Progeny Clustering: A Method to Identify Biological Phenotypes.

Area of Science:

• Computational biology
• Data science
• Bioinformatics

Background:

• Determining the optimal number of clusters is crucial for cluster analysis, particularly in high-dimensional biomedical datasets.
• Existing methods face challenges with complexity and computational efficiency.

Purpose of the Study:

• Introduce Progeny Clustering, a novel, stability-based, and computationally efficient algorithm.
• To accurately estimate the optimal number of clusters in diverse datasets, including complex biomedical data.

Main Methods:

• Progeny Clustering utilizes a novel Progeny Sampling method for cluster identity reconstruction.
• Employs a co-occurrence probability matrix to assess clustering stability.
• Incorporates reference datasets to mitigate inherent algorithmic and data space biases.

Main Results:

• Progeny Clustering demonstrated robust performance on synthetic, standard biological, and complex biomedical datasets.
• Outperformed existing methods on high-dimensional synthetic and cell phenotype datasets.
• Successfully identified clinically relevant patient clusters in an acute myeloid leukemia (AML) reverse phase protein array (RPPA) dataset.

Conclusions:

• Progeny Clustering offers an improved, efficient, and robust approach for determining the optimal number of clusters.
• The method shows particular promise for analyzing complex, high-dimensional biomedical data.
• It successfully identified clinically meaningful patient subgroups in AML RPPA data, highlighting its translational potential.