Neurodegenerative diseases can progress in newly identified patterns

Neurodegenerative diseases – such as amyotrophic lateral sclerosis (ALS or Lou Gehrig’s disease), Alzheimer’s and Parkinson’s disease – are complicated, chronic conditions that can present with a variety of symptoms, worsen at different rates, and have many underlying genetic and environmental causes, some of which are unknown. ALS in particular affects voluntary muscle movement and is always fatal, but while most people survive only a few years after diagnosis, others live with the disease for decades. Manifestations of ALS can also vary significantly; Often slower disease progression correlates with limb onset and impairs fine motor skills, while more severe bulbar ALS impairs swallowing, speech, breathing, and mobility. Therefore, understanding the progression of diseases like ALS is crucial for enrollment in clinical trials, analyzing possible interventions, and discovering root causes.

However, assessing the development of the disease is anything but easy. Current clinical studies typically assume that health deteriorates linearly down a symptom rating scale and use these linear models to assess whether drugs slow disease progression. However, the data indicate that ALS often follows non-linear courses, with periods of stable symptoms alternating with periods of rapid change. Because data can be sparse and health assessments often rely on subjective assessment metrics measured at uneven time intervals, comparisons between patient populations are difficult. This heterogeneous data and history, in turn, complicates analyzes of the invention’s efficacy and may obscure the origin of the disease.

Now, a new machine learning method developed by researchers at MIT, IBM Research and others aims to better characterize the patterns of ALS disease progression to inform clinical trial design.

“There are groups of individuals who share patterns of progression. For example, some seem to have really rapidly progressive ALS and others have slowly progressive ALS that changes over time,” says Divya Ramamoorthy PhD ’22, a research specialist at MIT and lead author of a new article about the work that was published this month in natural informatics. “The question we asked ourselves is: can we use machine learning to determine whether and to what extent these types of consistent patterns exist in people?”

Indeed, their technique identified discrete and robust clinical patterns in ALS progression, many of which are nonlinear. Furthermore, these subtypes of disease progression were consistent across all patient populations and disease metrics. The team also found that their method can also be used for Alzheimer’s and Parkinson’s disease.

Alongside Ramamoorthy on paper are MIT-IBM Watson AI Lab members, Ernest Fraenkel, professor in the MIT Department of Biological Engineering; research scientist Soumya Ghosh of IBM Research; and senior research scientist Kenney Ng, also of IBM Research. Additional authors include Kristen Severson PhD ’18, senior researcher at Microsoft Research and former member of Watson Lab and IBM Research; Karen Sachs PhD ’06 of Next Generation Analytics; a research team using Answer ALS; Jonathan D. Glass and Christina N. Fournier from Emory University School of Medicine; the Pooled Resource Open-Access ALS Clinical Trials Consortium; ALS/MND Natural History Consortium; Todd M. Herrington of Massachusetts General Hospital (MGH) and Harvard Medical School; and James D. Berry of MGH.

Redesign health care

After consulting with clinicians, the team of machine learning researchers and neurologists let the data speak for themselves. They designed an unsupervised machine learning model that used two methods: Gaussian process regression and Dirichlet process clustering. These derived the health trajectories directly from the patient data and automatically grouped similar trajectories without specifying the number of clusters or the shape of the curves, forming “subtypes” of ALS progression. Their method took into account prior clinical knowledge in the form of a negative trend—in line with expectations for neurodegenerative disease—but did not assume linearity. “We know that the linearity does not reflect what is actually observed,” says Ng. “The methods and models we use here were more flexible, in the sense that, They “capture what was seen in the data” without the need for expensive labeled data and preset parameters.

They primarily applied the model to five longitudinal data sets from ALS clinical trials and observational studies. These used the gold standard for measuring symptom development: the Revised ALS Functional Rating Scale (ALSFRS-R), which captures a global picture of the patient’s neurological impairment but can be a somewhat “messy metric.” In addition, performance related to survival probabilities, forced vital capacity (a measure of respiratory function) and fractional scores of ALSFRS-R, which examines individual bodily functions, were included.

New progression and utility regimes

When their population-level model was trained and tested with these metrics, four dominant disease patterns emerged from the many trajectories—sigmoidal rapid progression, stable slow progression, unstable slow progression, and unstable moderate progression—many with strong nonlinear characteristics. In particular, it captured trajectories where patients experienced a sudden loss of ability, known as the functional cliff, which would have a significant impact on treatments, enrollment in clinical trials and quality of life.

The researchers compared their method to other commonly used linear and nonlinear approaches in the field to separate the contribution of clustering and linearity to the accuracy of the model. The new work outperformed them, even patient-specific models, and found that the subtype patterns were consistent across all measurements. Impressively, the model was able to interpolate missing values ​​when data was withheld and, crucially, predict future health actions. The model could also be trained on an ALSFRS-R dataset and predict cluster membership in others, making it robust, generalizable, and accurate with scarce data. As long as data were available for 6-12 months, health histories could be derived with greater reliability than with conventional methods.

The researchers’ approach also provided insight into Alzheimer’s and Parkinson’s disease, both of which can present a range of symptom presentations and courses. In Alzheimer’s, the new technique could identify different disease patterns, specifically differences in conversion rates from mild to severe disease. The Parkinson’s analysis showed a relationship between courses of progression for scores without medication and disease phenotypes, such as e.g.

Work is making significant headway in finding the signal amidst the noise in the time series of complex neurodegenerative diseases. “The patterns we’re seeing are reproducible across studies that I don’t think have been shown before, and that may have implications for how we subtype them.” [ALS] illness,” says Frankel. Because the FDA has considered the implications of nonlinearity in clinical trial designs, the team finds their work to be particularly relevant.

As new ways of understanding disease mechanisms come online, this model provides another tool to distinguish diseases such as ALS, Alzheimer’s and Parkinson’s from a systems biology perspective.

“We have a lot of molecular data from the same patients, and so our long-term goal is to see if there are subtypes of the disease,” says Fraenkel, whose lab studies cellular changes to understand the etiology of diseases and possible targets for cures. “One approach is to start with the symptoms … and see if people with different patterns of disease progression are also different at the molecular level. This could lead you to therapy. Then there’s the bottom-up approach, where you start with the molecules’ and try to reconstruct biological pathways that might be affected. “We go [to be tackling this] from both ends… and figuring out if anything meets in the middle.”

This research was supported in part by the MIT-IBM Watson AI Lab, the Muscular Dystrophy Association, the Department of Veterans Affairs of Research and Development, the Department of Defense, the NSF Gradate Research Fellowship Program, Siebel Scholars Fellowship, Answer ALS, the acquisition activities US Army Medical Research, National Institutes of Health and NIH/NINDS.