1. Bioinformatics Pipeline for Fosmid based Molecular Haplotype Sequencing
Jorge Duitama1,2, Thomas Huebsch1, Gayle McEwen1, Sabrina Schulz1, Eun-Kyung Suk1, Margret R. Hoehe1
1. Max Planck Institute for Molecular Genetics, Berlin, Germany
2. Department of Computer Science and Engineering University of Connecticut, Storrs, CT, USA
Mapping slides

2. MHC: Key Region for Common Diseases & Transplant Medicine
MHC class I
MHC class III
MHC class II
29,74
31,59
32,34
33,21

3. MHC: Variation amongst Haplotypes
Variation of MHC Haplotypes against PGF reference
RCCX
CNV
HLA-DRB
CNV
Haplotype sequences
7 further MHC
PGF reference sequence
MHC class III
MHC class II
Variation amongst 8 MHC Haplotypes:
Substitutions
7.093 Short Indels
Variation and annotation map for eight MHC haplotypes, Horton et al. Immunogenetics (2008) 60,1-18

4. Experimental Approach
100 Individuals
100 Libraries
3x96-well = 288 fosmid pools
40 kb
haploid
molecules
5000 fosmids
One pool
SNP Mapping for Prioritization of MHC Informative Pools
Targeted Enrichment
SOLiD NGS Platform Shotgunning complete kb fosmids
Complete Fosmid Pool
Data Analysis Pipeline
Identification of kb fosmid sequences
Haplotype A
Haplotype B
Phasing molecular fosmid sequences
Contiguous MHC haplotype sequence T G A

5. Data Analysis Pipeline
Fosmid Detection
Program
Read Alignment
against Genome
Fosmid Specific Matching Algorithm
Pairing
Fosmid Sequences
Based Phasing
Consensus
Calling SNP Analysis
Visualization &
MHC Database
In House Project Specific Analysis Pipeline
SOLiD Standard
Pipeline

6. Data Analysis Pipeline
Fosmid Detection
Program
Read Alignment
against Genome
Fosmid Specific Matching Algorithm
Pairing
Fosmid Sequences
Based Phasing
Consensus
Calling SNP Analysis
Visualization &
MHC Database
In House Project Specific Analysis Pipeline
SOLiD Standard
Pipeline

7. Pool of 15.000 Fosmids 22 Mill. Reads 50bp
Mapping real data
Pool of Fosmids 22 Mill. Reads 50bp

8. Data Analysis Pipeline
Fosmid Detection
Program
Read Alignment
against Genome
Fosmid Specific Matching Algorithm
Pairing
Fosmid Sequences
Based Phasing
Consensus
Calling SNP Analysis
Visualization &
MHC Database
In House Project Specific Analysis Pipeline
SOLiD Standard
Pipeline

9. SNP calls: Haploid fosmids vs. genomic DNA
gDNA
# cov
ref
consen F3
coord
335 C Y
177/17
3345 T
3191/56
875 G A
862/25
1795 K
722/23
707 S
528/13
2643
1391/20
643
417/23
1074 R
554/21
606
226/21
639 M
167/15
158
89/14
1032
443/26 7
7/4
775
742/26 10
10/5
698
684/29 40
40/4 94
93/9
286
283/16
194
190/22
Fosmid
# cov
ref
consen F3
coord
595 C T
572/91
3418
3278/98
2089 G A
2048/98
2238
2194/98
1134
1107/73
3104
2922/98
1033
1014/83
1799
1753/98
1053
1049/83 54
39/22 32
27/23
1374
1355/95
973
946/97
2850
2745/98 49
48/33
1888
1845/95 37
36/20
923
901/97
8411 W
2006/78
253
253/47

10. SNP Calling Accuracy in the MHC
Affymetrix genotype information for 1583 SNP positions as reference standard:
- Homozygous identical with reference: 957
- Heterozygous: 562
- Homozygous different from reference: 64
Compared to variants called from the SOLiD sequenced genomic DNA sample (15x average read coverage)
Percentage of error in genotype calling: 3.66%
False positive rate: 0.1%
False negative rate: 9.25%
comparing the genotype calls from 1583 affy snps with the NGS calls
957 hom , 562 het, 64 hom non ref
FP: proportionof hom reference for which NGS call is either heterozygous of homozygous non reference
FP number of genotypes where the NGS call shows a variant and affy dos not
FN number of genotypes whre affy calls a variant and NGS does not, the reason could be that NGS does not have coverage in all affy genomic positions

11. Data Analysis Pipeline
Fosmid Detection
Program
Read Alignment
against Genome
Fosmid Specific Matching Algorithm
Pairing
Fosmid Sequences
Based Phasing
Consensus
Calling SNP Analysis
Visualization &
MHC Database
In House Project Specific Analysis Pipeline
SOLiD Standard
Pipeline

