A new artificial intelligence (AI)-driven platform developed by researchers from the New York Stem Cell Foundation (NYSCF) Research Institute and Google Research successfully identified a cellular signal that differentiates skin cells of Parkinson’s disease patients from healthy controls.
The platform uses a combination of automated cell culture, high-quality imaging, Cell painting—using a range of different fluorescent dyes to label different cell parts, and deep learning—a type of AI and machine learning that imitates the way humans learn. The researchers hope it will help improve the drug development process for conditions such as Parkinson’s where many drug trials have failed in the past.
“Traditional drug discovery isn’t working very well, particularly for complex diseases like Parkinson’s,” explained Susan Solomon, Co-Founder and CEO of the NYSCF, in a press statement. “The robotic technology NYSCF has built allows us to generate vast amounts of data from large populations of patients, and discover new signatures of disease as an entirely new basis for discovering drugs that actually work.”
The study, which is published in Nature Communications, used the extensive patient cell repository and robotic cell culture system built by NYSCF to help train deep learning algorithms built by Google Research scientists.
The researchers tested the system on fibroblast skin cells from 91 individuals, half of whom were Parkinson’s patients and half matched, healthy controls. The cells underwent ‘painting’ with fluorescent dyes and the large number of images created were fed into the AI algorithm for assessment.
“These artificial intelligence methods can determine what patient cells have in common that might not be otherwise observable,” said Samuel Yang, Research Scientist at Google Research and co-lead author on the study. “What’s also important is that the algorithms are unbiased—they do not rely on any prior knowledge or preconceptions about Parkinson’s disease, so we can discover entirely new signatures of disease.”
The test was able to accurately identify cells from Parkinson’s patients, both genetic and sporadic cases, and differentiate them from healthy controls with a high level of accuracy (a receiver operating characteristic area under curve of 0.79). The algorithm was able to differentiate the different types of Parkinson’s and also some level of individual variation between patients.
“Excitingly, we were able to distinguish between images of patient cells and healthy controls, and between different subtypes of the disease,” noted Bjarki Johannesson, a NYSCF Senior Investigator and study author. “We could even predict fairly accurately which donor a sample of cells came from.”
This platform can now be used to help drug screening for Parkinson’s by assessing if the disease-specific features on the cells can be reversed by new drugs in development. The platform could also be used to research other diseases, as it is not specific to Parkinson’s.
“This is the first tool to successfully identify disease features with this much precision and sensitivity,” said NYSCF Senior Vice President of Discovery and Platform Development Daniel Paull. “Its power for identifying patient subgroups has important implications for precision medicine and drug development across many intractable diseases.”