Most phenotyping algorithms rely upon supervised or unsupervised statistical methods. Supervised phenotyping methods rely upon significant quantities of expensive expert-labels. Meanwhile, unsupervised methods are prone to learning spurious phenotypes. To address these limitations, we propose the Semi-Supervised Mixed Membership Model (SS3M) – a model for learning disease phenotypes from clinical data with few expert labels. We show SS3M can learn interpretable, disease-specific phenotypes which capture the clinical characteristics of the diseases specified by the labels provided.

Learning Objective: After participating in this session, the learner should:

- Understand the role of semi-supervised models in learning disease phenotypes from clinical data.
- Learn how Semi-Supervised Mixed Membership Models (SS3M) are related to popular models like Latent Dirichlet Allocation, and why they’re well suited for phenotype inference.
- Understand how a limited set of labels can be used to guide SS3M toward learning desired, disease-specific phenotypes from observational clinical data.

Authors:

Victor Rodriguez (Presenter)
Columbia University

Adler Perotte, Columbia University

Presentation Materials:

• Abstract/Manuscript - 1