The complexity of linkage analysis with neural networks

Area of Science:

• Genetics
• Computational Biology
• Bioinformatics

Background:

• Genome-wide scans are shifting towards complex traits.
• New linkage detection methods are needed beyond single-locus effects.

Purpose of the Study:

• To investigate the repeatability of a neural network method for detecting disease loci in complex traits.
• To assess the reliability of neural network results in genetic linkage analysis.

Main Methods:

• Utilized a neural network trained on genome-wide data.
• Analyzed the non-linear relationship between identity-by-descent sharing and disease status in affected siblings.
• Performed multiple runs of the neural network to assess result consistency.

Main Results:

• Neural network results showed considerable variation across multiple runs.
• This variability suggests potential issues with the optimization process.

Conclusions:

• The neural network method for complex trait linkage analysis may lack repeatability.
• The presence of multiple local minima during neural network training likely contributes to result variability, impacting its reliability for genetic studies.