Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease characterised by considerable heterogeneity in both its underlying biological mechanisms and clinical presentation. High-dimensional transcriptomic datasets offer an opportunity to characterise this variation at the molecular level; however, traditional statistical methods struggle with their scale and complexity. Machine learning approaches can reduce dimensionality and uncover latent patterns, enabling the identification of molecular subtypes that may refine prognosis and support patient stratification.Recent transcriptomic studies employing unsupervised machine learning have identified ALS subtypes with distinct molecular and clinical characteristics. Redefining ALS into more homogeneous molecular and clinical subtypes could transform all areas of ALS research by supporting novel experimental designs and precision medicine approaches. In this review, we summarise and critically assess these studies, discussing their findings, strengths, and limitations, and highlighting research gaps and challenges that should be addressed to enable their translation into biomedical and clinical practice.