Phenomapping of Patients with Primary Breast Cancer Using Machine Learning-Based Unsupervised Cluster Analysis

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Primary breast cancer (PBC) presents significant heterogeneity across clinical, histopathological, and molecular features.
• Accurate classification of PBC is crucial for identifying patient subgroups and optimizing management strategies.
• Machine learning offers potential for deciphering complex relationships within heterogeneous clinical data.

Purpose of the Study:

• To explore the utility of unsupervised learning techniques for enhancing the classification of primary breast cancer.
• To identify distinct patient subgroups within PBC based on biological prognostic parameters.

Main Methods:

• Utilized a dataset comprising 712 women diagnosed with primary breast cancer.
• Applied four unsupervised clustering methods: K-means, self-organising maps, hierarchical agglomerative clustering (HAC), and Gaussian mixture models.
• Evaluated clustering performance, with HAC demonstrating superior results.

Main Results:

• Hierarchical agglomerative clustering identified two distinct patient clusters with differing clinical profiles.
• Cluster 1 comprised younger patients with lower estrogen receptor (ER) and progesterone receptor (PgR) values and lower cathepsin D levels.
• Cluster 2 included older patients with higher ER and PgR values. Age, ER, and PgR were identified as the most significant variables by HAC.

Conclusions:

• Unsupervised learning, particularly HAC, is a valuable tool for analyzing heterogeneous PBC data.
• This approach offers new perspectives for dissecting clinical heterogeneity in breast cancer research.
• The identified clusters highlight distinct patient profiles that may inform personalized treatment strategies.