Unsupervised machine learning identifies distinct ALS molecular subtypes in post-mortem motor cortex and blood
Heather Marriott1,2, Renata Kabiljo2, Guy P Hunt1,2,3,4
1Department of Basic and Clinical Neuroscience, Maurice Wohl Clinical Neuroscience Institute, Institute of Psychiatry, Psychology and Neuroscience, King?s College London, London, SE5 9NU, UK.
Acta Neuropathologica Communications
|December 22, 2023
Summary
Machine learning identified three molecular subtypes of amyotrophic lateral sclerosis (ALS) based on gene expression. These subtypes reflect distinct disease mechanisms and offer potential for personalized ALS treatments.
Area of Science:
- Neuroscience
- Genetics
- Computational Biology
Background:
- Amyotrophic lateral sclerosis (ALS) exhibits significant clinical and genetic variability.
- Previous machine learning efforts for ALS patient stratification lacked independent validation across diverse populations and tissues.
- This study addresses the need for robust patient stratification in ALS research.
Purpose of the Study:
- To stratify amyotrophic lateral sclerosis (ALS) patients into distinct molecular subtypes using gene expression data.
- To validate these subtypes across different tissue types (motor cortex, blood) and populations.
- To explore the potential of these subtypes for personalized treatment strategies in ALS.
Main Methods:
- Hierarchical clustering was applied to gene expression data from sporadic ALS patients (KCL BrainBank) focusing on the 5000 most variable autosomal genes.
- Linear discriminant analysis was used for cluster validation on independent datasets, including US, Italian, and Dutch cohorts (motor cortex and blood).
- Logistic regression classifiers were built to assess the specificity and discriminatory power of identified gene expression signatures for ALS and motor cortex involvement.
Main Results:
- Three molecular phenotypes were identified: synaptic and neuropeptide signalling, oxidative stress and apoptosis, and neuroinflammation.
- High assignment probabilities (80-90%) confirmed subtype validity across independent datasets.
- Gene expression signatures effectively distinguished ALS cases from controls (AUC 0.88±0.10) and reflected motor cortex specificity, with perfect discrimination from other brain regions.
- Cell type proportions within clusters aligned with known biological processes, supporting biological interpretation.
- Distinct cluster-related outcomes were observed, correlating with disease onset and progression measures.
Conclusions:
- The findings support the existence of distinct pathogenic mechanisms underlying ALS in patient subgroups.
- The identified molecular subtypes have the potential to guide the development of personalized therapeutic approaches for ALS.
- A validated computational method for ALS gene expression clustering is publicly available for the scientific and clinical community.
