1. Bioinformatics: A Statistician’s Persepctive
Ho Kim
School of Public Health
Seoul National University

2. Contents Case-crossover design Microarray Data Analysis
SNP and haplotype analysis

3. Case-crossover design
Case-control + crossover
Popular tool for estimating the effects of acute outcomes by environmental and occupational exposures
Only cases are sampled, estimates are based on within-subject comparisons of exposures at failure times vs. control times

4. Id=A
Exposure
time
Control
Case E1 E2

5. Controls for time-invariant confounders by design
Conditional logistic regression & matching can be applied
Problems:
confounding by time-varying factors -> selection bias
Time trends in exposure of interest -> bias

6. References, case-crossover design
Maclure(1992) The case-crossover design : a method for studying transient effects on the risk of acute events, Am J Epi 133:
Mittleman, Maclure, Bobinson(1995), Control sampling strategies for case-crossover studies :an assessment of relative efficiency, Am J Epi 142:91-98
Lee, Kim, Schwartz(2000) Bidirectional Case-crossover studies of air pollution : Bias from skewed and incomplete waves, Environ Health Persp 108:
Hong, Rha, Lee, Kwon, Ha, Kim, Ischemic Stroke Onset associated with Decrease in Temperature, Epidemiology (in press)

7. Microarray Data Analysis
대량의 유전자에 대해 발현현상을 동시관찰
유전자의 regulation과 interaction의 이해에 기여
Full Yeast Genome in a Chip
(Brown Lab, Stanford. Univ.)

8. Microarray Experiment
cDNA Microarray Experiment using Apo AI

9. Microarray & Statistical Method
생물학적 의문점
-유전자 발현양상의 차이
-유전자 및 샘플의 분류
Experimental Design
Microarray Experiment
Image Processing
Normalization
Statistical problems in all the procedures
Validation
Estimation
Testing
Clustering
Classification
생물학적 해석 및 확인

10. Experimental Design
Reference Design Loop Design

11. Image Analysis

12. A Statistician’s tip Use robust (nonparametric) estimator data mean
median
1,2,3,4
2.5
1,2,3,4,1000
220 3

13. Steps in Image analysis
Addressing: locate centers
Segmentation: classification of pixels either as signal or background.
Information extraction: for each spot of the array, calculates signal intensity pairs, background and quality measures.

14. Normalization
Microarray자료에서 발현수준에 영향을 미치는 기술적 변이(systematic variation)를 찾아내서 제거함

15. M&A Plot
M = log2(R / G) A = log2(R*G) / 2

16. Global normalization: add constant to make mean zero
Intensity dependent normalization: add values that depend on the intensity

17. Print-tip normalization

18. Effect of Normaization
Print-tip normalization (just location)
Location+scale normalization

19. Cluster Analysis K-means Hierarchical Clustering
SOM (Self-Organizing Map)
Gene Shaving
SVM (Support Vector Machine)

20. K-Means Clustering
Select k initial seed
K is pre-deoermined

21. K-Means Clustering For each data point, assign cluster
Calculate cluster mean
Seed is changed to the mean of the cluster

22. K-Means Clustering
Repeats until seeds don’t change

24. Hierarchical Clustering
An evolutionary tree
Hypohippus osborni
Nannipus minor
Neohipparion occidentale
Calippus placidus
Pliohippus mexicanus
Merychippus seversus
Merychippus secundus
Parahippus pristinus
Mesohippus barbouri

25. We need a measure of similarity or distance(dissimilarity)
Agglomerative a
a,b b
a,b,c,d,e c
c,d,e e
d,e d
Divisive
We need a measure of similarity or distance(dissimilarity)

26. 3 5 2 dAB=(d13+d14+d15+d23+d24+d25)/6 Group average distance Cluster A
Cluster B 1 2
dAB=(d13+d14+d15+d23+d24+d25)/6
Group average distance

27. Dendogram indicating two groups

28. Dendogram Hierarchical Clustering(Eisen et al. 1998)
Clustered display of data from time course of serum stimulation of primary human fibroblasts

29. SNP and Haplotype Analysis
SNP (Single Nucleotide Polymorphism)
Haplotypes
Linkage & Linkage disequilibrium
Association study design
SNP vs. Haplotype for association study
Haplotype estimation
Data analysis

30. SNPs in the Human Genome
All humans share 99.9% the same genetic sequence
SNPs occur about every 1000 base pairs
90% of human genome variation comes SNPs
The human genome contains more than 2 million SNPs
~21,000 SNPs are found in genes
SNPs are not evenly spaced along the sequence
SNP-rich regions
SNP-poor regions

31. SNP Locations SNPs within a gene
May alter protein structures
SNPs in the regulatory region outside a gene
May affect when and how the gene is turned on, which affects the quantity of the protein produced
SNPs not within the vicinity of a gene
genetic markers for locating disease-causing genes

32. SNPs as DNA Landmarks Help in DNA sequencing
Help in the discovery of genes responsible for many major diseases:
asthma, diabetes, heart disease, schizophrenia and cancer among others

33. SNP & Haplotype SNP: Single Nucleotide Polymorphism
Haplotype: A set of closely linked genetic markers present on one chromosome which tend to be inherited together (not easily separable by recombination). G A C
Set of SNP polymorphisms: a SNP haplotype

