Haplotypes Personalized Medicine: Understanding Your Own Genome Fall 2014

Transcription

1 Haplotypes Personalized Medicine: Understanding Your Own Genome Fall 2014

2 Terminology Review llele: different forms of genecc variacons at a given gene or genecc locus Locus 1 has two alleles, and C, and Locus 2 has two alleles, T and G Individual 1 T G Locus 2 Genotype: specific allelic make- up of an individual s genome Individual 1 has genotype at Locus 1 and genotype TG at Locus 2 Locus 1 Individual 2 T T Heterozygous/Homozygous Locus 1 of Individual 1 is homozygous, and Locus 2 is heterozygous C Locus 1 Locus 2

3 Single Nucleo<de Polymorphism (SNP) GTCTTCGTCTGGT GTCTTCGTCTGGT GTTTTCGTCGGT C p GTTTTCGTCTGGT C m GTCTTCGTCTGT GTTTTCGTCGGT a diploid individual GTTTTCGTCGGT GTCTTCGTCTGT SNP: Binary nucleotide sustitutions at a single locus on a chromosome" each variant is called an "allele "

4 From SNPs to Haplotypes GTTTTCGTCGGT C p GTTTTCGTCTGGT C m GTCTTCGTCTGT GTTTTCGTCGGT a diploid individual GTTTTCGTCGGT GTCTTCGTCTGT chromosome" Haplotype: a stretch of consecutive nucleotides that lie on the same chromosome" What are the alleles here? " GTCTTCGTCTGGT GTCTTCGTCTGGT

5 Haplotypes from SNP rray? C m C p T G C T TGC sequencing Heterozygous diploid individual TC TG Genotype g pairs of alleles with association of alleles to chromosomes unknown T G C T T T C G haplotype h (h 1, h 2 ) possile associations of alleles to chromosome

6 Why Haplotypes? Haplotypes have a greater power for discriminacng genomic regions Consider J inary markers (e.g., SNPs) in a genomic region There are 2 J possile haplotypes SNPs have only two alleles, whereas haplotypes have a larger numer of alleles Good genecc marker for populacon, evolucon and hereditary diseases

7 Haplotypes and SNPs GTCTTCGTCTGGT GTCTTCGTCTGGT GTTTTCGTCGGT GTTTTCGTCTGGT GTCTTCGTCTGT GTTTTCGTCGGT GTTTTCGTCGGT GTCTTCGTCTGT chromosome" Haplotype CTG 3/8 TG 3/8 CT 2/8 SNPs can discnguish etween two groups of individuals (a group with C, another group with T) Haplotypes can discnguish etween three groups of individuals (each group with CTG, TG, and CT)

8 Haplotypes and SNPs GTCTTCGTCTGGT GTCTTCGTCTGGT GTTTTCGTCGGT GTTTTCGTCTGGT GTCTTCGTCTGT GTTTTCGTCGGT GTTTTCGTCGGT GTCTTCGTCTGT chromosome" Haplotype CTG 3/8 TG 3/8 CT 2/8 healthy healthy disease X Haplotypes can have a greater power to detect disease- related genome region

9 Inferring Haplotypes from SNP rray Data Genotype: C////TG Maternal genotype: C////TT Paternal genotype: CC////TG Then the haplotype is C/TG. Genotype: C////TG Maternal genotype: C////TG Paternal genotype: C////TG Cannot determine unique haplotype Prolem: How can we determine haplotypes without parental genotypes

10 Phasing: Inferring Haplotypes from SNP Data Given mulclocus genotypes at a set of SNPs for many individuals, phasing means Reconstruct haplotypes for all individuals EsCmate frequencies of all possile haplotypes Haplotype reconstruccon algorithm Clark s parsimony algorithm (Clark, Mol. Biol. Evol. 1990)

