Combining location-and-scale batch effect adjustment with data cleaning by latent factor adjustment.

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• High-throughput molecular data often contains distinct groups (batches) due to varying experimental conditions.
• Systematic differences between batches, known as batch effects, can distort analysis results if not addressed.
• Existing batch effect correction methods may be too simplistic or rely on restrictive assumptions.

Purpose of the Study:

• To introduce FAbatch, a general, model-based method for correcting batch effects in analyses with a binary target variable.
• To evaluate FAbatch's performance against common competitors using various metrics.
• To demonstrate FAbatch's utility in prediction modeling for eliminating batch effects from new data.

Main Methods:

• FAbatch integrates location-and-scale adjustment with data cleaning via latent factor adjustment.
• The method is extensively compared to existing batch effect correction techniques.
• Implementation is available in the R package bapred.

Main Results:

• FAbatch demonstrates competitive to above-average performance across various metrics.
• The method adequately preserves biological signal, except in cases of extremely outlying batches or very weak batch effects.
• FAbatch shows successful application in prediction modeling with real and simulated data.

Conclusions:

• Real-world batch effect structures are diverse, necessitating flexible correction methods.
• FAbatch's general underlying model and strong performance make it a reliable tool for most practical batch effect adjustment scenarios.
• The method is suitable for diverse datasets and prediction modeling tasks.