Reproduce “Clustering Interval-Censored Time-Series for Disease Phenotyping” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Irene Y. Chen, Rahul G. Krishnan, David Sontag
Researchers from MIT and the University of Toronto developed SubLign, a deep generative model that simultaneously clusters time-series data and corrects for interval censoring. The method accurately identifies underlying disease subtypes and their true progression patterns, achieving an Adjusted Rand Index of 0.94 on synthetic sigmoid data and successfully recovering known heart failure phenotypes from clinical records.
Key insights
- The model provides theoretical guarantees for the identifiability of both latent subtypes and alignment values under specific noiseless conditions.
- SubLign offers a flexible and model-agnostic approach to handling delayed entry, avoiding strong assumptions about its distribution beyond its extrema.
- By modeling time-series in a continuous manner and accommodating variable sequence lengths, it naturally addresses complexities inherent in real-world clinical data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.