Reproducibility of gene expression signature-based predictions in replicate experiments

Area of Science:

• Genomics
• Biotechnology
• Computational Biology

Background:

• Gene expression profiling is crucial for understanding cellular responses and developing targeted therapies.
• Reproducibility and robustness of gene expression data are essential for reliable diagnostic and prognostic predictions.
• Pharmacogenomic predictors aim to personalize treatment by predicting therapeutic response based on an individual's genetic makeup.

Purpose of the Study:

• To assess the concordance of gene expression results from replicate experiments using Affymetrix U133A chips.
• To evaluate the prediction agreement of multigene pharmacogenomic predictors on replicate gene expression data.
• To determine the robustness of these predictions when faced with experimental noise.

Main Methods:

• Gene expression profiling was performed on 97 breast cancer fine-needle aspiration biopsies using Affymetrix U133A chips.
• Replicate experiments included intra-laboratory, inter-laboratory, and manual versus robotic labeling comparisons.
• 111 pharmacogenomic predictors were developed and tested on replicate data, with introduced noise to assess robustness.

Main Results:

• High concordance correlation coefficients (0.96-0.99) were observed for all replicate types, indicating excellent reproducibility.
• Pharmacogenomic predictors achieved high prediction agreement on replicate data, with support vector machines showing 97% agreement.
• Prediction accuracy in the test set varied by method, with support vector machines achieving 77% accuracy.

Conclusions:

• Gene expression profiling with Affymetrix U133A chips is highly reproducible within and between high-volume laboratories.
• Pharmacogenomic predictions demonstrate strong agreement (>90%) on replicate data, supporting their reliability.
• The findings confirm the robustness of gene expression-based predictors for therapeutic response assessment.