AI Used to Assure Quality of Cell Therapy

Download PDF

Researchers at the NEI have developed a simple method using artificial intelligence (AI) to assure quality control of a cell therapy for patients with age-related macular degeneration (AMD).¹ In a proof-of-principle study, they con­firmed that their methodology reliably, quickly, and noninvasively evaluated their autologous cell therapy product.

Their approach should increase tissue production and speed its delivery to the clinic for replacement of degenerated retinal pigment epithelial (RPE) cells.

“This AI-based method of validating stem cell-derived tissues is a significant improvement over conventional assays, which are low-yield, expensive, and re­quire a trained user,” said Kapil Bharti, PhD, at the NEI Ocular and Stem Cell Translational Research Section. “The current technology brings the autolo­gous cell therapy a step closer to AMD patients.”

MICROGRAPH. This image shows the fiber-based scaffold (blue) and cultured iPSC-RPE cells (gray). The hair-like structures on top of each cell are their apical processes that confirm their polarity and maturity.

Seeking confirmation. With a gar­den-variety microscope programmed with deep learning algorithms, a technician will be able to verify that the replacement RPE cells are correctly manufactured just prior to transplan­tation in patients. Specifically, the AI methodology allows validation of “patches” of stem cell–derived RPE cells. The RPE “patch” is made from induced pluripotent stem cells (iPSCs) that are made from the patient’s blood.

The need for validating healthyreplacement cells. This need was underscored by the researchers, who noted that at least 11 investigations are underway using RPE cells to treat AMD.² In fact, they are awaiting FDA approval of a phase 1 trial to transplant RPE cells in AMD patients. Pending approval, they will begin manufactur­ing patient cells, likely this year.

A two-step methodology. Dr. Bharti’s team first had to validate the ability of quantitative bright-field absorbance microscopy (QBAM) to make a precise, reproducible measurement of tissue quality. Next, they had to employ AI to analyze QBAM images for predicting multicellular function.

To that end, Dr. Bharti’s team trained deep neural networks (DNNs) to assess QBAM images of iPSC-RPE created from both healthy and diseased donors. They found that deep learning could determine the sensitivity of QBAM to biological variation. The DNNs also identified borders of cells in QBAM images. And DNNs determined if the cells came from the same donor.

Confirming the identity of each patient’s dose is essential, because the lab will be manufacturing cells from multiple patients simultaneously. “For every patient, we need to manufacture this product over and over again, and functionally validate it every time,” Dr. Bharti said. “This will be a live and noninvasive method to confirm identi­ty of given donor’s cells.”

Toward a clinical application. While awaiting the green light from the FDA, the team has begun implementing its deep learning software onto micro­scopes that they plan to install in their manufacturing facility. “Once that’s completed, we are ready to go,” Dr. Bharti said. “With our new AI-based method to functionally validate patient cell–derived transplants, we are more confident that we are manufacturing the correct, safe, and functional clinical product.”

—Miriam Karmel

___________________________

1 Schaub NJ et al. J Clin Invest. Published online Nov. 12, 2019.

2 Aijaz A et al. Nat. Biomed Eng. 2018;2(6):362-376.

___________________________

Relevant financial disclosures—Dr. Bharti: None. This study was supported by the National Institute of Standards and Technology.

For full disclosures and the disclosure key, see below.

• New ROP Screening Criteria Validated

• Dry Eye, Migraine, and Visual Quality of Life

• OCT Provides Fuller Picture of Tamoxifen Retinopathy