11 Iden<fiaility Genotypes of 14 individual Genotype representations 0/0! 0 1/1! 1 0/1!

12 Iden<fiaility " 10" 7" Parsimonious solution" " 1" 1" 1" 8" 1" 1" 6" 1"

13 Haplotype Reconstruc<on lgorithm y Clark (1990) Choose individuals that are homozygous at every locus (e.g. TT////CC) Haplotype: TC Choose individuals that are heterozygous at just one locus (e.g. TT//// CG) Haplotypes: TC or TG Tally the resulcng known haplotypes. For each known haplotype, look at all remaining unresolved cases: is there a cominacon to make this haplotype? Known haplotype: TC Unresolved pa`ern: T////CG Inferred haplotype: TC/G. dd to list. Known haplotype: TC and TG Unresolved pa`ern: T////CG Inferred haplotypes: TC and TG. dd oth to list. ConCnue uncl all haplotypes have een recovered or no new haplotypes can e found this way.

14 Prolems: Clark (1990) Many unresolved haplotypes at the end Ignores recominacon Error in haplotype inference if a crossover of two actual haplotypes is idenccal to another true haplotype Frequency of such errors depends on recominacon rate Clark (1990): algorithm "performs well" even with small sample sizes.

15 RECOMBINTION & LINKGE DISEQUILIBRIUM

16 Morgan s FruiYly Experiment Morgan s frui.ly data (1909): 2,839 flies Eye color : red a: purple Wing length B: normal : vescgial BB x aa" ab x aa" ab a aab aa" Exp " Os 1, ,195"

17 Linked Genes When two genes lies on the same chromosome, they are transmi`ed to offspring in a non independent manner B a

18 Morgan s Explana<on: Recomina<on B B a a F1: a B a a F2:" B a a a a a B a Recomination has taken place"

19 Recomina<on Parental types: ab, aa Recominants: a, aab The proporcon of recominants etween the two genes (or characters) is called the recomina*on frac*on etween these two genes.

20 Review: Correla<on GP and TV in hours per week are negacvely correlated Mean How can we quancfy the level of correlacon?

21 Covariance and Correla<on Degree of associacon etween two variales x and y Given oservacons x 1,, x n and y 1,, y n Covariance CorrelaCon coefficient: (Variance of x i s) x (n- 1) (Variance of y i s) x (n- 1) Falls etween - 1 and +1, with sign indicacng direccon of associacon

22 Correla<on etween X 1 and X 2 X1 X 2

23 Basic Concepts B a B a High LD -> No Recomination (r 2 = 1) SNP1 tags SNP2 B B B a a a Low LD -> Recomination Many possiilities a B a B B a B etc B B X OR Parent 1 Parent 2

24 Linkage Disequilirium (LD) LD reflects the relaconship etween alleles at different loci. Omen, r 2 (squared correlacon coefficient) is used as a measure of LD. Locus Locus B

25 How to Compute r 2 on SNP Data Individuals SNP1 SNP2 SNP r 2 =1.0 SNP1 SNP2 SNP3 r 2 matrix SNP1 SNP2 SNP r 2 =0.0 R 2 =0.0

26 Linkage Disequilirium in SNP Data r 2 in SNP data from a populacon of individuals (Black: r 2 =1, white: r 2 =0) genome PopulaCon 2 PopulaCon 2 genome PopulaCon 1 PopulaCon 1

27 Reducing Genotyping Costs with Tag SNPs Neary SNPs in the genome are in linkage disequilirium (LD), and thus contain redundant informacon. If we knew which SNPs are in LD, we can pre- select the representacve SNPs for each LD lock of chromosome, and genotype only for those SNPs. r 2 values (lack: r 2 =1, white: r 2 =0) Genome These two SNPs are in high LD and thus are redundant

28 Reducing Genotyping Costs with Tag SNPs Two- stage data colleccon process Stage 1: Collect genotype data for a dense set of SNPs for mulcple individuals Select a non- redundant set of tag SNPs y examining the LD pa`ern Stage 2: Collect genotype data only for the tagsnps for a large numer of individuals

