summary: Researchers used stem cell images to enable machine learning to accurately predict subtypes of Parkinson’s disease. This breakthrough shows a computer model that classifies his four subtypes of Parkinson’s disease, reaching his highest accuracy of 95%. This could revolutionize personalized medicine and aid in more targeted drug research for Parkinson’s disease.
Important facts:
- This study used machine learning to classify stem cell images into four different Parkinson’s disease subtypes.
- The most predictive features of subtyping were intracellular mitochondria and lysosomes.
- This technology may enable customized treatment of Parkinson’s disease based on specific subtypes.
sauce: Francis Crick Institute
Researchers from the Francis Crick Institute and the UCL Queen Square Neurological Institute, in collaboration with technology company Faculty AI, show that machine learning can accurately predict Parkinson’s disease subtypes using images of patient-derived stem cells Ta.
their work was published today nature machine intelligence, showed that a computer model could accurately classify four subtypes of Parkinson’s disease, one of which reached 95% accuracy. This could pave the way for personalized medicine and targeted drug discovery.
Parkinson’s disease is a neurodegenerative disease that affects movement and cognition. Symptoms and disease progression vary from person to person because of differences in the underlying mechanisms that cause the disease.
Until now, there has been no way to accurately distinguish between subtypes. This meant that people received non-specific diagnoses and did not always have access to targeted treatment, support and care.
Parkinson’s disease is associated with critical protein misfolding and dysfunctional clearance of defective mitochondria, the energy generators within the cell. Although the majority of Parkinson’s cases begin sporadically, some may be associated with genetic mutations.
Researchers generated stem cells from the patient’s own cells and chemically engineered four different subtypes of Parkinson’s disease. Two involve pathways that lead to toxic accumulation of a protein called alpha-synuclein, and two involve pathways that lead to dysfunctional mitochondria, creating ‘human’. A model of brain disease in a dish.
We then imaged the disease model in detail microscopically and labeled key cellular components, such as lysosomes, involved in the degradation of worn out parts of the cell. The researchers “trained” a computer program to recognize each subtype, and were able to predict the subtype when presented with images it had never seen before.
Mitochondria and lysosomes are the most important features in predicting the correct subtype, and the involvement of mitochondria and lysosomes in the pathogenesis of Parkinson’s disease has been confirmed. It turns out that the sides are also important. Please don’t explain yet.
James Evans, PhD student at Crick and UCL and co-first author with Karishma Dasa and Gurvil Virdi, said: Manually select some features of interest.
“By using AI in this study, we were able to assess more cellular features and assess the importance of these features in discriminating disease subtypes. This allowed us to extract far more information from the images than conventional image analysis, and we are now extending this approach to explore the implications of these cellular mechanisms for other subtypes of Parkinson’s disease. We want to understand how we contribute.”
Sonia Gandhi, Assistant Research Director and Group Leader, Crick University Neurodegenerative Biology Laboratory, said: However, since we do not know what mechanism is occurring while we are alive, we cannot provide accurate treatment.
“Currently, there is no treatment that makes a significant difference in the progression of Parkinson’s disease. We used a model of the patient’s own neurons and combined this with a large number of images to generate an algorithm that classifies specific subtypes. This is a powerful approach that opens the door to identifying disease subtypes in life.
“Taking this a step further, using our platform, we can first test drugs in stem cell models to predict whether a patient’s brain cells are likely to respond to the drug before enrolling in a clinical trial. It is hoped that one day this will fundamentally change the way personalized medicine is delivered.”
The project was developed while research in the lab was suspended due to the pandemic. The entire team took an intensive coding course to learn to code in Python and develop skills that they apply to their current projects.
Crick Chief Information Officer James Fleming, who worked with Faculty AI on this project, said: “AI is a fascinating and powerful technology, but it is often obscured by hype and jargon.
“This paper was developed as a result of a unique industry partnership with faculty to help a group of complete AI novices learn best practices in a very short time frame and see if they can be applied directly to their own science. Was born.
“The success of this project has not only proven that new insights can be derived in the process, but it has also been a catalyst for the rapid expansion of our own AI and software engineering team, with over 25 projects ‘in progress’. It also helped drive investment. There are various laboratories across the Crick River, with new projects starting each month. “
The researchers’ next step will be to understand disease subtypes in people with other genetic mutations, and see if sporadic Parkinson’s cases (i.e., no mutations) can be classified in a similar way. It is to be.
About this AI and Parkinson’s research news
author: claire green
sauce: Francis Crick Institute
contact: Claire Greene – Francis Crick Institute
image: Image credited to Neuroscience News
Original research: open access.
“Predicting Mechanistic Subtypes of Parkinson’s Disease Using Patient-derived Stem Cell Modelsby James Evans et al. nature machine intelligence
abstract
Predicting Mechanistic Subtypes of Parkinson’s Disease Using Patient-derived Stem Cell Models
Parkinson’s disease is a common and intractable neurodegenerative disease and is clinically heterogeneous, so different cellular mechanisms may drive the pathology in different individuals. So far, we have not been able to define the cellular mechanisms underlying neurodegenerative diseases in life.
We have generated a machine learning-based model that can simultaneously predict the presence of disease in human neurons and its major mechanistic subtypes. We have used stem cell technology to induce control or patient-derived neurons to generate different disease subtypes by chemical induction or the presence of mutations.
Multidimensional fluorescent labeling of organelles was performed on healthy control neurons and four different disease subtypes, independently training a set of classifiers using both quantitative single-cell fluorescent features and images. and built a deep neural network.
A quantitative cellular profile-based classifier achieves 82% accuracy, while an image-based deep neural network predicts controls and four different disease subtypes with 95% accuracy.
A machine learning-trained classifier exploits mitochondrial organelle features and the additional contribution of lysosomes to achieve accuracy across all subtypes, demonstrating the biological importance of these pathways in Parkinson’s disease. I back it up.
Taken together, we show that the machine learning approach applied to patient-derived cells is highly accurate in predicting disease subtypes, and the potential for this approach to enable mechanical stratification and precision medicine approaches in the future. provide proof of concept that there is
