Pattern discovery and disentanglement on relational datasets.

Area of Science:

• Biomedical Machine Learning
• Data Mining
• Genomics

Background:

• Machine learning shows promise in cognitive tasks but struggles with relational datasets.
• Challenges include low volume, imbalanced groups, mislabeled data, and lack of output interpretability.
• Subtle functional and statistical relation entanglement hinders performance in complex datasets.

Purpose of the Study:

• To introduce the Pattern Discovery and Disentanglement System (PDD) for explicit pattern discovery.
• To address limitations in handling data with varying sizes, imbalanced groups, and anomalies.
• To improve transparency and interpretability in machine learning models for biomedical data.

Main Methods:

• Development of the Pattern Discovery and Disentanglement System (PDD).
• Application of PDD to four diverse biomedical datasets.
• Creation of the PDD Knowledge Base for explicit representation linking sources, patterns, and patients.

Main Results:

• PDD successfully discovers explicit patterns from data with various sizes and imbalanced groups.
• The system effectively screens out anomalies, improving data quality.
• Case studies demonstrate enhanced prediction accuracy and transparent interpretation of findings.

Conclusions:

• PDD offers a robust solution for machine learning challenges in relational biomedical datasets.
• The system facilitates transparent interpretation through the PDD Knowledge Base.
• PDD holds significant potential for broad applications in genomic and biomedical machine learning.