Identifying homogeneous subgroups of patients and important features: a topological machine learning approach

Area of Science:

• Computational topology
• Data science
• Machine learning

Background:

• Topological data analysis (TDA) offers advanced methods for understanding complex datasets.
• The Mapper algorithm is a key TDA tool for reducing high-dimensional data to a graph representation.

Purpose of the Study:

• To present a computational pipeline for data clustering using the Mapper algorithm.
• To leverage statistically significant topological features for robust cluster identification and summarization.

Main Methods:

• Utilizing the Mapper algorithm for data dimensionality reduction and graph construction.
• Applying statistical methods to identify significant topological features within the generated graph.
• Integrating prior knowledge and machine learning for cluster optimization and analysis.

Main Results:

• A functional pipeline for clustering point cloud data using Mapper.
• Identification and summarization of clusters based on robust topological features.
• Demonstrated ability to incorporate mixed data types and prior knowledge.

Conclusions:

• The pipeline effectively clusters data by identifying statistically significant topological features.
• Key strengths include integration of prior knowledge, bootstrap for feature robustness, machine learning for inspection, and mixed data type handling.
• The pipeline is available as open-source software under the GNU GPLv3 license.