Scientists from the Francis Crick Institute and UCL Queen Square Institute of Neurology, in partnership with Faculty AI, have demonstrated the ability of machine learning to accurately predict subtypes of Parkinson’s disease using images of patient-derived stem cells. Their research, published in Nature Machine Intelligence, showcases computer models’ potential to classify four Parkinson’s subtypes with an impressive accuracy of 95% in one instance. This breakthrough could open doors for personalized medicine and targeted drug discovery.
Parkinson’s disease, a neurodegenerative condition affecting movement and cognition, presents diverse symptoms and progression due to various underlying mechanisms. Until now, distinguishing between subtypes has been difficult, leading to vague diagnoses and limited access to tailored treatments.
By creating stem cells from patients’ cells and developing four distinct Parkinson’s subtypes, researchers established a “human brain disease model in a dish.” Using microscopic imaging of these models, which highlighted cell components such as lysosomes involved in cellular breakdown, a computer program learned and predicted each subtype, focusing on critical features like mitochondria and lysosomes.
James Evans, a Crick and UCL Ph.D. student, highlighted how AI enabled the analysis of multiple cell features, providing deeper insights than traditional methods. The team plans to expand this approach to understand other Parkinson’s subtypes and their cellular mechanisms.
Sonia Gandhi, assistant research director at the Crick, noted that while the causes of Parkinson’s are understood, pinpointing mechanisms in living individuals remains challenging. This advancement holds promise for more precise future treatments.
Generated nerve cells in the brain’s cortex from patients’ stem cells – the type of image used by the computer model. Image Credit: D’Sa, K. et al. Nature Machine Intelligence. (2023).
“We lack effective treatments that significantly alter the course of Parkinson’s disease. By utilizing patient-specific neuron models and combining them with extensive image data, we developed an algorithm capable of categorizing specific subtypes. This potent method could potentially revolutionize the identification of disease subtypes during a patient’s lifetime.”
“Taking this a step further, our platform could enable us to pretest drugs using stem cell models, predicting the likelihood of a patient’s brain cells responding to the treatment before embarking on clinical trials. The aspiration is that this approach might ultimately reshape how personalized medicine is administered.”
James Fleming, Chief Information Officer at the Crick, collaborated with Faculty AI on this endeavor and remarked, “AI is an enthralling and formidable technology, often obscured by exaggeration and technical jargon. This study emerged from a distinctive industry partnership with Faculty, exploring whether AI novices could rapidly learn and apply best practices to their scientific work.”
“The triumph of this project not only demonstrated their capability to do so, unveiling novel insights along the way, but also propelled investment into the swift expansion of our own AI and software engineering team. With over 25 ongoing projects spanning various labs at the Crick and new initiatives commencing each month, this success has been instrumental.”
Going forward, the research team aims to understand disease subtypes among individuals with different genetic mutations and explore the possibility of classifying sporadic cases of Parkinson’s disease (those without genetic mutations) similarly.
Source The Francis Crick Institute
