An unsupervised machine learning method for discovering patient clusters based on genetic signatures.

Area of Science:

• Genomics
• Machine Learning
• Personalized Medicine

Background:

• Chronic disorders exhibit significant patient-to-patient variability in etiology, progression, and treatment response.
• This variability necessitates identifying clinically relevant patient subgroups for personalized medicine.
• Existing unsupervised methods often struggle with identifying robust clusters and require a priori parameter input.

Purpose of the Study:

• To develop an unsupervised machine learning method for clustering patients based on genomic data.
• To enable algorithmic identification of the number of clusters using internal validity metrics.
• To avoid the need for researchers to specify input parameters beforehand.

Main Methods:

• Utilizes an unsupervised machine learning approach for patient clustering.
• Employs internal validity metrics to determine the optimal number of clusters.
• Leverages linkage disequilibrium between single nucleotide polymorphisms for enhanced analysis.
• Performs gene pathway analysis to interpret biological relationships between clusters.

Main Results:

• The proposed method demonstrated superior performance compared to existing benchmarks.
• Successfully identified genetically distinct patient clusters in a multiple sclerosis dataset without a priori parameters.
• Discovered significant genetic variants enriched in immune processes and cell adhesion pathways via Gene Ontology analysis.

Conclusions:

• The developed method effectively clusters patients based on genomic makeup, aiding personalized medicine.
• Identified clusters show enrichment in biologically relevant pathways, suggesting clinical significance.
• Future work can link these clusters to clinical outcomes for predictive value in chronic diseases.