Procrustes is a machine-learning approach that removes cross-platform batch effects from clinical RNA sequencing data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Personalized oncology relies on gene expression profiling.
• Standardizing RNA sequencing (RNA-seq) data from different protocols (exome capture [EC] and poly-A) is crucial for accurate comparisons.
• Formalin-fixed, paraffin-embedded samples present unique challenges for RNA-seq.

Purpose of the Study:

• To optimize an EC-based RNA-seq protocol for formalin-fixed, paraffin-embedded samples.
• To develop a machine-learning algorithm (Procrustes) to correct batch effects between EC and poly-A RNA-seq data.
• To enhance the comparability of gene expression measurements across different RNA-seq protocols.

Main Methods:

• Developed and optimized an EC-based RNA-seq protocol for FFPE samples.
• Created a machine-learning algorithm, Procrustes, for batch effect correction.
• Applied Procrustes to RNA-seq data from EC and poly-A protocols and benchmarked against other methods.

Main Results:

• Procrustes significantly improved gene expression correlation between EC and poly-A protocols from 26% to 61% (N=20,062).
• Correlation for cancer-specific and microenvironment genes increased from 36% to 84% (N=1,438).
• Procrustes outperformed other batch correction methods and can project single-sample data onto larger cohorts.

Conclusions:

• The optimized EC protocol and Procrustes algorithm enhance RNA-seq data comparability.
• Procrustes effectively mitigates batch effects, improving gene expression analysis accuracy.
• This approach supports gene expression-based treatment decisions in personalized oncology by enabling robust analysis of single tumor samples.