Extensive increase of microarray signals in cancers calls for novel normalization assumptions

Area of Science:

• Bioinformatics
• Genomics
• Cancer Research

Background:

• Microarray data normalization commonly assumes stable gene expression distributions between sample groups.
• This assumption may be violated in complex diseases like cancer, where widespread gene expression changes occur.
• Evaluating the impact of normalization on biological discoveries is crucial for accurate cancer research.

Purpose of the Study:

• To assess the validity of standard microarray normalization assumptions in cancer datasets.
• To evaluate the performance of widely used normalization algorithms (RMA, MAS5.0, dChip) in identifying differentially expressed genes (DE genes) in cancer.
• To compare these algorithms with a less assumption-reliant method (LVS).

Main Methods:

• Analysis of 7 large Affymetrix datasets of paired normal and cancer samples from the NCBI GEO database.
• Investigation of probe intensity distributions across normal and cancer samples.
• Comparison of DE gene selection results from RMA, MAS5.0, dChip, and LVS algorithms.

Main Results:

• Median probe intensities significantly increased in cancer samples across 6 of 7 datasets.
• RMA, MAS5.0, and dChip algorithms produced a high number of false down-regulated DE genes and missed many up-regulated DE genes.
• LVS algorithm showed better performance due to fewer reliance on normalization assumptions.

Conclusions:

• Standard normalization methods assuming uniform probe intensity distributions may be misleading for cancer studies.
• Current normalization techniques can distort biological differences between normal and cancer samples.
• Widespread gene up-regulation may characterize most human cancers, challenging traditional normalization assumptions.