Machine Learning Used to Predict Parkinson’s Subtypes Using Stem Cell Images, with Accuracy Reaching 95%
Researchers at the Francis Crick Institute and UCL Queen Square Institute of Neurology, in collaboration with technology company Faculty AI, have made a significant breakthrough in predicting subtypes of Parkinson’s disease using stem cell images. Their findings, published in Nature Machine Intelligence, demonstrate that computer models can accurately classify four subtypes of Parkinson’s disease, with top accuracies reaching 95%. This advancement has the potential to revolutionize personalized medicine and facilitate more targeted drug research for Parkinson’s.
Parkinson’s disease is a neurodegenerative condition that affects both movement and cognition. The symptoms and progression of the disease vary from person to person due to different underlying mechanisms. Currently, there is no accurate way to differentiate between subtypes, resulting in non-specific diagnoses and limited access to tailored treatments and support.
The research team generated stem cells from patients’ own cells and created four different subtypes of Parkinson’s disease in the lab. They then used advanced imaging techniques to examine microscopic details of the disease models and labeled key components within the cells, such as mitochondria and lysosomes. By training a computer program to recognize each subtype, the team was able to accurately predict the subtype when presented with new images.
The study revealed that the most predictive features for subtype classification were the mitochondria and lysosomes, confirming their role in the development of Parkinson’s disease. Other areas of the cell, including the nucleus, were also found to be important. The researchers also noted that there were aspects of the images which they could not yet explain.
James Evans, a PhD student at the Crick and UCL, stated that the use of AI in this study allowed them to evaluate a larger number of cell features and understand their importance in distinguishing disease subtypes. He expressed a desire to expand this approach to explore how different cellular mechanisms contribute to other subtypes of Parkinson’s.
Sonia Gandhi, assistant research director and group leader of the Neurodegeneration Biology Laboratory at the Crick, emphasized the need for precise treatments in Parkinson’s disease. She stated that the team’s platform could potentially identify disease subtypes in living patients and enable the testing of drugs in stem cell models to predict whether a patient’s brain cells would respond to the treatment.
The project was developed during the disruptions caused by the pandemic, with the research team undergoing an intensive coding course to learn Python. Their success not only validated their ability to apply AI best practices to their scientific work but also contributed to the expansion of the Crick’s AI and software engineering team.
The next steps for the researchers include studying disease subtypes in people with different genetic mutations and investigating whether sporadic cases of Parkinson’s disease can be classified similarly.
This breakthrough in using machine learning to predict Parkinson’s disease subtypes based on stem cell images offers new opportunities for personalized medicine and targeted drug development. With further advancement in understanding cellular mechanisms, it is hoped that this approach will lead to transformative changes in how Parkinson’s disease is diagnosed and treated.