Cross-platform normalization enables machine learning model training on microarray and RNA-seq data simultaneously.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression data compendia are vital for discovering biological relationships.
• Microarray and RNA-sequencing (RNA-seq) are common gene expression assay platforms.
• Differences in data structure and distribution hinder direct integration of microarray and RNA-seq data.

Purpose of the Study:

• To evaluate existing normalization methods for combining microarray and RNA-seq data.
• To identify the best normalization techniques for cross-platform machine learning applications.

Main Methods:

• Supervised and unsupervised machine learning evaluations were performed.
• Comparison of various normalization techniques including quantile, Training Distribution Matching, nonparanormal normalization, and z-scores.
• Assessment of suitability for pathway analysis using Pathway-Level Information Extractor (PLIER).

Main Results:

• Quantile and Training Distribution Matching normalization facilitate simultaneous supervised and unsupervised model training on both microarray and RNA-seq data.
• Nonparanormal normalization and z-scores are suitable for specific applications like PLIER pathway analysis.
• Effective cross-platform normalization is achievable for machine learning.

Conclusions:

• Existing normalization methods can successfully integrate microarray and RNA-seq data.
• This integration enables robust machine learning applications across different gene expression platforms.
• The findings support the combined use of historical and current gene expression data for broader biological insights.