This document summarizes methods for characterizing structural variation and segmental duplication using next-generation sequencing data. It describes using mrFAST to align Illumina sequencing data to the reference genome to discover duplication regions. VariationHunter and CommonLAW were used to detect deletions by considering alignments across multiple samples simultaneously. Future work is proposed to extend the analysis to larger datasets, additional structural variant types, validation using long-read technologies, and application to other sequencing datasets.

1. Structural Variation and Segmental
Duplication Characterization
Can Alkan
Department of Computer Engineering
Bilkent University
Ankara, Turkey
Iman Hajirasouliha
Stanford University, Stanford, CA

2. Pre-Summary
 SD:
mrFAST to align the
roughly 70Gb Illumina
data for each
individual HG002,
HG003, HG004 to the
build37 reference
genome
mrcanavar to discover
the duplication regions
 SV:
mrFAST to align Illumina
data (~40X) them to
version build37 reference
genome
VariationHunter /
CommonLAW for deletions

3. Mapping to duplications
 Seed-and-extend based read mappers for Illumina
 mrFAST: with indel; mrsFAST: Hamming distance only
 All map locations are reported

4. mrcanavar
Reimplementation of :
 Alkan et al, “Personalized copy number and segmental
duplication maps using next-generation sequencing.”,
Nature Genetics, 2009
Also used in:
 Sudmant et al, Science, 2010; 1000 Genomes Project and SV companions
(2010,2011,2012,2015), Great Ape Diversity Project (Prado-Martinez et al,
Nature, 2013), several genome projects (gorilla, bonobo), Neandertal and
Denisova projects, dog domestication (Freedman et al, 2014), cat
domestication (Montague et al, 2014 and Tamazian et al 2014), etc.

5. Segmental duplications
Absolute copy numbers of genes are available

6. VariationHunter / CommonLAW
 Instead of detecting structural variation in samples
independently, considers all samples simultaneously.
 Uses all/multiple map locations and generalizes the
maximum parsimony structural variation into multiple
donor genomes.
 Finds potential cluster of deletions based on discordant
paired-end alignments among all individuals.
 Given a collection of discordant reads from the trio, find
a unique assignment of each read to one potential
deletion under maximum parsimony criteria.

7. Deletions (>50 bp)

8. Future directions
 Extend to full 300X dataset
 Include inversions, MEIs, insertions
 More in-depth analyses on de novo deletion and
duplication calls.
 Clean up false positives
 PacBio/nanopore in silico validation
 Start playing with other data, from 10X
Genomics, etc.