Impact of Pre-Processing Methods on Microarray Data Analysis
Author Information
Author(s): Verhaak Roel GW, Staal Frank JT, Valk Peter JM, Lowenberg Bob, Reinders Marcel JT, de Ridder Dick
Primary Institution: Erasmus Medical Center
Hypothesis
How do different pre-processing methods affect the analysis outcomes of microarray data in patient-cohort studies?
Conclusion
The choice of pre-processing method has a relatively minor influence on the final analysis outcome of large microarray studies, but can significantly affect smaller studies.
Supporting Evidence
- The study used two publicly available datasets to assess the impact of pre-processing methods.
- Results showed that the choice of pre-processing method had a more profound effect on the CNS dataset than on the AML dataset.
- Statistical tests indicated that no single pre-processing method consistently outperformed the others across all analyses.
Takeaway
This study looked at how different ways of preparing data from gene tests can change the results. For big studies, the method doesn't matter much, but for smaller ones, it can make a big difference.
Methodology
The study compared four pre-processing methods (MAS, RMA, GCRMA, dChip) using two datasets: one with 285 AML patients and another with 42 CNS tumor samples.
Potential Biases
The use of special-purpose datasets for calibration may create a risk of overfitting.
Limitations
The study did not use data with accompanying ground truth to determine which method provides the closest estimates to actual expression values.
Participant Demographics
The study involved 285 patients with Acute Myeloid Leukemia and 42 samples of tumor tissue from the embryonal central nervous system.
Statistical Information
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website