Computational Approaches to High-Dimensional Omics-Driven Alzheimer’s Disease Subtyping: A Scoping Review

Introduction 

Alzheimer’s disease (AD) is biologically heterogeneous. High-dimensional omics may identify molecular strata relevant to precision medicine, but it remains unclear how studies differ in molecular inputs, computational methods, subtype outputs, and validation approaches. This scoping review mapped computational approaches to omics-driven AD and AD-spectrum subtyping. 

Methods 

A scoping review was conducted using Joanna Briggs Institute methodology. MEDLINE, EMBASE, and Web of Science searches identified 8,238 records. Two reviewers independently screened records using Covidence. From the broader review set, 44 omics studies were evaluated against refined criteria: an AD or AD-spectrum population, high-dimensional or broad-domain molecular input, computational individual-level subgrouping, and molecular subtype or stratification output. Data were extracted from 15 included studies regarding population, modality, algorithms, subtype outputs, validation, and clinical translation. Synthesis was descriptive. No meta-analysis was planned. 

Results 

Included studies were published between 2019 and 2025; 13/15 were published from 2022 onwards. Primary molecular data modalities were transcriptomics or RNA sequencing (7/15), multi-omics (2/15), genetics or imaging-genetics (2/15), cerebrospinal fluid proteomics (1/15), plasma lipidomics (1/15), microbiome sequencing (1/15), and high-dimensional blood or plasma markers (1/15). All studies used clustering; 13/15 used hybrid analytical pipelines, and 6/15 explicitly incorporated supervised machine learning (ML).  

Derived subtype numbers ranged from 2 to 5: two subtypes in 5/15 studies, three in 5/15, four in 3/15, and five in 2/15. Recurrent subtype themes involved immune/inflammatory, metabolic, synaptic, autophagy/programmed-cell-death, blood-brain barrier/vascular, genetic, and atrophy patterns. All studies reported internal evaluation. External validation was reported in 10/15 but often assessed diagnostic signatures or risk scores rather than independently replicating subtype solutions. All studies discussed potential clinical utility; none reported clinical implementation. 

Conclusions 

High-dimensional omics-driven stratification reveals recurrent molecular heterogeneity across AD and AD-spectrum cohorts, but evidence remains discovery-focused. Key translational barriers include validation targets, subtype reproducibility, and clinical usability.