A Rule-Guided Community Detection Method for Identifying Subpopulations in Medical Data

Area of Science:

• Medical Informatics
• Network Science
• Data Mining

Background:

• Identifying homogeneous subpopulations is crucial for understanding disease subtypes in heterogeneous populations.
• Existing community detection methods often overlook association rules between attributes, which are vital for medical diagnosis.

Purpose of the Study:

• To propose a novel rule-guided community detection (RGCD) method for precise identification of homogeneous subpopulations in medical data.
• To integrate association rules into community detection to improve the accuracy of disease subtype identification.

Main Methods:

• Developed RGCD by incorporating association rules to construct an augmented network.
• Enhanced the transition probability matrix using rule-guided biased random walks, creating a rule-augmented matrix.
• Applied matrix decomposition and clustering to the rule-augmented matrix for subpopulation identification.

Main Results:

• RGCD demonstrated superior performance compared to six state-of-the-art community detection methods across 10 real-world medical datasets.
• Achieved up to a 22.62% increase in weighted F1 score compared to existing methods.
• Provided qualitative depictions of identified subpopulations, yielding medically significant insights.

Conclusions:

• RGCD is the first method to incorporate association rules into community detection for medical data analysis.
• The proposed method enables more precise identification of homogeneous subpopulations, aiding in disease subtype understanding.
• RGCD offers a promising approach for advancing medical diagnosis and personalized medicine.