Bayesian pedigree inference with small numbers of single nucleotide polymorphisms via a factor-graph representation

Area of Science:

• Computational Biology
• Population Genetics
• Bioinformatics

Background:

• Pedigree inference is crucial for understanding genetic relationships and population structures.
• Existing methods often rely on strict assumptions of complete sampling and error-free genotyping.
• Limited numbers of single nucleotide polymorphisms (SNPs) pose challenges for accurate pedigree reconstruction.

Purpose of the Study:

• To develop a flexible computational framework for pedigree inference using a limited number of SNPs.
• To relax common assumptions of complete sampling and genotyping errors in pedigree analysis.
• To enable efficient Bayesian inference of pedigree structures using Markov Chain Monte Carlo (MCMC) methods.

Main Methods:

• Developed a factor graph representation for inferred pedigrees.
• Utilized the Sum-Product algorithm for efficient computation of data probabilities under pedigree rearrangements.
• Employed MCMC sampling for Bayesian inference of pedigree structures.
• Applied the method to infer full sibling groups in Chinook salmon using 95 SNPs.

Main Results:

• The developed computational framework successfully infers pedigrees with fewer SNPs (80-400) and relaxed assumptions.
• The method demonstrates superior point estimates and uncertainty quantification compared to maximum-likelihood approaches for sibling reconstruction.
• Accurate inference of full sibling groups was achieved in a large Chinook salmon sample (n=1157).

Conclusions:

• The novel Bayesian framework offers a robust and accurate method for pedigree inference, particularly in scenarios with limited genetic markers and imperfect data.
• This approach enhances the reliability of genetic relationship reconstruction in population genetics studies.
• The methodology shows promise for extensions to more complex pedigree structures and scenarios.