Exploiting the functional and taxonomic structure of genomic data by probabilistic topic modeling

Area of Science:

• Computational Biology
• Microbiome Research
• Genomics

Background:

• Understanding the functional core of microbial communities and genomes is crucial for biological research.
• Existing homology-based and composition-based approaches have limitations in comprehensively identifying these cores.
• Generative topic modeling offers a novel approach to extract information from unlabeled biological data.

Purpose of the Study:

• To present a unified method using generative topic modeling for studying both microbial and gene cores.
• To leverage generative topic modeling for analyzing taxon abundance and N-mer features.
• To identify and explain the functional roles of latent topics in microbial and genomic data.

Main Methods:

• Applied generative topic modeling to taxon abundance data from a homology-based approach to study the microbial core.
• Modeled samples as documents with mixtures of functional groups (latent topics), where each topic is a mixture of species.
• Applied generative topic modeling to N-mer features from a composition-based approach to study the gene core.
• Modeled genomes as mixtures of latent genetic patterns (topics), where each pattern is a mixture of N-mer features.
• Analyzed mutual information between latent topics and gene regions to interpret functional roles.

Main Results:

• Generative topic modeling successfully modeled taxon abundance, revealing the distribution of latent functions within microbial samples.
• The method identified common N-mer features indicative of core genomes by modeling genomes as mixtures of genetic patterns.
• Functional roles of uncovered latent genetic patterns were explained through analysis of mutual information with gene regions.
• Experimental results validated the effectiveness of the proposed generative topic modeling approach.

Conclusions:

• Generative topic modeling provides a powerful, unified framework for studying functional cores in both microbial and genomic data.
• The method effectively extracts meaningful functional and genetic patterns from unlabeled biological datasets.
• This approach enhances our ability to understand microbial community structure and genome composition.