1. How to Build a Horse: Final Report
Megan Smedinghoff

2. Project Goals Goal 1 Goal 2 Goal 3
Produce an assembly of the horse genome using
the Celera Assembler
Goal 2
Reconcile my assembly with the Arachne assembly done by Broad Institute
in February 2007
Goal 3
Produce a parallelized version of the UMD Overlapper to help facilitate
easier assembly of large organisms in the future

3. DNA Cells in all organisms have a DNA molecule in their nucleus
DNA is an abbreviation for DeoxyriboNucleic Acid
DNA is translated into proteins, which determine the structure, function, and behavior of the cell

4. What does DNA look like? Linear, double-stranded, helical molecule.
Consists of four types of nucleotides (bases):
adenine (A)
thymine (T)
guanine (G)
cytosine (C)

5. What is a genome?
A genome is the linear sequence of bases of the DNA molecule: …..GATGACATGTAT…..
Bacteria: ~2-5 million bases
Insects (fly): ~200 million bases
Mammals: ~3 billion bases

6. The Horse Genome
In February 2007, Broad Institute released a draft assembly of the horse genome
The sequencing cost $15 million
The assembly contains approximately 2.7 billion bases and was done using 6.8-fold coverage
The horse that was sequenced

7. Why Sequence the Horse?
Allows scientists to study diseases that primarily affect horses such as Glanders
Over 80 known genetic conditions exist in both horses and humans; comparative genomics can lead to better treatments for both species
Work done on the horse will lead to improvements in the assembly pipeline and allow easier assembly of large mammals in the future

8. Review of Genome Sequencing
DNA target sample
SHEAR
SIZE SELECT
e.g., 10Kbp
± 8% std.dev.
Primer
End Reads (Mates)
SEQUENCE
750bp
Vector
LIGATE & CLONE
Slide courtesy of Art Delcher

9. Review of Assembly Process
Calculating Overlaps
We compare each read to every other read and record pairs whose ends
overlap by a certain amount (usually ~40 nucleotides)
AGTGATTAGATGATAGTAGA
| | | | | | | | | | | |
GATGATAGTAGAGGATAGATTTA
We use the University of Maryland overlapper to calculate the overlaps
UMD Overlapper is advantageous because it corrects for sequencing
errors and does additional trimming to help produce the most accurate
list of overlaps possible

10. Review of Assembly Process
Unitigs
We use overlap information to
combine reads into unitigs
Unitig
Contigs
We use mate pair information to
combine unitigs into contigs
Contig
Scaffolds
We further use mate pair
information to determine how
far apart contigs should be
(the result is called a scaffold)
Scaffold

11. Goal 1: Creating an Assembly
Running the Assembler
It took nine days for the Celera Assembler to produce an assembly
Celera generated over a terabyte of data
The final assembly used 24.9 million reads (7.9x coverage)
Results
Assembly
Celera
Arachne
Number of Scaffolds
59,044
9,687
Total Contigs In Scaffolds
126,810
55,316
Total Span of Scaffolds
2.5 billion bases
2.4 billion bases
N50 Contig Size
77,479 bases
112,381 bases

12. Goal 2: Produce a Reconciled Assembly
Step 1: Match the Celera assembly to the Arachne assembly
Step 2: Compare scaffolds from the two assemblies
Step 3: Run reconciliation software
Contig A
Contig B
Contig C
Contig D
Contig A’
Contig B’
Contig C’
Contig D’
Mapping of Celera Scaffold to Arachne Scaffold
Celera
Scaffold
Arachne

13. Orientation Problems Definition:
An orientation problem occurs when a set of Celera contigs matches a set of Arachne contigs, but one of the contigs is placed in the opposite orientation
Contig A
Contig B
Contig C
Contig D
Contig A’
Contig B’
Contig C’
Contig D’
Illustration of an orientation problem between the two assemblies
Celera
Scaffold
Arachne

