1. Population Structure Analysis
using STRUCTURE software
Chang Bum Hong
kt Bioinformatics TF, hongiiv@gmail.com, twitter @hongiiv, hongiiv.tistory.com
Permissions: you are free to blog or live-blog about this presentation as long as you attribute the work to its authors
Friday, August 12, 11
2. Genetic test
일반적으로 알콜을 섭취하게 되면 알콜은 아세트알데히드(얼굴을 붉게 만들고, 가슴도 콩닥
거리고, 구토를 일으키는 독성 물질)로 변하게 되고 이것이 다시 ALDH 에 의해 인체에 무해
한 젖산으로 분해되는 과정을 거치게 됩니다. 이때 ALDH2라는 유전자가 바로 아세트알데히
드가 조금이라도 생성되면 분해하는데 관여하게 이때 유전자형에 따라서 3가지 유형으로 나
타나게 됩니다.
Friday, August 12, 11
7. East Asia - Public genotype data
SNP Individual Population
PASNP 54,794 1,928 75
HGDP a 2,834~ 1,056 52
HapMap 1,481,135 1,397 11
b
SGVP 268,667 292 3
Korean 58,625 159 10
China(Yanbian) 58,625 16 1
Japan(Kobe) 58,625 5 1
Korea-Japan 58,625 6 1
Vietnam 58,625 16 1
Korean-Vietnam 58,625 8 1
Cambodia 58,625 16 1
Mongol 58,625 16 1
a. Pan-Asian SNP Consortium(http://www4a.biotec.or.th/PASNP)
b. Singapore Genome Variation Project(http://www.nus-cme.org.sg/SGVP)
Friday, August 12, 11
8. Korean Data
16
YeonCheon
16
Pyeong
Chang
MW
JeCheon
16 16
Cheonan
average >70 year old
long settlement
Affymetrix 50K Xba
GyeongJu
16 16
GimJe 15 China(Yanbian)
Goryeong UlSan
Japan(Kobe)
16 Korea-Japan
Vietnam
Korean-Vietnam
SW 16
NaJu
SE
Cambodia
Mongol
16 58,960 SNPs
Jeju
Friday, August 12, 11
9. Missing genotype individuals
GimJe
GoRyeong
Gyeong
Text Ju
Before QC 58,960 SNPs Before QC 58,960 SNPs
All Asian Korean
Friday, August 12, 11
10. Relatedness between the 153
Korean(10 region) Individuals
YeonCheon
PyeongChang
JeCheon
CheonAn GyeongJu
UlSan
GimJe GoRyeong
NaJu
JeJu
PCA analysis using autosomal 46,559 SNP markers (n=153, Korean)
Friday, August 12, 11
11. PCA analysis of East Asian descent
Mongol
Yanbian
Kobe JPT-
Jeju HapMap
CHB-
HapMap
Vietnam
Cambodia
illustration of geographic correspondence of ethnic group
Korea-Vietnam Korea-Japan
locations
Friday, August 12, 11
12. Relationship between Eigenvector
values and Latitude
47.81
39.98
37.53
2
R = 0.8621
y = 36.65 + 166.33x
14.72
Friday, August 12, 11
13. STRUCTURE software
• A model-based clustering method (Pritchard et al. 2000)
• Free software
(http://pritch.bsd.uchicago.edu/structure.html)
• Bayesian approach (MCMC: Markov Chain Mote Carlo)
• Detects the underlying genetic population among a set of individuals genotyped at multiple
markers
• Computes the proportion of the genome of an individual originating from each inferred
population (quantitative clustering method)
Friday, August 12, 11
14. Input data
• A matrix where the data for individuals are in rows, the loci
are in column
• n consecutive rows have the data for each individual of n-
ploid species
• Integer should be used for coding genotype
• Missingoccur should be indicated by(e.g. -1) which
doesn’t
data
elsewhere in the data
a number
• The dataSTRUCTUREbe a text file (.txt) not an excel (.xls) for
running
file should
Friday, August 12, 11
15. Input format
1 consecutive rows for alleles
MarkerName...
Label PopID Flag Location Genotype...
genotype (1,2,5)
AA = 11
AB = 12
BB = 22
missing = 55
Information of user-defined populations
Lable : 각 개인의 고유한 ID로 숫자 또는 문자 어떤것이든 상관없다.(예, CEPH1334.10)
PopID: 개인이 속한 민족의 고유한 번호 (예, 중국인(CHB)인 경우 5, 유럽인(CEU)인 경우 1과 같이 자신이 직접 부여)
Flag: 해당 PopID 정보를 STRUCTURE 프로그램 실행시 사용할 것인가?(1= 사용한다, 2= 사용하지 않는다.)
Location: 해당 개인의 위치정보(예, 동아시아(EAS)인경우 1번, 유럽(EURA)인 경우 2번과 같이 자신이 직접 부여)
Friday, August 12, 11
27. Configuring a parameter set (cont.)
Length of Burnin Period : how long to run the simulation before collecting data to minimize the
effect of the starting configuration, 목표함수로 수렴할 때까지의 반복 숫자
Number of MCMC Reps after Burnin : how long to run the simulation after burnin to get
accurate parameter estimates
Friday, August 12, 11
37. Analysis of genome-wide SNP data
• For very may become impractically slow
settings
large data sets, the runtime of structure using default
• reduced data sets (ex, pruned)
• get accurate resultsNUMREPS) shorter runs than default
(ex, small values of
using much
• download themachine) and compile it on your machine
(using 64-bit
source code
• use the command-line version of structure
Friday, August 12, 11
38. An example of MCMC convergence
Friday, August 12, 11
39. Inference of true K
(number of population)
• The log likelihood for each K, Ln P(D) = L(K)
• Two approaches to determine the best K
• Use of L(K) : When K is approaching a true value, L(K) plateaus
and has high variance between runs
• Use of an ad hod quantity (∆K) the likelihood (∆K).on the
second order rate of change of
: calculated based
The ∆K
shows a clear peak at the true value of K
Friday, August 12, 11
44. We may not always be able to know the TRUE value
K, but we should aim for the smallest value of K
that captures the major structure in the data
Pritchard et al. (2000)
Friday, August 12, 11