Unsupervised machine learning identifies distinct molecular and phenotypic ALS subtypes in post-mortem motor cortex and blood expression data

• DOI: 10.1101/2023.04.21.23288942
• Publication Date: 2023-04-25T18:55:12Z
• Authors: Heather Marriott; Renata Kabiljo; Guy P Hunt; Ahmad Al Khleifat; Ashley Jones; Claire Troakes; ; ; Abigail L Pfaff; John P Quinn; Sulev Koks; Richard J Dobson; Patrick Schwab; Ammar Al-Chalabi; Alfredo Iacoangeli
• Keywords: Hierarchical clustering

ABSTRACT Background Amyotrophic lateral sclerosis (ALS) displays considerable clinical, genetic and molecular heterogeneity. Machine learning approaches have shown potential to disentangle complex disease landscapes they been utilised for patient stratification in ALS. However, lack of independent validation different populations pre-mortem tissue samples greatly limited their use clinical research settings. We overcame such issues by performing a large-scale study over 600 post-mortem brain blood people with ALS from four datasets the UK, Italy, Netherlands US. Methods Hierarchical clustering was performed on 5000 most variably expressed autosomal genes identified motor cortex expression data sporadic KCL BrainBank (N=112). The architectures each cluster were investigated gene enrichment, network cell composition analysis. Methylation also used assess if other omics measures differed between individuals. Validation these clusters achieved applying linear discriminant analysis models based TargetALS US (N=93), as well Italian (N=15) Dutch (N=397) datasets. Phenotype cluster-specific differences outcomes. Results three phenotypes, which reflect proposed major mechanisms pathogenesis: synaptic neuropeptide signalling, excitotoxicity oxidative stress, neuroinflammation. Known risk among informative cluster, suggesting profiling phenotypes. Cell types are known be associated specific phenotypes found higher proportions those clusters. These validated distinct cluster-related outcomes progression, survival age death. developed public webserver ( https://alsgeclustering.er.kcl.ac.uk ) that allows users stratify our model uploading data. Conclusions driven types, pathogenesis. Our results support hypothesis biological heterogeneity where underly pathogenesis subgroup patients can signature. show trials, development biomarkers personalised treatment approaches.