Disease2Vec: Encoding Alzheimer's progression via disease embedding tree.

Area of Science:

• Neuroscience
• Computational Biology
• Medical Informatics

Background:

• Traditional Alzheimer's Disease (AD) prediction models often overlook the continuous nature of disease progression.
• Existing methods primarily focus on binary or multi-class classification, failing to capture nuanced transitions between clinical stages.
• Previous AD progression models primarily aimed to order biomarkers rather than predict individual patient status along a continuum.

Purpose of the Study:

• To develop a novel learning-based embedding framework for modeling Alzheimer's Disease (AD) progression.
• To represent the intrinsic relationships among AD clinical stages in a continuous latent space.
• To enable accurate prediction of individual patient status across the full spectrum of AD development.

Main Methods:

• Developed Disease2Vec, a novel learning-based embedding framework to encode relationships among AD clinical stages.
• Generated a disease embedding tree (DETree) to represent AD progression as a continuous trajectory.
• Utilized DETree to predict individual patient clinical status by projecting them onto the continuous trajectory.

Main Results:

• The Disease2Vec framework and its generated DETree effectively represent AD progression.
• The model enables efficient and accurate prediction of patient status across five fine-grained clinical groups.
• The approach provides richer status information by analyzing patient positions within the continuous progression trajectory.

Conclusions:

• The developed disease embedding framework (Disease2Vec) offers a significant advancement in modeling Alzheimer's Disease (AD) progression.
• DETree provides a continuous trajectory for predicting individual patient status more accurately and with greater detail.
• This novel approach addresses the understudied problem of predicting patient status within the spectrum of continuous AD development.