34. From SNP to Haplotype Phenotype Black eye GATATTCGTACGGA-T Brown eye
GATGTTCGTACTGAAT
GATATTCGTACGGAAT
SNP
Phenotype
Black eye
Brown eye
Blue eye
AG- 2/6
GTA 3/6
AGA 1/6
Haplotypes
SNP Simple to measure & understand
Haplotype have the advantage in the appropriate circumstances of carrying more information about the genotype-phenotype link than do the underlying SNPs.
DNA Sequence

35. Linkage and Linkage Disequilibrium (1)
Linkage: the tendency of genes or other DNA sequences at specific loci to be inherited together as a consequence of their physical proximity on a single chromosome.
Linkage disequilibrium (allelic association): particular alleles at two or more neighboring loci show allelic association if they occur together with frequencies significantly different from those predicted from the individual allele frequencies.
Linkage is a relation between loci, but association is a relation between alleles.

36. Linkage and Linkage Disequilibrium (2)
( = recombination fraction)
No linkage:  = 0.5
Perfect linkage:  = 0
Linkage disequlibrium: 0   1
( = probability of allelic association)
Linkage equilibrium:  = 0
Complete linkage disequilibrium:  = 1

37. Allelic Association (LD) Morton et al. (2001)
Locus B
Locus A Allele 1 Allele 2 Allele frequency
Allele 1
Allele 2
Allele
frequency 1
A, B: diallelic loci; 11, 12, 21, 22: haplotypes; : association probability

38. Measures of LD Covariance D = | 11 22 - 12 21 | Association
 = D/Q(1-R)
All other measures are functions of Q, R, .

39. New Findings on Linkage Disequilibrium
In the chromosome, there are blocks of limited haplotype diversity in which more than 80% of a global human sample can typically be characterized by only three common haplotypes (Patil et al., Science 2001).
Haplotype blocks are the more precise units to reflect genetic variation.
Identification of haplotype structure, i.e., construction of a haplotype map, provides a basis for accurate and efficient association studies.

40. Daly et al. (2001). LD by distance from two markers

41. Association Analysis So, what information do we need?
For the group under study:
Relationships between individuals (family structure)
For each individual in the study:
Disease status (or health status)
Environmental information
Haplotypes for the chromosomes where the candidate gene lies

42. The Problem
It’s not yet easy to measure an individual’s (only two) haplotypes
Molecular haplotyping (nucleotide sequencing) is the gold standard
A more efficient strategy:
Focus on regions, such as certain genes
Estimate haplotypes from SNP data (genotypes)
Use LD map, and reduce the number of loci to represent the haplotype
Use haplotype map (DB) = key SNPs + haplotype blocks with strong LD

43. Haplotyping: Phase ProblemC
SNP1
SNP2
Diploid
Observed: SNP1 G/T SNP2 A/C
Possible Haplotypes: GA, TC or GC, TA
n SNPs  2n possible haplotypes

44. Molecular Haplotyping
Hetero-duplex analysis, mismatch detection, allele-specific PCR:
Have potential to get high-throughput
Only practical for short haplotypes (2-5 kb vs kb)
Costly
Rolling Circle amplification method, etc:
Can handle larger size
Difficult to automate

45. In-silico Haplotyping
Alias: Haplotype Reconstruction, Haplotype Inference,
Computational Haplotyping, Statistical Haplotyping, etc.
Advantages:
Cost effective
High-throughput
Difficulty:
Phase Ambiguity: Haplotypes increase exponentially with SNPs

46. In-silico Haplotyping: Two Tasks
Reconstruction of the haplotypes of the sampled individuals
II. Estimation of haplotypes frequencies in a population

47. In-silico Haplotyping: Approaches
Clark’s algorithm
E-M algorithm (expectation-maximization algorithm)
Bayesian algorithm

48. Haplotype Inference
A: SNP data: 0 (MM), 1 (Mm), 2 (mm) for a single locus
B: Haplotype data: 0(M), 1 (m) for a single locus

49. #1 1, 2
00000
00100
#2 1, 3
00010
#3 1, 4
01001
#4 1, 5
00001
#5 1, 1
#6 1, 1

50. An Example Data 169 cases, 231 controls 11 haplotypes
sex, age information

52. Logistic Regression Results
Without adjusting for age, sex:
Haplotype 7 is most strongly associated, but not statistically significant (p=0.07)
Adjusting for age, sex:
Haplotype 11 is most strongly associated (p=0.03)
Slightly stronger association with accounting for repeated measures (2 haplotypes per person) by GEE procedure (p=0.02)

53. Other Examples

54. Drysdale et al. PNAS 2000, 97(19) 10483–10488

57. Wallenstein, Hodge, and Weston, Genetic Epidemiology 15:173–181 (1998)

58. Cohort study
Case-control study

59. Shaw et al. Am J of Medical Genet 114 205-213 (2002)

60. IMT2000 소개

61. Ultimate Human Genome ….. …… Genetic epi Genomic epi , medicine
Functional or structural units (coding regions, gene,…)
Biology
Chromosome
…..
Epidemiology
1 2 ….. SNP … SNP i
Sub 1 ……
Various phenotype
Sequence alignment
Id, naming
Exposure
Ethnic group

62. Thank you !
This file is available at
/~hokim
열린 강의실, 세미나자료