A tree-based model for homogeneous groupings of multinomials.

Area of Science:

• Statistics
• Bioinformatics
• Computational Biology

Background:

• Grouping multinomial data is crucial for various analytical tasks.
• Existing methods may face challenges with large datasets or complex probability distributions.
• Understanding patterns in genetic sequences requires effective data clustering.

Purpose of the Study:

• To develop a tree-based methodology for classifying multinomial data based on probability vectors.
• To offer a scalable approach for handling a large number of multinomial observations.
• To facilitate the analysis and alignment of biological sequences through homogeneous grouping.

Main Methods:

• Binary recursive partitioning to construct an initial classification tree.
• Maximizing the likelihood function to determine optimal data splits.
• Bottom-up tree pruning using hypothesis testing, Bayesian Information Criterion (BIC), and Wilcoxon rank-sum test.

Main Results:

• A robust tree-based model for grouping multinomial data was developed.
• The method effectively handles large datasets by ordering and reducing potential splits.
• The model successfully identified homogeneous groupings in genetic sequence data.

Conclusions:

• The proposed tree-based method provides an effective way to cluster multinomial data.
• This technique offers new possibilities for understanding and aligning genetic sequences.
• The approach demonstrates utility in bioinformatics and computational biology applications.