Improving DNA Methylation Clustering Methods
Author Information
Author(s): Marjoram Paul, Chang Jing, Laird Peter W, Siegmund Kimberly D
Primary Institution: University of Southern California
Hypothesis
The fewer parameters required by the threshold detection model may translate into a lower classification error than the Bernoulli-lognormal model.
Conclusion
Using a detection threshold model improves clustering performance for DNA methylation data compared to traditional methods like k-means.
Supporting Evidence
- The study shows that the single threshold model outperforms the Bernoulli-lognormal model in clustering accuracy.
- Results indicate that the number of genes measured significantly affects clustering performance.
- Clustering methods that respect the biological properties of the data yield better results.
Takeaway
This study shows that when measuring DNA methylation, using a special model that accounts for undetectable values can help group cancer samples better.
Methodology
The study compares a Bernoulli-lognormal model and a single threshold model for clustering DNA methylation data.
Limitations
The performance of the models varies significantly between different cancer types and the number of genes measured.
Participant Demographics
The study involved DNA methylation data from lung cancer cell lines and colorectal cancer tissues.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website