PFClust: a novel parameter free clustering algorithm.

Area of Science:

• Data Science
• Bioinformatics
• Computational Biology

Background:

• Automated data clustering and determination of the optimal number of clusters remain significant challenges.
• Existing clustering methodologies often require user-defined parameters, limiting their applicability.
• PFClust is introduced as a parameter-free algorithm for automated data clustering.

Purpose of the Study:

• To develop and validate an automated clustering algorithm that does not require user-specified parameters.
• To assess the performance of PFClust against established clustering methods.
• To evaluate PFClust's ability to cluster complex biological data.

Main Methods:

• PFClust partitions datasets into clusters based on shared attributes like minimum expectation value and intra-cluster similarity variance.
• The algorithm was tested on synthetic 2D vector datasets and real-world protein domain structures from the CATH database.
• Performance was compared against six other leading clustering methodologies.

Main Results:

• PFClust demonstrated clustering performance at least equal to, and on average slightly better than, six other leading methods.
• Five of the compared methods required pre-specified cluster numbers, unlike PFClust.
• PFClust showed excellent agreement with known classifications for protein domain structures.

Conclusions:

• PFClust successfully automates data clustering and optimal cluster number identification without parameter input.
• The algorithm generates meaningful clusters for synthetic data and accurately classifies protein domain structures.
• PFClust offers a robust and user-independent solution for data clustering challenges.