Robotics combined with AI machine learning spots Parkinson’s disease signatures.
- Over 10 million people worldwide live with Parkinson’s disease, including nearly a million Americans.
- A new study uses AI deep learning that finds cellular disease signatures to help accelerate the discovery of novel therapeutics for Parkinson’s.
- This unique AI deep learning platform solution is not limited to Parkinson’s disease. It can be repurposed for other disease signatures.
Artificial intelligence (AI), machine learning, and robotics are accelerating precision medicine for neurodegenerative diseases and brain disorders.
A new study published in Nature Communications reveals a high-throughput screening platform using AI deep learning that finds cellular disease signatures to help accelerate the discovery of novel therapeutics for Parkinson’s disease (PD).
“To our knowledge, this is the first successful demonstration in which automated, unbiased deep learning-based phenotypic profiling is able to discriminate between primary cells from PD patients (both sporadic and LRRK2) and healthy controls,” wrote the study authors affiliated with The New York Stem Cell Foundation Research Institute (NYSCF) and Google Research.
Over 10 million people worldwide are living with Parkinson’s disease, including nearly a million Americans, according to the Parkinson’s Foundation, a nonprofit organization.
Parkinson’s disease is an incurable, progressive nervous system disorder that affects movement. Symptoms of Parkinson’s disease include tremors, written and spoken communication changes, posture and balance issues, muscle rigidity and stiffness, loss or reduction of unconscious movements, and bradykinesia (slowed movement). The exact cause of Parkinson’s disease has not been identified.
The researchers used images of millions of skin cells from Parkinson’s disease patients and health controls using The NYSCF Global Stem Cell Array® — a large dataset of patient cells and a state-of-the-art robotic system. The system helped to identify and label via Cell Painting fibroblasts, which are cells that produce and maintain connective tissue, as well as generate microscopy images.
“The scale of this unbiased high-content profiling experiment is, to our knowledge, unprecedented: it provides the scientific community with the largest publicly available Cell Painting dataset to date (in terms of pixel count) at 48 terabytes in size, compared with the next largest dataset at 13 terabytes (RxRx19a),” the researchers reported.
To analyze these images, the researchers used a custom-developed AI deep learning algorithm consisting of a deep convolutional neural network that was pretrained using the object recognition dataset ImageNet. The team used multiple supervised machine learning models, such as multilayer perceptron (MLP), logistic regression classifier, ridge regression, and random forest for modeling.article continues after advertisement
The scientists report that their automated platform has the ability to simultaneously profile ninety-six primary cell lines with consistent growth rates and minimal variation.
“In summary, we employed a robust experimental design that successfully minimized the effect of potential covariates,” the researchers wrote. “We also established a comprehensive image analysis pipeline where multiple machine learning models were applied to each classification task, using both computed deep embeddings and extracted cell features as data sources.”
What sets this AI deep learning platform solution apart is that it is not limited to Parkinson’s disease; it can be repurposed for other disease signatures.
Our ability to identify Parkinson’s-specific disease signatures using standard cell labeling and deep learning-based image analysis highlights the generalizable potential of this platform to identify complex disease phenotypes in a broad variety of cell types,” the scientists wrote. “This represents a powerful, unbiased approach that may facilitate the discovery of precision drug candidates undetectable with traditional target- and hypothesis-driven methods.