A novel method for subgroup discovery in precision medicine based on topological data analysis.

Area of Science:

• Computational biology
• Data mining
• Topological data analysis

Background:

• The Mapper algorithm visualizes patient data as similarity graphs for disease understanding.
• Current limitations include complex parameter selection and manual graph analysis.
• This hinders routine application in real-world settings.

Purpose of the Study:

• To develop a novel algorithm for subgroup discovery within Mapper graphs.
• To address the challenge of optimal parameter selection for the Mapper algorithm.
• To identify homogenous and distinct patient subgroups (hotspots).

Main Methods:

• Proposed a hotspot detection algorithm for subgroup discovery in Mapper graphs.
• Integrated hotspot existence as a criterion for parameter space searching.
• Applied the algorithm to artificial and real-world breast cancer datasets.

Main Results:

• Successfully identified homogenous communities in the Two Circles dataset.
• Discovered a hotspot of ER+ breast cancer patients with poor prognosis and distinct gene expression.
• Confirmed findings in an independent breast cancer dataset.

Conclusions:

• The proposed method effectively facilitates subgroup discovery with pathology data.
• Enables identification of optimal Mapper algorithm parameters.
• Bridges the gap between Mapper's potential and translational research applications.