Practical impacts of genomic data "cleaning" on biological discovery using surrogate variable analysis.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Genomic data production is rapidly increasing, providing vast resources for research.
• Systematic heterogeneity, technical or biological, can confound genomic analyses.
• Batch correction methods aim to adjust for such widespread systematic noise.

Purpose of the Study:

• To explore the impact of "batch" correction on biological discovery using surrogate variable analysis (SVA).
• To assess the consequences of data cleaning for identifying biological effects in public expression datasets.
• To demonstrate artifact discovery and preservation of biological signals.

Main Methods:

• Utilized surrogate variable analysis (SVA) for data analysis.
• Applied SVA to two public expression datasets: differentiating pluripotent cells (GSE32923) and human brain tissue (GSE30272).
• Investigated artifact discovery considering biological heterogeneity, secondary questions, and non-linear effects.

Main Results:

• SVA sharpened differential expression analysis in stem cell systems.
• Non-linear effects of age across the lifespan were preserved in brain tissue data.
• Data cleaning with SVA removed other potentially relevant biological information, such as sex and cell line-specific effects.

Conclusions:

• Data cleaning is valuable for high-throughput genomic analysis of data with technical artifacts.
• Caution is needed to prevent removal of important biological signals.
• Supervised cleaning limits open data exploration; R code and processed data are provided for transparency and reproducibility.