Using Artificial Neural Networks to Estimate Haplotype Frequencies
Author Information
Author(s): Kevin C Cartier, Daniel Baechle
Primary Institution: Case Western Reserve University, Department of Epidemiology and Biostatistics, Cleveland, OH, USA
Hypothesis
Can an artificial neural network (ANN) be designed to estimate haplotype frequencies from population genotype data?
Conclusion
The study successfully demonstrated that an ANN can estimate haplotype frequencies from population genotype data, correlating well with estimates from the EM algorithm.
Supporting Evidence
- The ANN design produced provisional results that correlated well with estimates from the expectation maximization algorithm.
- The correlation between the ANN and EM algorithm was 0.98 for the most frequent haplotypes.
- The ANN approach is promising due to its compatibility with parallel computing architectures.
Takeaway
The researchers created a computer program that helps figure out how common different gene combinations are in a group of people, using a method that mimics how our brains learn.
Methodology
The study used an artificial neural network to classify haplotypes based on genotype data, training the network with random samples from the population.
Potential Biases
The assumption of a uniform distribution of haplotype frequencies given genotype is likely unrealistic.
Limitations
The ANN's performance could not be properly analyzed due to the lack of haplotype information in the simulated data, and the training process was somewhat arbitrary.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website