Bayesian integrative model for multi-omics data with missingness.

Area of Science:

• Genomics
• Biostatistics
• Computational Biology

Background:

• Multi-omics data integration offers insights into complex diseases.
• Existing methods struggle with missing data, leading to power loss.
• Complete case analysis discards valuable samples with incomplete omics profiles.

Purpose of the Study:

• To develop a full Bayesian model for multi-omics data integration that accommodates missing samples.
• To enhance statistical power and marker detection in complex disease studies.
• To propose a decision scheme for incorporating samples with missing data.

Main Methods:

• A full Bayesian model inspired by iBAG (Integrative Bayesian Analysis of Genomics data) was developed.
• The model incorporates samples with missing omics data.
• A self-learning cross-validation (CV) decision scheme was proposed to manage missingness.

Main Results:

• The Bayesian approach improved prediction accuracy and feature selection with limited sample size and high missingness.
• Incorporating missing data can decrease performance when sample size is large or missingness is low.
• The CV decision scheme demonstrated superior performance across various missing data mechanisms.

Conclusions:

• The proposed Bayesian model effectively integrates incomplete multi-omics data.
• The CV decision scheme provides a robust method for determining the inclusion of samples with missing data.
• This approach enhances the analysis of complex diseases by maximizing data utilization.