On data normalization and batch-effect correction for tumor subtyping with microRNA data.

Area of Science:

• Bioinformatics
• Genomics
• Cancer Research

Background:

• Transcriptomics profiling aids tumor subtype discovery.
• Data artifacts from experimental handling can cause irreproducible subtypes.
• Normalization and batch-effect correction methods are used for subtyping but their effectiveness for sample clustering is unclear.

Purpose of the Study:

• To evaluate the effectiveness of data normalization and batch-effect correction for sample clustering in tumor subtyping.
• To investigate the impact of these methods on the reproducibility of microRNA-based tumor subtype discovery.

Main Methods:

• A re-sampling-based simulation study using microRNA microarray data.
• Examination of normalization techniques (e.g., quantile normalization).
• Assessment of batch-effect correction methods and various clustering algorithms (e.g., Prediction Around Medoid).

Main Results:

• Normalization generally benefited sample clustering and subtype discovery, with quantile normalization performing best.
• Batch-effect correction was found to be harmful when artifacts confounded biological signals.
• Clustering method choice significantly influenced performance, with Prediction Around Medoid being a consistent performer.

Conclusions:

• Data normalization is beneficial for reproducible tumor subtyping using microRNAs.
• Batch-effect correction should be applied cautiously, as it can hinder discovery.
• Optimal results depend on the interplay between normalization, batch-effect correction, array-to-sample assignment, and clustering method selection.