Bayesian clustering of DNA sequences using Markov chains and a stochastic partition model

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Clustering DNA sequences is vital for identifying organismal units like species in biological applications.
• Metagenomics requires grouping short, error-containing DNA sequences, necessitating robust statistical methods.

Purpose of the Study:

• To introduce a novel stochastic partition model for clustering DNA sequences based on Markov chains.
• To address the challenges of clustering short, error-prone sequences in metagenomics.

Main Methods:

• Developed a Dirichlet process prior-based stochastic partition model for clustering Markov chains.
• Employed conjugate priors for Markov chain parameters to enable analytical marginal likelihood comparisons.
• Utilized a hybrid EM-algorithm and greedy search for efficient posterior mode approximation.

Main Results:

• The novel method demonstrates faster performance and higher accuracy than existing clustering techniques for metagenomics.
• The model effectively clusters Markov chains, applicable to various sequence data types.
• Experiments on Escherichia coli strain shotgun sequence data validate the model's utility.

Conclusions:

• The proposed stochastic partition model offers an accurate and efficient solution for DNA sequence clustering in metagenomics.
• The model's generic nature allows for broader applications in sequence data analysis.
• This method advances the ability to classify microbial communities and analyze genomic data.