Finishing Circular Assemblies. J Fass UCD Genome Center Bioinformatics Core Thursday April 16, 2015

Transcription

1 Finishing Circular Assemblies J Fass UCD Genome Center Bioinformatics Core Thursday April 16, 2015

2 Assembly Strategies de Bruijn graph Velvet, ABySS earlier, basic assemblers IDBA, SPAdes later, multi-k integration, multiple technologies Overlap-Layout-Consensus (OLC) wgs-assembler (aka Celera Assembler) PBcR (replaces pacbiotoca) used in Human Genome Project; used in PB s HGAP, MHAP (publication), FALCON SGA (String Graph Assembler) much faster overlap detection, lower memory but slower than de Bruijn graph assemblers

3 Bacterial Isolates Eight clones, each with agricultural importance Clones picked and grown on new plates, picked again so, should be pure (survey says - nope!) Genome size: 4-7 Mbp Each isolate has: 2-4 SMRT-Cells = ,000 filtered sub-reads ~ X base coverage 6-10M PE150 MiSeq reads ~ 500-1,000 X base coverage(!)

4 Vertebrate BACs BAC clones picked to cover a specific region where there s a suspected misassembly / novel gene(s) BAC sizes: Kbp Each BAC sequenced on a single SMRT-Cell (potentially ~10,000x coverage!)

5 Circularizing Assemblies Circular assemblers are a little like flying cars - you d think we d have one by now. geneious has one (as of early 2014), that builds circular sequences during assembly (rather than detecting circularity after the fact). But, how exactly does it work? Haven t evaluated AMOS circularization tool yet. Graph-based assemblers naturally can handle circularity (and one can view the graphs, in some cases), but all (?) current assemblers ignore circularity when simplifying to contigs / scaffolds.

6 Circularizing Assemblies by post-processing. Assuming the assembly process builds the circular sequence at least once, and then a little more, the 5 - and 3 -ends of the sequence will be nearly identical. So: 1. Find overlap by self-alignment 2. Trim one of the ends 3. (If desired) re-center sequence 4. Verify that expected depth of reads align across breakpoint

7 Finding Overlaps (self alignment) sprai (single pass read accuracy improver) is a new read correction tool, and has scripts / instructions for other parts of an assembly pipeline. check_circularity.pl script (uses megablast under the hood) did not work for any of the bacterial chromosomes or plasmids that had clear 3-5 Kbp overlaps.

8 Finding Overlaps (self alignment) One way to examine self-similarity is a dotplot: B A C T E R I U M B A C T E R I U M

9 Finding Overlaps (self alignment) which get confusing when all singlet matches are shown: B A N A N A R A M A B A N A N A R A M A

10 Finding Overlaps (self alignment) Better to show only, say, matching 10-tuples: Krumsiek 2007 Bioinformatics 23:1026

11 Finding Overlaps (self alignment) Zoom in until you can determine overlap coordinates:

12 Trim and Spin One of the circular (?) chromosomes showed an overlap of the first and last ~6.2 Kbp. Trim, 100 K and re-center ( spin ), pushing (for example) the last 100 Kbp to the front, so that the putative circularization point is now between positions 100,000 and 100, K

13 Verify circularization point What does an incorrect join look like? Align reads back to wrongly joined sequence? First, let s try this with BWA MEM ( bwa mem -M -x pacbio... ), and the filtered_subreads.fastq files produced in most SMRT-Portal protocols.

14 Bad join, filtered subreads

15 Bad join, filtered subreads

16 Verify circularization point What does an incorrect join look like? Align reads back to wrongly joined sequence? Er, let s try this with BWA MEM ( bwa mem -M... ), and the corrected.fastq file produced in the HGAP SMRT-Portal protocol (or wgs-assembler s PBcR script).

17 Bad join, corrected reads

18 Good join, corrected reads

19 OK, but what about hands-free? (a) (p)erfect (c)ircle? Tests for full end overlap using the LAST aligner, trims one version of overlap, joins ends and moves join to the center of output permuted sequence.

20 apc Comments, pull requests, rewrites in Python all welcome!

21 Back to BACs They re circular, but there s a fixed reference frame the genome (chromosome). 11.6kbp vector B C D A B C A B C D

22 Back to BACs Need a more general version of apc.pl Same functionality, but follows by linearizing starting after the vector sequence. What if the vector overlaps the ends? Partly overlaps?

23 Back to BACs Vertebrate BACs chosen to overlap across a region of interest, with three BACs. HGAP (SMRT-Portal) was not universally successful. perfect assembly BAC in pieces? + lots of E coli contigs gene 1 gene 2 just E coli contigs

24 BACs assembly BAC 2 resolved by removing reads that align to E coli. BAC 3 resolved by downsampling raw reads, then assembling with PBcR script (wgs-assembler). perfect assembly BAC in pieces? + lots of E coli contigs gene 1 gene 2 just E coli contigs

25 Thoughts on the process Permuting join to center is most general (easiest testing via alignments), but for comparison, a consistent start is better origin of replication? What s general for bacteria? Assemblers and aligners need to work on circular (or more topologically complicated) sequences graphs? Assemblers need to adapt to different coverage ranges better Finishing genomes is still hard :(