Batch correction evaluation framework using a-priori gene-gene associations: applied to the GTEx dataset.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Heterogeneous datasets often contain artefacts from multiple confounders, necessitating batch effect correction in bioinformatics.
• Removing technical variation can inadvertently eliminate biologically meaningful signals, posing a challenge for researchers.
• Assessing the impact of batch correction on biological data integrity is essential.

Purpose of the Study:

• To introduce a novel framework, B-CeF, for evaluating the effectiveness of batch correction methods.
• To assess the tendency of batch correction methods toward over- or under-correction.
• To determine if batch correction preserves biological signals in RNA-seq data.

Main Methods:

• The B-CeF framework compares gene-gene co-expression in adjusted datasets against a gold standard of known associations.
• It involves data adjustment, co-expression measurement calculation, and performance evaluation against an external reference.
• The framework was applied to evaluate five batch correction methods on GTEx RNA-seq tissue datasets.

Main Results:

• B-CeF effectively evaluates batch correction methods and their impact on biological signals.
• Multiple linear regression models for correcting known confounders demonstrated superior performance compared to factor analysis-based methods.
• The study identified that linear regression better preserves biological signals than methods estimating hidden confounders.

Conclusions:

• The B-CeF framework provides a robust method for evaluating batch correction techniques in bioinformatics.
• It highlights the superiority of linear regression models in preserving biological signals during batch effect removal.
• The developed framework and code are publicly available as an R package for broader scientific use.