Artificial intelligence (AI) has helped researchers reveal that Parkinson’s disease is not uniform at a cellular level, paving the way for personalised treatments that match a patient’s individual cell biology.
Neuroscience Research Australia (NeuRA) researchers Senior Research Fellow Dr Gautam Wali and Professor Carolyn Sue, and their colleagues used machine learning and advanced cell imaging to identify patterns of cellular dysfunction in Parkinson’s.
“Parkinson’s disease affects each person differently, including the symptoms they experience, how quickly the disease progresses and how they respond to treatment,” Prof Sue said.
“This study provides new insight into the biological differences that may help explain why.
“With the assistance of AI, we were able to analyse hundreds of features inside patient-derived cells at the same time and detect biologically meaningful differences between patient groups. This brings us a step closer to personalised treatments for patients.”
Study focus and key findings
The study focused on two subcellular systems: the cell’s energy producers, mitochondria, and its waste-clearance system, lysosomes. Researchers applied cell morphomics, a machine-based learning framework combining high content imaging with quantitative feature extraction to analyse the morphology at single-cell resolution in Induced pluripotent stem (iPS) cell-derived cortical neurons from people with Parkinson’s disease and healthy controls.
Researchers found the pattern of dysfunction in these systems differed across patient groups. PRKN-associated neurons had a predominantly mitochondrial pattern of dysfunction, while LRRK2 neurons showed greater lysosomal involvement, and SNCA neurons showed features of both.
“Using advanced cell imaging and machine learning, we have uncovered differences that suggest that different forms of Parkinson’s disease may ultimately require different treatment strategies,” Dr Wali said.
“In the future, this could help us better classify Parkinson’s disease, identify new cellular biomarkers, and determine which treatments are most likely to correct the underlying pathology is different patient groups.
“Importantly, these findings were reproduced across two independent laboratories, strengthening confidence that the platform is robust and reproducible.
“The next step is to test our machine-learning-based approach in larger preclinical studies using patient samples collected from multiple sites around the world.”
What this means for people with Parkinson’s disease
Around 90% of people with the disease have idiopathic Parkinson’s, meaning there is no known disease-causing mutation and genetics alone cannot explain what’s driving their disease. Dr Wali said this is where this research could be particularly valuable.
The research will not immediately change the treatment people receive in the clinic, but it is an important step towards personalised medicine.
“For now, treatment will continue to be based on each person’s symptoms and individual clinical needs,” Dr Wali said.
“In the short term, this has established a powerful preclinical research platform that gives researchers a more sensitive way to identify biologically distinct patient groups and test whether potential treatments can correct the cellular abnormalities seen in patient-derived cells.
“In the long term, this could guide the development of new treatments and support more targeted clinical trials by identifying which groups are most likely to respond to a particular therapy. Ultimately, it may help move treatment decisions beyond symptoms or genetics alone, and towards the cellular biology driving each person’s disease.”
The study is part of a larger project funded by the Michael J. Fox Foundation focused on understanding how Parkinson’s disease-associated genes affect human neurons, and also supported by funds from NeuRA’s Rising Star Award to Dr Wali.
The report was published as “Machine learning image analysis predicts Parkinson disease genotype and mitochondrial-lysosomal abnormalities in iPS neurons” in npj Parkinson’s Disease. It is available here.
