Using stem cell images, researchers have successfully applied machine learning to distinguish between Parkinson's disease subtypes with high accuracy. This method could dramatically improve personalized treatment strategies and accelerate the discovery of new targeted drugs. Credit: SciTechDaily.com
Machine learning has been used to accurately predict Parkinson's disease subtypes from stem cell images.
This advancement offers the potential for personalized medicine by classifying the four subtypes with up to 95% accuracy and leveraging stem cell models and AI to revolutionize treatment and drug testing approaches.
AI-driven breakthrough in Parkinson's disease research
Researchers from the Francis Crick Institute and the Queen Square Institute of Neurology, University of London, working with technology company Faculty AI, have shown that machine learning can accurately predict Parkinson's disease subtypes using images of patient-derived stem cells.
A study published in the journal Nature Machine Intelligence found that a computer model could accurately classify four subtypes of Parkinson's disease, one of which reached 95% accuracy, potentially paving the way for personalized medicine and the discovery of new targeted drugs.
Image of a neuron from the brain cortex generated from the patient's stem cells (left) – the type of image the computer model used, divided into panels showing different parts inside the cell (right). Credit: D'Sa, K. et al. Nature Machine Intelligence. (2023) Understanding variability in Parkinson's disease
Parkinson's disease is a neurodegenerative disorder that affects movement and cognition. Due to differences in the underlying mechanisms that cause the disease, symptoms and disease progression vary from person to person.
Until now, there was no way to accurately distinguish between subtypes, so people regularly received a non-specific diagnosis and were unable to receive targeted treatment, support and care.
Parkinson's disease involves the misfolding of key proteins and a malfunction in the removal of defective mitochondria, the source of energy production in cells. The majority of Parkinson's disease cases begin sporadically, but some may be linked to genetic mutations.
The researchers generated stem cells from the patients' own cells and chemically engineered four different subtypes of Parkinson's disease (two involved in pathways that lead to toxic buildup of a protein called alpha-synuclein, and two involved in pathways that lead to mitochondrial dysfunction), creating a “human model of brain disease in a dish.”
Brain cortex neurons generated from a patient's stem cells – the type of image used by the computer model. Credit: D'Sa, K. et al. Nature Machine Intelligence. (2023) Disease Classification with AI
The researchers then imaged the disease models in microscopic detail and labeled key cellular components, such as lysosomes, which are responsible for breaking down worn-out parts of cells. They “trained” a computer program to recognize each subtype, allowing it to predict the subtype when presented with images it had never seen before.
Although mitochondria and lysosomes were the most important features in predicting the correct subtype, supporting their involvement in Parkinson's disease pathogenesis, other regions of the cell, such as the nucleus, and as yet unexplained aspects of the image, were also found to be important.
Leveraging AI to gain deeper insights into disease
“The use of more advanced imaging techniques generates huge amounts of data, much of which is discarded when manually selecting a few features of interest,” said James Evans, a doctoral student at the Crick Institute and University College London and lead author of the paper with Karishma Dasa and Gurvir Virdi.
“Using AI in this study enabled us to assess many more cellular characteristics and evaluate the importance of these characteristics in identifying disease subtypes. By using deep learning, we were able to extract much more information from the images than was possible with traditional image analysis. In the future, we hope to extend this approach to understand how these cellular mechanisms contribute to other subtypes of Parkinson's disease.”
The Potential of Precision Medicine
“We understand many of the processes that cause Parkinson's disease in people's brains, but we have no way of knowing what mechanisms are happening while the patient is alive, so we can't treat them precisely,” said Sonia Gandhi, deputy research director and group leader in the Laboratory of Neurodegenerative Biology at the Crick Institute.
“Currently, there are no treatments that make a significant difference in the progression of Parkinson's disease. By using a model of a patient's own neurons and combining this with a large amount of imagery, we have created an algorithm that classifies the specific subtype. This is a powerful approach that opens the door to identifying disease subtypes in life. Taking this a step further, our platform will allow us to first test drugs on stem cell models and predict whether a patient's brain cells are likely to respond to the drug, before they enter clinical trials. In future, we hope this will bring about a fundamental change in how personalized healthcare is delivered.”
Developments amid the pandemic
The project was developed during the lab's research hiatus due to the COVID-19 pandemic. The entire team took an intensive coding course, learning to code in Python, gaining skills that they are now applying to their current project.
James Fleming, Crick's Chief Information Officer, who collaborated with Faculty AI on the project, said: “AI is a fascinating and powerful technology, but the hype and jargon often make it difficult to understand. This paper is the result of a unique industry partnership with the Faculty to explore whether a group of complete novices to AI could learn best practices in a very short timeframe and apply them directly to science. The success of this project not only proved it could be done, unlocking new insights in the process, but also helped to catalyze investment in the rapid expansion of our AI and software engineering team, which has over 25 projects 'in the works' across Crick's various labs, with new projects starting every month.”
Future directions for Parkinson's disease research
The team's next step is to understand disease subtypes in people with other gene mutations and to see whether cases of sporadic Parkinson's disease (i.e. without the gene mutation) can be classified in a similar way.
Reference: Karishma D'Sa, James R. Evans, Gurvir S. Virdi, Giulia Vecchi, Alexander Adam, Ottavia Bertolli, James Fleming, Hojong Chang, Craig Leighton, Mathew H. Horrocks, Dilan Athauda, Minee L. Choi, Sonia Gandhi, “Predicting mechanistic subtypes of Parkinson's disease using patient-derived stem cell models,” 10 August 2023, Nature Machine Intelligence.
DOI: 10.1038/s42256-023-00702-9
