Evaluating Cross-Platform Batch Correction Methods for Integrated Microarray and RNA-seq Data Analysis

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Integrating gene expression data from different platforms (microarray, RNA-seq) is crucial for complex trait analysis.
• Platform-specific technical differences pose significant challenges for cross-platform data integration.

Purpose of the Study:

• To evaluate ten batch-effect correction methods for combining microarray and RNA-seq data.
• To compare unsupervised (sample-wise, gene-wise) and supervised methods for cross-platform integration.

Main Methods:

• Classified methods into unsupervised sample-wise, unsupervised gene-wise, and supervised approaches.
• Assessed performance using distribution alignment, clustering, outcome prediction, and differential expression (DE) analysis.
• Utilized paired real datasets and simulation studies for evaluation.

Main Results:

• Supervised methods showed good alignment but introduced information leakage and inflated Type I error.
• Gene-wise methods generally maintained Type I error control and offered higher power, with limma excelling in DE analysis.
• Meta-analysis also controlled Type I error well; limma and QN performed best for outcome prediction.

Conclusions:

• limma is recommended as the top general-purpose method for cross-platform integration due to its strong performance and error control.
• Gene-wise methods are superior to supervised methods for unbiased discovery in integrated gene expression studies.
• QN offers practical advantages for normalizing new samples without refitting.