What is it about?
Several diseases such as Parkinson's manifest in a variety of different ways in different individuals. The 'subtype' that an individual manifests, and the appropriate treatment for their specific subtype, may depend on a variety of factors or symptoms that interact with each other and vary in time. These factors may be phenotypic, relating to an individual's demographic, genetic etc. Here we develop a novel, data-driven network based quantitative method that identifies subtypes by studying complex interactions between genetic information, demographics and symptoms of various types (motor function, cognitive function, sleep quality etc.). Not only does our method identify clinically relevant and stable subtypes, but is also predictive in nature. Specifically, we were able to identify the correct Parkinson's subtype that a person would develop into, FIVE years in advance with around 80% accuracy, which is pretty impressive given the slow timescale of the disease.
Featured Image
Photo by Milad Fakurian on Unsplash
Why is it important?
Complex diseases such as Parkinson's have several subtypes which are often difficult to identify. Early identification of an individual's subtype allows doctors to treat the person specifically for their subtype. While there is a variety of research on Parkinson's subtypes from a clinical perspective, there is no consensus, and much of this research is not predictive in nature. Our method provides very specific and detailed information about an individual's future predicted subtype. For instance, using our method, I might be able to look at a person that has recently been diagnosed with Parkinson's and predict, with somewhat high confidence, that they are likely to develop visual problems in approximately two years time, and memory issues in four years, however, they will likely not develop problems with reciting the language backwards...these types of individual-specific predictions, that are medically validated, allow us to make significant advances in early prediction and personalized preventive and pre-emptive treatment of Parkinson's! In addition, our methods are appropriate for a wide variety of other progressive diseases where a variety of factors interact in complex time-varying ways!
Perspectives
This article is really exciting to me since it involves a strong collaboration between people like myself with modeling, theoretical, and computational skills, as well as others that are Parkinson's clinicians that treat patients as part of their regular job. Working with clinicians made me deeply appreciate the importance of transparent and interpretable approaches that answer the kind of questions that people in biomedicine care about. This article led to a variety of clinicians working on other diseases contacting me, and ultimately led to several wonderful collaborations! I hope this article brings you new perspectives!
Sanjukta Krishnagopal
University of California Los Angeles
Read the Original
This page is a summary of: Identifying and predicting Parkinson’s disease subtypes through trajectory clustering via bipartite networks, PLoS ONE, June 2020, PLOS,
DOI: 10.1371/journal.pone.0233296.
You can read the full text:
Resources
Blogpost
Blogpost written by a member of the audience at my talk at the Oxford-Imperial seminar
Original paper
Free copy of the paper
Follow-up paper
Follow up paper elucidating some more involved and robust methods involving multi-layer networks
Paper on Stroke
Extension of these approaches to Stroke recovery subtypes in another paper.
Contributors
The following have contributed to this page
