Clustering biological sequences with dynamic sequence similarity threshold.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Biological sequence clustering faces challenges due to high computational costs of sequence alignments and difficulties in parameter selection.
• Existing methods often rely on a single, fixed sequence identity threshold, potentially leading to low-quality clusters.
• Limited user support exists for selecting appropriate thresholds for diverse biological sequence datasets.

Purpose of the Study:

• To develop a novel, efficient, and robust biological sequence clustering approach.
• To overcome limitations of fixed-threshold methods in sequence clustering.
• To provide automated threshold determination for improved cluster quality.

Main Methods:

• Implemented ALFATClust, a novel sequence clustering approach.
• Utilized rapid, alignment-free pairwise sequence distance calculations.
• Employed community detection in graphs for cluster generation.
• Dynamically determined cluster-specific cut-off thresholds based on cluster separation and intra-cluster similarity.

Main Results:

• ALFATClust outperforms existing sequence clustering approaches on benchmark datasets.
• The method simultaneously maintains cluster robustness and substantial cluster separation.
• Dynamically determined thresholds enhance the quality of generated clusters.
• An evaluation report is provided for verifying non-singleton cluster quality.

Conclusions:

• ALFATClust effectively generates high-quality sequence clusters with strong intra-cluster similarity and clear inter-cluster separation.
• The approach eliminates the need for users to manually specify precise similarity cut-off thresholds.
• ALFATClust offers an automated and robust solution for biological sequence clustering.