29 lgorithm for Selec<ng Tag SNPs Greedy algorithm Genome Randomly select a tag SNP Iterate uncl the set of candidate tag SNPs is empty Genome Find the SNPs with a high LD with the previously selected tag SNP (r 2 >0.8) and remove those SNPs from the set of candidate tag SNPs

30 Recomina<on and Haplotypes Rememer Clark s method does not take into account recominacon How can we find haplotypes from SNP data collected for a populacon of individuals under recominacon? ssume haplotypes of ancestor chromosomes and treat modern individuals chromosomes as a mosaic of ancestor chromosomes However, ancestor chromosomes cannot e oserved! Key idea: Haplotype of each individual is a mosaic of other individuals haplotypes unresolved haplotypes are similar to known haplotypes

31 Recomina<on and Haplotypes h 1, h 2, h 3 : unoserved ancestral haplotypes we have no SNP data h 4, h 4B : unoserved haplotypes for modern individuals Haplotypes are unoserved, however, we have SNP data Circles: mutacons TCGTTTTCGTTCGTGTGTTTCTGTTCTGTGTCGTTC TCGTTTTTTCTTTTGCGTGTTTCTGCTGCTTCTGTGTCGTTC Mosaic of ancestor chromosomes

32 PHSE Model as an HMM Inferring the unoserved state laels for each of the oserved SNP amounts to haplotype reconstruccon TCGTTTTCGTTCGTGTGTTTCTGTTCTGTGTCGTTC h3h3h3h3h3h3h3h3h3h3h3h3h2h2h2h2h2h2h2 TCGTTTTTTCTTTTGCGTGTTTCTGCTGCTTCTGTGTCGTTC h3h3h3h3h3h1h1h1h1h2h2h2h2h2h2h2h2h2h3h3h3.

33 PHSE Model as an HMM States: h 1, h 2, h 3, unoserved ancestral haplotypes State space with possile transicons h 1 h 2 h 3 TransiCon proailices (from SNP X l to X l+1 ) are dependent on distance etween adjacent SNPs d l RecominaCon rate etween adjacent SNPs ρ l Emission proailices: mutacon model Task: infer hidden state laels for each locus of each individual (h 4, h 4B )

34 INTERNTIONL HPMP PROJECT (HPMP.ORG)

35 HapMap Phase 3 Samples lael population sample # samples QC+ Draft 1 SW* frican ancestry in Southwest US CEU* Utah residents with Northern and Western European ancestry from the CEPH collection CHB Han Chinese in Beijing, China CHD Chinese in Metropolitan Denver, Colorado GIH Gujarati Indians in Houston, Texas JPT Japanese in Tokyo, Japan LWK Luhya in Weuye, Kenya MEX* Mexican ancestry in Los ngeles, California MKK* Maasai in Kinyawa, Kenya TSI Toscans in Italy YRI* Yorua in Iadan, Nigeria ,301 1,115 * Population is made of family trios

36 Haplotype Structure and Recomina<on Rate Es<mates: HapMap I vs. HapMap II

37 HapMap: llele Frequencies in Different Popula<ons Comparison of allele frequencies for individuals from pairs of populacons The red regions show that there are many SNPs that have similar low frequencies in each pair of analysis panels/ populacons. CHB (Chinese) and JPT (Japanese) have similar allele frequencies

38 Why Haplotypes? Haplotypes have a greater power for discriminacng genomic regions Consider J inary markers (e.g., SNPs) in a genomic region There are 2 J possile haplotypes ut in fact, far fewer are seen in human popula<on SNPs have only two alleles, whereas haplotypes have a larger numer of alleles Good genecc marker for populacon, evolucon and hereditary diseases

39 Summary Haplotype: a set of genecc markers that lie on the same chromosome How can we find haplotypes from SNPs? RecominaCon, linkage disequilirium, and how to take advantage of them Haplotypes as a set of linked SNPs with a greater discriminacve power Tag SNPs for saving the genotyping cost HapMap Project