Genetic Classification of Populations Using Supervised Learning
Author Information
Author(s): Michael Bridges, Elizabeth A. Heron, Colm O'Dushlaine, Ricardo Segurado, Derek W. Morris, Aiden Corvin, Michael Gill, Carlos Pinto
Primary Institution: Trinity College, Dublin, Ireland
Hypothesis
Can supervised learning methods effectively classify genetic populations compared to traditional unsupervised methods?
Conclusion
Supervised learning methods, particularly artificial neural networks and support vector machines, significantly outperform traditional principal components analysis in classifying genetic populations.
Supporting Evidence
- Supervised methods showed higher sensitivity than traditional PCA.
- ANNs can classify populations with fewer SNPs than PCA requires.
- SVMs performed comparably to ANNs in classifying genetic data.
- Both supervised methods can detect genetic differences that PCA cannot.
- Supervised learning methods can localize genetic differences along chromosomes.
Takeaway
This study shows that using smart computer programs can help scientists tell different groups of people apart based on their genes, much better than older methods.
Methodology
The study used artificial neural networks and support vector machines to classify genetic data from three populations, comparing their performance to principal components analysis.
Limitations
The study primarily focused on three populations and may not generalize to all genetic datasets.
Participant Demographics
The study involved controls from Scotland and Bulgaria, with a total of 3322 cases and 3587 controls.
Statistical Information
P-Value
0.022
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website