Stratification Bias in Low Signal Microarray Studies
Author Information
Author(s): Parker Brian J, Günter Simon, Bedo Justin
Primary Institution: NICTA, Australia
Hypothesis
Stratification bias can significantly affect the performance measures in low signal microarray studies.
Conclusion
Stratification bias can substantially affect several performance measures, and using stratified repeated holdout and balanced cross-validation methods can help avoid this bias.
Supporting Evidence
- Stratification bias can lead to large pessimistic biases in performance measures.
- Using averaging strategies for AUC calculation avoids stratification bias.
- Balanced, stratified cross-validation is recommended to mitigate bias.
Takeaway
When scientists study small datasets, like those from microarrays, they need to be careful because mistakes in how they split their data can lead to wrong conclusions.
Methodology
The study analyzed stratification bias through simulations and real-world datasets, comparing various cross-validation methods.
Potential Biases
The study highlights the risk of underestimating model performance due to stratification bias in small datasets.
Limitations
The analysis primarily focused on small sample sizes and may not generalize to larger datasets.
Participant Demographics
The study used the van 't Veer breast cancer dataset, which included 98 samples with varying prognostic outcomes.
Statistical Information
P-Value
<0.001
Statistical Significance
p<0.001
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website