Health & Medicine, UK (Commonwealth Union) – A collaborative effort between researchers at the Francis Crick Institute and the University College London (UCL), Queen Square Institute of Neurology, in partnership with technology company Faculty AI, has demonstrated the remarkable potential of machine learning in accurately predicting distinct subtypes of Parkinson’s disease by analyzing images of stem cells derived from patients.
Their recent study, published in Nature Machine Intelligence, has unveiled the capability of computer models to proficiently categorize 4 distinct subtypes of Parkinson’s disease, achieving an impressive accuracy rate of 95%. This breakthrough has the potential to open doors to personalized medical treatments and targeted exploration of new drugs.
Parkinson’s disease is a neurodegenerative disorder affecting both movement and cognitive functions. Its symptoms and progression manifest differently in individuals due to variations in the underlying mechanisms driving the condition.
Until now, the accurate differentiation of these subtypes has remained a challenge, resulting in generic diagnoses that often restrict access to tailored treatments, support, and care.
Parkinson’s disease involves the misfolding of crucial proteins and dysfunction in the clearance of defective mitochondria, the cell’s energy producers. While most instances of Parkinson’s disease emerge sporadically, certain cases can be associated with genetic mutations.
The researchers cultivated stem cells from patients’ own cellular material and artificially generated four distinct subtypes of Parkinson’s disease. Two subtypes were associated with pathways leading to the accumulation of a protein called α-synuclein, while the other two involved pathways connected to malfunctioning mitochondria. This innovative approach essentially created a ‘brain disease model in a dish.’
Subsequently, they meticulously imaged the disease models at a microscopic level, specifically labeling vital cellular components like lysosomes, which play a role in breaking down worn-out cellular parts. The researchers then ‘trained’ a computer program to recognize each subtype, enabling it to accurately predict the subtype when presented with previously unseen images.
The analysis highlighted the pivotal role of mitochondria and lysosomes in effectively predicting the correct subtype, thus reaffirming their significance in the development of Parkinson’s disease. Other parts of the cell, such as the nucleus, were also identified as significant contributors, along with aspects of the images that are not yet fully understood.
James Evans, a doctoral candidate at both the Crick Institute and UCL, and co-primary author alongside Karishma D’Sa and Gurvir Virdi, indicated that with their adoption of more sophisticated imaging methods, they now produce extensive volumes of data, a significant portion of which is discarded during the manual selection of a handful of noteworthy attributes.
“Using AI in this study enabled us to evaluate a larger number of cell features, and assess the importance of these features in discerning disease subtype. Using deep learning, we were able to extract much more information from our images than with conventional image analysis. We now hope to expand this approach to understand how these cellular mechanisms contribute to other subtypes of Parkinson’s.”
Sonia Gandhi, Assistant Research Director and head of the Neurodegeneration Biology Laboratory at the Crick Institute, indicated that they have gained insight into numerous processes that underlie Parkinson’s disease within the human brain. However, during their lifetimes, they lack the means to pinpoint the specific mechanisms at play, which consequently prevents them from administering targeted treatments.
The project came into fruition amidst the disruptions caused by the pandemic, which had temporarily halted the lab’s ongoing research. In response, the entire team collectively engaged in an immersive coding program, mastering the intricacies of Python programming. These newly acquired skills have since been effectively leveraged in ongoing work.
Moving forward, the research team aims to delve into the realm of disease subtypes among individuals exhibiting different genetic mutations. Additionally, they aspire to ascertain the feasibility of categorizing sporadic Parkinson’s disease cases—those devoid of genetic mutations—using a comparable approach.
