mStruct: Structure under mutations

1
mStruct Structure under mutations
mStruct Inference of population structure in the
presence of genetic admixing and allele mutations

• Suyash Shringarpure and Eric Xing
• Carnegie Mellon University

2
Significance
3
Genetic Population Structure

• Structure (Pritchard et al, 2000)?

Ancestral proportion
Africa
Europe
Mid-East
Cent./S. Asia
East Asia
Oceania
Genetic structure of Human Populations (Rosenberg
et al. 2002)?
4
Generative model- Structure
All the alleles observed at this locus
a (for the dataset)
0.8
0.2
0.8
0.2
0.3
0.7
5
Modeling allele similarity

• Microsatellite
• Repeats of a small DNA unit, say

Allele - 2
Allele - 9
Allele - 10

• Allele 9 is much more similar to allele 10 than
  allele 2.
• Allele 10 might be a mutation of allele 9.
• Mathematically encode the idea in the model
• mStruct Structure under mutations

6
Hypothesis

• Individual genomes in modern populations are a
  result of
• Admixture of ancestral populations.
• Mutations from ancestral alleles.
• Ancestral populations have fewer alleles
• (Mostly) True for microsatellites

7
Generative model- mStruct
All the alleles observed at this locus
a (for the dataset)
d1
0.8
0.2
0.8
0.2
d2
0.3
0.7
8
Mutation models

• How to derive descendant alleles from ancestral
  alleles?
• Distribution based on the single step model
• P(ba) a dabs(b-a) , d lt 1
• Computationally easy
• NOT conventional mutation rate.

9
Finding ancestral alleles

• Fit mixtures of mutation distributions
• Try using 1,2,3.. ancestral alleles
• Use information theory to decide how many
  ancestral alleles are appropriate

Histogram of observed alleles
10
Comparing population structure maps
11
Phylogenetic Trees from the Structural Maps
12
Phylogenetic Trees from the Structural Maps
mStruct
Structure
13
HGDP SNP results
14
Implications of Inconsistency

• Simplistic mutation model
• SNP mutations harder to discover from data
• The model reduces to Structure
• Fundamental difference
• Different markers treated differently
• Structures treatment of alleles is almost
  categorical

15
Contour of Empirical Mutation
16
Conclusion

• Generative model for population structure
• Modeling mutations from ancestral alleles
• Gives mutational information apart from
  population structure.
• (in press) Genetics
• Online version up now.

17
Graphical model representations
Structure
mStruct