svaseq: removing batch effects and other unwanted noise from sequencing data

Area of Science:

• Genomics
• Bioinformatics
• Statistical Genetics

Background:

• Unwanted noise and batch effects in genomic data reduce statistical inference accuracy.
• Accurate measurement of biological variability requires modeling and removal of these artifacts.
• High-throughput genomic analysis necessitates robust methods for noise reduction.

Purpose of the Study:

• To describe updated versions of surrogate variable analysis (sva) for genomic data.
• To introduce methods specifically for count data and FPKM values from sequencing experiments.
• To present supervised sva (ssva) for artifact identification using control probes.

Main Methods:

• Surrogate variable analysis (sva) identifies artifacts by subsetting data and estimating noise with principal components or singular vectors.
• A new sva version is adapted for count data and FPKMs using appropriate data transformations.
• Supervised sva (ssva) incorporates control probes to refine artifact identification.

Main Results:

• The study compares updated sva versions and supervised sva (ssva) against other batch effect estimation methods.
• Performance is evaluated on simulated, count-based, and FPKM-based genomic datasets.
• The updated sva methods demonstrate improved artifact estimation and correction capabilities.

Conclusions:

• The described sva updates provide effective tools for removing technical artifacts in genomic data.
• These methods enhance the accuracy of statistical inference in sequencing experiments.
• The updated sva approaches are available via the sva Bioconductor package with reproducible analysis scripts.