Alignment constrained sampling.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Generating realistic randomized biological sequences is crucial for benchmarking computational tools.
• Existing methods often require extensive multiple sequence alignment data for training.
• Accurate homology detection is vital for understanding gene function and evolutionary relationships.

Purpose of the Study:

• To introduce ALICO (ALIgnment COnstrainted), a novel framework for generating randomized multiple sequence alignments.
• To develop a method that preserves essential alignment features, including dependence structure and pairwise sequence identities.
• To demonstrate ALICO's utility in evaluating and improving homology detection algorithms.

Main Methods:

• ALICO generates randomized alignments by preserving pairwise sequence identities and dependence structures.
• The framework utilizes only pairwise alignment training data, reducing data requirements.
• ALICO's performance was evaluated using the MacIsaac orthologous yeast dataset and compared against established homology finders.

Main Results:

• ALICO samples approximately preserve average pairwise percent identities (PIDs) of the input alignment.
• Generated sequences exhibit k-mer composition resembling genomic training data.
• ALICO-derived p-values enhance the performance of combined homology detection methods, often surpassing individual tools.

Conclusions:

• ALICO provides a flexible and efficient framework for generating biologically relevant randomized alignments.
• The method effectively supports the evaluation of homology detection tools and facilitates improved data analysis.
• ALICO's ability to leverage pairwise data makes it broadly applicable in bioinformatics research.