Machine learning predicts distinct biotypes of amyotrophic lateral sclerosis
Nicholas Pasternack1,2, Ole Paulsen2, Avindra Nath3
1Section of Infections of the Nervous System, National Institute of Neurological Disorders and Stroke (NINDS), National Institutes of Health (NIH), Bethesda, MD, USA.
European Journal of Human Genetics : EJHG
|August 7, 2025
Summary
Researchers identified three distinct subtypes of amyotrophic lateral sclerosis (ALS) using machine learning. These subtypes, characterized by synaptic dysfunction, neuronal regeneration, or degeneration, offer potential new therapeutic targets for this fatal neurodegenerative disease.
Area of Science:
- Neuroscience
- Genomics
- Computational Biology
Background:
- Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease with no cure.
- Heterogeneity in clinical presentation and pathology hinders therapeutic development for ALS.
Purpose of the Study:
- To identify biologically distinct subtypes of ALS using transcriptomic data.
- To investigate potential novel therapeutic targets within identified ALS subtypes.
- To develop a predictive model for ALS subtypes based on clinical and demographic data.
Main Methods:
- Analysis of bulk and single-cell transcriptomic data from postmortem ALS patient cohorts.
- Application of unsupervised machine learning to identify patient subgroups.
- Development of a supervised machine learning model for subtype prediction.
Main Results:
- Identification of three ALS patient subtypes: synaptic dysfunction (34%), neuronal regeneration (47%), and neuronal degeneration (19%).
- Each subtype exhibits unique transcriptional dysregulation patterns.
- A supervised model achieved approximately 80% accuracy in predicting ALS subtype from clinical and demographic data.
Conclusions:
- Established three biologically distinct ALS subtypes.
- Identified potential novel therapeutic targets associated with each subtype.
- Demonstrated the feasibility of predicting ALS subtypes using clinical and demographic information.