12. Fosmids Detection Fosmid Detection Algorithm
Assign each read to a single 1kb long bin. Select bins with more than 5 reads
Perform allele calls for each heterozygous SNP. Mark bins with heterozygous calls
Cluster adjacent bins as belonging to the same fosmid if:
The gap distance between them is less than 10kb and
There are no bins with heterozygous SNPs between them
Keep fosmids with lengths between 3kb and 60kb
The silde shows (A) the HT Mapping of 14 fosmids in the MHC region. (B) shows the same fosmids but detected though SOLid sequencing. Also shown in (B) some landmark genes where HT mapping are not able to reliable predict the precence of fosmids in that region. (C) shows a detailed read statistics for all 19 detected fosmids of this individual well. The start and end position marks the beginning and end of the fosmid. The bins column show the length in kb. the reads column list the number of reads for this fosmid. the cov/base column shows the average coverage per base for each fosmid.
UCSC Genome browser Kent et al Genome Res. 12(6):

13. Size distribution of read-contigs
Fosmids Detection
Size distribution of read-contigs
20 – 50 kb
fosmid sized contigs

14. Data Analysis Pipeline
Fosmid Detection
Program
Read Alignment
against Genome
Fosmid Specific Matching Algorithm
Pairing
Fosmid Sequences
Based Phasing
Consensus
Calling SNP Analysis
Visualization &
MHC Database
In House Project Specific Analysis Pipeline
SOLiD Standard
Pipeline

15. Haplotyping
Locus
Event
Alleles Hap 1
Alleles Hap 2 1
SNV T C 2
Deletion - 3 A G 4
Insertion GC
Locus
Event
Alleles 1
SNV
C,T 2
Deletion
C,- 3
A,G 4
Insertion
-,GC
The process of grouping alleles that are present together on the same chromosome copy of an individual is called haplotyping

16. Single Individual Haplotyping
Input: Matrix M of m fragments covering n loci
Locus 1 2 3 4 5
... n f1 - f2 f3 fm

17. Single Individual Haplotyping
Input: Matrix M of m fragments covering n loci
Locus 1 2 3 4 5
... n f1 - f2 f3 fm

18. Single Individual Haplotyping
Input: Matrix M of m fragments covering n loci
Locus 1 2 3 4 5
... n f1 - f2 f3 fm

19. Single Individual Haplotyping
Input: Matrix M of m fragments covering n loci
Locus 1 2 3 4 5
... n f1 - f2 f3 fm

20. ReFHap Problem Formulation
For two alleles a1, a2
For two rows i1, i2 of M f1 - 1 f2
Score -1
s(M,1,2) = 1

21. ReFHap Problem Formulation
For a cut I of rows of M

22. ReFHap Algorithm
Reduce the problem to Max-Cut.
Solve Max-Cut
Build haplotypes according with the cut
Locus 1 2 3 4 5 f1 - f2 f3 f4 4 1 -1 3 1 2 1 -1 3 h h

23. ReFHap Algorithm Build G=(V,E,w) from M
Sort E from largest to smallest weight
Init I with a random subset of V
For each e in the first k edges
I’ ← GreedyInit(G,e)
I’ ← GreedyImprovement(G,I’)
If s(M, I) < s(M, I’) then I ← I’

24. ReFHap Algorithm
Classical greedy algorithm 1 3 4 2 1 3 4 2

25. ReFHap Algorithm
Edge flipping 1 2 3 4 2 1 3 4

26. Phasing the MHC: Mixed Diploid vs Fosmid-Based NGS
Libraries
Mate Pair & Paired End Genomic DNA
Paired End
16 Barcoded Pools
Uniquely Mapped
47 Gb
15 Gb
1/3rd
Number of Blocks
407 40
1/10th
Av. Block Length
438 bp
85 kb
194 x
Max. Block Length
3.7 kb
691 kb
186 x
Total Length all Blocks
178 kb
3.4 Mb
19 x
% of Phased SNPs
12 %
66 %
5 x

27. Phasing MHC: Preliminary Results
Number of blocks: 8
N50 block length: 793 kb
Maximum block length: 1.6 MB
Total extent of all blocks: 3.8 MB
Fraction of MHC phased into haplotype blocks: 95%
Number of heterozygous SNPs: 8030 SNPs
Fraction of SNPs phased: 86%

28. Acknowledgements Thank You! The Life Tech Team:
Margret
Hoehe
Anita
Suk
Thomas Hübsch
Roger Horton
Sabrina
Schulz
Steffi
Palczewski
Britta Horstmann
Gayle
McEwen
The Life Tech Team:
Kevin McKernan Alexander Sartori
Clarence Lee Dustin Holloway
Jessica Spangler Heather Peckham
Tristen Weaver Stephen McLaughlin
Tamara Gilbert Tim Harkins
Thank You!

29. Comparison Mapping algos
COX Haplotype simulated reads

30. Phasing MHC