14. Ordering Problems Definition:
An ordering problem occurs when a set of Celera contigs map to a set of
Arachne contigs, but the matches occur in a different order in each scaffold
Contig A
Contig B
Contig C
Contig D
Contig A’
Contig C’
Contig B’
Contig D’
Illustration of an ordering problem between the two assemblies
Celera
Scaffold
Arachne

15. Gap Size Problems Definition:
A gap problem occurs when the size of the Arachne gap is more than four
standard deviations away from the estimated Celera gap size
Contig A
Contig B
Contig C
Contig D
Contig A’
Contig B’
Contig C’
Contig D’
Illustration of an gap size problem between the two assemblies
Celera
Scaffold
Arachne
Gap

16. Results of the Comparison
Total Celera Scaffolds
59,044
Celera Scaffolds compared to Arachne
2,252 (96.6% of bases)
Celera Scaffolds with Orientation Problems 42
Celera Scaffolds with Ordering Problems
109
Celera Scaffolds with Gap Size Problems
162
Celera Scaffolds with at least one problem
174 (93.1% of bases)
Note:
In most Celera scaffolds, there was at most one contig that matched
a contig in the Arachne assembly. In these scaffolds, there was no
way to look for discrepancies between the two assemblies.

17. Reconciliation Gaps Compressions Assembly A Assembly B Assembly A
Matching sequences
Assembly A
Assembly B
Gap in Assembly B
Compressions
Assembly A
Assembly B
Matching sequences
Compression error in Assembly A or expansion error in Assembly B

18. Reconciliation Results
Assembly
Celera
Arachne
Reconciled
Number of Compressions
N/A
687
136
Assembly Size
2.512 billion bases
2.428 billion bases
2.433 billion bases
N50 Contig Size
77,479 bases
112,729 bases
130,123 bases

19. Goal 3: Parallelizing the Overlapper
R O M
M O R
Get associated hex value
for each 20-mer in read
Record “minimizer” (20-mer
with minimum hex value) for
sliding 20-base window
Create Read, Offset,
Minimizer file
Filter by certain minimizers
and then sort to get
candidate pairs for overlap
Overlaps
Additional
Post-Processing
Run “checker” on candidate pairs
and produce edit-transcripts
Output Overlap List

20. Parallelization of Overlapper
The parallelization occurs during the “checker” process when the reads with common minimizers are compared
Since the reads are already grouped by minimizers, there is no overhead for doing the comparisons in parallel
I used the Sun Grid Engine routines installed on the genome cluster to submit the jobs to available processors

21. Results of Parallelization
Bacteria
Current Overlapper
30 Minutes
Parallelized Overlapper
14 Minutes
Fly
Current Overlapper
21.5 Hours
Parallelized Overlapper
6.25 Hours
Horse
Current Overlapper
9 Days
Parallelized Overlapper
3.07 Days

22. Summary
Goal 1
I created an assembly of the horse using the Celera assembler. The
assembly size was 2.5 billion bases.
Goal 2
I created a reconciled assembly that fixed 551 out of 687
compressions in the Arachne assembly and increased the
assembly size by 4.3 million bases
Goal 3
I created a parallelized version of the overlapper that significantly
reduces the running time on larger genomes

23. Acknowledgements
Thanks to Aleksey Zimin, Guillaume Marçais, Mike Roberts, James Yorke,
and Poorani Subramanian for instruction and advice on this project
Thanks also to Matthew Pragel for moral support

24. More Technical Overlapper Slide
R O M
M O R
Get associated hex value
for each 20-mer in read
Record “minimizer” (20-mer
with minimum hex value) for
sliding 20-base window
Create Read, Offset,
Minimizer file
Filter by certain minimizers
and then sort to get
Candidate pairs for overlap
Overlaps
10-8
Run “checker” on candidate pairs
And produce edit-transcripts
Output Overlap List
Do Multi-compare and split
overlaps when they have
a certain number of differences
Do Poisson test and eliminate
any overlaps that have less than
a 10-8 probability of occurring