LACE-UP: An ensemble machine-learning method for health subtype classification on multidimensional binary data.

Area of Science:

• Biomedical research
• Machine learning
• Data science

Background:

• Subtype identification is crucial in biomedical research.
• Existing clustering methods struggle with multidimensional binary data.
• Lack of robust statistical methods limits subtype discovery.

Purpose of the Study:

• Introduce LACE-UP (Latent Class Analysis ENsembled with UMAP and PCA) for robust binary data clustering.
• Develop a method that does not require prespecifying the number of clusters.
• Address challenges like correlated and unrelated variables in subtype discovery.

Main Methods:

• Ensemble machine-learning approach combining Latent Class Analysis (LCA), Principal Components Analysis (PCA), and Uniform Manifold Approximation and Projection (UMAP).
• LCA provides model-based clustering.
• PCA offers spectral signal processing, and UMAP provides model-free dimensionality reduction.

Main Results:

• LACE-UP demonstrates superior performance compared to gold-standard techniques in simulations across various realistic data settings.
• The method is robust to correlated and extraneous variables.
• Application to UK Biobank dietary data revealed interpretable dietary subtypes linked to cardiovascular risk.

Conclusions:

• LACE-UP offers a powerful and robust solution for clustering multidimensional binary data.
• The method facilitates more accurate and interpretable disease and behavior subtype discovery.
• This approach has significant implications for understanding health behaviors and associated risks.