Optimal Bayesian supervised domain adaptation for RNA sequencing data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Limited next-generation sequencing data hinders complex disease subtyping.
• Existing domain adaptation methods often require label correspondence or cannot use label information.
• Prior biological information, like gene networks, is typically not leveraged.

Purpose of the Study:

• To develop a generative Optimal Bayesian Supervised Domain Adaptation (OBSDA) model for improved prediction accuracy in target domains.
• To enable integration of RNA sequencing (RNA-Seq) data from multiple domains, accommodating shared or distinct labels.
• To incorporate prior biological knowledge, such as gene interaction networks, into the domain adaptation process.

Main Methods:

• Developed a hierarchical Bayesian negative binomial model with parameter factorization.
• Implemented an efficient Gibbs sampler for parameter inference in OBSDA.
• Utilized gene-gene network prior information with a flexible variational family for posterior inference.

Main Results:

• OBSDA demonstrated superior performance in identifying cancer subtypes by integrating cross-domain RNA-Seq data.
• The model successfully handles domains with shared or different label spaces.
• Incorporating gene network information further improved prediction accuracy.

Conclusions:

• OBSDA offers a robust framework for leveraging multi-domain RNA-Seq data for enhanced disease subtyping.
• The integration of prior biological networks significantly boosts the model's predictive power.
• This approach addresses limitations of existing methods by utilizing label information and network priors effectively.