The abstract is read at the publisher; the summary is JClub's.
ADSMoE, a multimodal mixture-of-experts transformer autoencoder, identifies four Alzheimer’s disease subtypes from incomplete real-world data in 5,422 participants, with distinct neuroimaging, genetic, and proteomic patterns.
- Why it matters: AD heterogeneity limits precision medicine, and prior clustering or biomarker-trajectory methods struggle when modalities are missing or combinations vary across patients.
- What they did: The model uses a Fleximodal MoE encoder for arbitrary available modality sets and a subtype-specific MoE decoder, trained on ADNI and validated externally in NACC and UKB without retraining.
- The result: Benchmarking shows robust subtype discovery with biological and clinical relevance, and external validation finds consistent subtype-associated characteristics across independent cohorts.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.