generate clinical release criteria potentially suitable for various stem cells and their derivatives from human or animal tissue.
Claims Coverage
The patent claims include one independent method claim and one independent computing system claim, each featuring non-invasive prediction of retinal pigment epithelial (RPE) maturation using a trained machine learning model.
Non-invasive prediction of retinal pigment epithelial maturation
Obtaining a trained machine learning model trained on cell images and characteristics; receiving noninvasive test cell images acquired by fluorescent microscopy; providing these images to the model; predicting transepithelial resistance (TER) of test cells; and generating release criteria for clinical cell preparations for implantation based on predicted TER.
Use of deep neural networks for segmentation and classification
Performing machine learning using deep neural networks that segment test images into individual cells and classify test cells based on characteristics.
Prediction based on classification of cell properties
Predicting aspects such as cell identity, cell function, drug effect, disease state, and similarity to technical replicates or previous samples based on classification results.
Prediction on single cells or multiple cell fields
Capability to predict transepithelial resistance on a single cell or a field of view of multiple cells in a test image.
Feature extraction to improve prediction
Visually extracting features from test images, including cell boundaries, shapes, and texture metrics; training the model using these features; and predicting TER using the identified features.
Acquisition of images using quantitative bright-field absorbance microscopy (QBAM)
Receiving microscopy images, converting pixel intensities to absorbance values, calculating absorbance confidence, establishing microscope equilibrium via benchmarking, and filtering color during image acquisition.
Segmentation of test images for feature extraction
Segmenting test images into individual cells using a deep neural network to extract visual features from segmented cells for prediction.
Prediction of multiple cell functions and classifications
Predicting polarized vascular endothelial growth factor secretion, cell function, cell maturity, donor identity, and detection of outliers relative to known classifications.
Generation of release criteria for drug discovery and toxicity
Extending the trained machine learning model to generate release criteria related to drug discovery and drug toxicity.
Broad applicability to various stem cell types and derivatives
Applicability to embryonic stem cells, induced pluripotent stem cells, neural stem cells, retinal pigment epithelial stem cells, mesenchymal stem cells, hematopoietic stem cells, cancer stem cells, and cells derived therefrom from human or animal tissue.
Handling of large images by tiling and recombining outputs
Dividing large test images into tiles for individual processing by the model, combining outputs for overall prediction corresponding to the large image.
Computing system for implementing the method
A computing system comprising memory and processors configured to implement the non-invasive prediction method using trained machine learning models and fluorescent microscopy images to predict TER and generate release criteria.
The claims collectively cover a non-invasive machine learning-based method and system for predicting retinal pigment epithelial maturation and related cell characteristics from noninvasive cell images, employing deep neural networks for segmentation and classification, incorporating QBAM imaging, enabling prediction of functional and identity parameters, and generating release criteria applicable to various stem cell types and clinical applications.
Stated Advantages
Non-invasive and non-destructive analysis of cells and tissues.
Automation eliminating the need for trained users and reducing human bias and error.
High throughput compatibility enabling large scale biomanufacturing and drug screening.
Cost reduction and faster time for relevant prediction of tissue function and donor identity.
Reproducible image acquisition across different microscopes using quantitative bright-field absorbance microscopy.
Capability to longitudinally monitor cells during development and maturation without disturbing them.
Prediction reliability in evaluating cell maturity, function, disease states, and drug/toxin effects.
Ability to classify cell lines, detect developmental outliers and identify donor identity for quality assurance in clinical cell therapy manufacturing.
Documented Applications
Cell therapy, including validation and release criteria for stem cells and their derivatives for implantation.
Drug discovery and drug toxicity testing using non-invasive monitoring of cellular responses.
High-throughput screening in biological and clinical research.
Quality assurance and identity verification in biomanufacturing of cell therapies.
Longitudinal assessment of cell health, function, and maturation in vitro prior to transplantation.
Interested in licensing this patent?