Handling missing rows in multi-omics data integration: multiple imputation in multiple factor analysis framework.

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Omics data integration often suffers from missing individuals across datasets.
• Missing data prevents direct application of most statistical methods, hindering analysis.
• Multiple Factor Analysis (MFA) is a key tool for integrating multi-omics data.

Purpose of the Study:

• To propose a novel multiple imputation (MI) approach within a multivariate framework for omics data integration.
• To address challenges posed by missing individuals in multi-omics datasets using Multiple Factor Analysis (MFA).
• To develop a method that accurately estimates individual coordinates in MFA despite missing data.

Main Methods:

• Multiple Imputation (MI) was employed to fill missing rows, creating M completed datasets.
• Multiple Factor Analysis (MFA) was applied to each completed dataset.
• The M resulting configurations were combined to generate a single consensus solution (MI-MFA).

Main Results:

• The MI-MFA method was validated on real omics datasets with artificially introduced missingness.
• MI-MFA performance was compared against regularized iterative MFA (RI-MFA) and mean variable imputation (MVI-MFA).
• MI-MFA configurations closely approximated the true MFA configuration, even with significant missing data. Uncertainty due to missingness was quantified using confidence ellipses and convex hulls.

Conclusions:

• MI-MFA offers a robust method for estimating individual coordinates in MFA with incomplete omics data.
• The approach effectively handles missing rows, maintaining accuracy even with substantial data gaps.
• MI-MFA accounts for imputation uncertainty, enabling reliable evaluation of results.