Stability of Variable Importance Scores in Genetic Studies
Author Information
Author(s): Kristin K Nicodemus, Wenyi Wang, Yin Yao Shugart
Hypothesis
How stable are variable importance scores and rankings when using statistical learning tools on genetic data?
Conclusion
The study found that random forests and generalized boosted regression provided more stable estimates of variable importance and rankings compared to Monte Carlo logic regression.
Supporting Evidence
- The study tested three statistical learning tools to assess variability in rankings and importance scores.
- Random forests and generalized boosted regression showed more stable estimates of variable importance.
- The results suggest that repeated analyses are necessary to ensure reliability in variable importance measures.
Takeaway
This study looked at how reliable certain methods are for figuring out which genes are important in diseases. It found that some methods are better at giving consistent results than others.
Methodology
The study used three statistical learning methods (Monte Carlo logic regression, random forests, and generalized boosted regression) to analyze simulated genetic data for interactions in complex diseases.
Limitations
Results are specific to the simulation model used and may differ under alternative models.
Want to read the original?
Access the complete publication on the publisher's website