Evolutionary Multiobjective Clustering Algorithms With Ensemble for Patient Stratification.

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Patient stratification is crucial for effective disease treatment, but faces challenges due to data dimensionality and interpretability.
• Existing stratification models struggle with high diagnostic ability and generalization.

Purpose of the Study:

• To propose novel evolutionary multiobjective clustering algorithms with ensemble (NSGA-II-ECFE and MOEA/D-ECFE) for improved patient stratification.
• To enhance ensemble diversity and develop a robust ensemble clustering fitness evaluation (ECFE) method.

Main Methods:

• Developed two novel evolutionary multiobjective clustering algorithms (NSGA-II-ECFE, MOEA/D-ECFE) using four cluster validity indices.
• Introduced an ensemble construction method for diversity and an ECFE method using hybrid co-association matrices for consensus clustering.
• Dynamically selected clustering algorithms based on the co-association matrix.

Main Results:

• The proposed algorithms demonstrated superior performance compared to seven clustering, twelve ensemble clustering, and two multiobjective clustering algorithms.
• Experiments on 55 synthetic and 35 real patient stratification datasets confirmed the algorithms' effectiveness.
• Successfully applied the algorithms to identify cancer subtypes from five single-cell RNA-seq datasets.

Conclusions:

• The novel ensemble clustering algorithms offer significant improvements in patient stratification and diagnostic ability.
• These algorithms provide a powerful tool for identifying disease subtypes, including cancer, from complex biological data.