Feature specific quantile normalization enables cross-platform classification of molecular subtypes using gene...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Transcriptomic profiling reveals cancer and autoimmune disease subtypes, aiding in understanding pathogenesis and treatment.
• Technical biases across gene expression profiling platforms complicate multi-study data analysis.
• Existing methods lack effectiveness in eliminating platform-based bias.

Purpose of the Study:

• To develop and validate a method for normalizing and classifying RNA-sequencing (RNA-seq) data.
• To address platform-based bias in gene expression profiling.
• To enable comparison of RNA-seq data with existing DNA microarray datasets.

Main Methods:

• Utilized machine learning classifiers trained on DNA microarray data.
• Applied feature specific quantile normalization (FSQN) to RNA-seq data.
• Tested the method on breast invasive carcinoma (BRCA) and colorectal cancer (CRC) datasets.

Main Results:

• FSQN successfully removed platform-based bias from RNA-seq data.
• Achieved high accuracy (up to 98% for BRCA, 97% for CRC) in assigning molecular subtypes.
• Maximum accuracy was observed with RNA-seq datasets containing at least 25 samples.

Conclusions:

• FSQN enables the comparison of RNA-seq data with historical DNA microarray datasets.
• This method allows for the effective utilization of existing gene expression data across different platforms.
• The developed technique overcomes platform-specific limitations in transcriptomic analysis.