Alignment Approaches II: Long Reads
2 likes1,428 views
The document discusses advancements in genomic sequencing, particularly the challenges and opportunities presented by de novo assembly and long-read mapping using PacBio technology. It highlights the evolution of genome sequencing methods since the 1000 Genomes Project, emphasizing the need for high-contiguity human assemblies and the potential for better understanding genetic variations. Additionally, it addresses ongoing research topics, assembly statistics, and future tools for improving genome analysis.
Alignment Approaches II: Long Reads
- 1. FIND MEANING IN COMPLEXITY © Copyright 2014 by Pacific Biosciences of California, Inc. All rights reserved.For Research Use Only. Not for use in diagnostic procedures. Jason Chin (@infoecho) / Sept. 20 2014, GRC Workshop, Cambridge, UK Learning Genomic Structures From De Novo Assembly and Long-read Mapping de novol
- 2. Cost per Genome Dilemma 2 Sequencing cost is down for sure, but getting a de novo human genome that has the same scientific standard as the initial work does NOT follow Moore’s law. PacBio® CHM1: 4378 kb from just single random fragment library HGP, N50 ~100kb NCBI-34 Contig N50 29Mb HuRef: 107kb BGI YH: 7.4kb KB1: 5.5kb NA12878: 24kb CHM1: 144kb RP11: 127kb According to the NHGRI website, the definition of “sequencing a genome” changed in 2008. The 1000 Genomes Project starts in 2008, too.
- 3. Question Asked!! • Since the 1000 Genomes Project, we have learned a lot of about point mutations. Can we go beyond that? • What if we have 50, 100 or more human assemblies so we can address all genetic variations as much as possible? • Will one day all human genome sequencing be done in de novo fashion? – If so, how can we get ready for that as bioinformatists? 3 Evan Eichler , In Future Opportunities for Genome Sequencing and Beyond, July 28-29, 2014
- 4. Where We Are Now • One PacBio® human data set is publicly available, more are likely to come • Multiple groups have successfully assembled the public CHM1 data set independently with new algorithms from raw data • With new alignment/assembly tools from Gene Myers: one can assemble a genome in ~ 20,000 CPU-hours. (20X faster than 400,000+ CPU-hours from previous effort.) 4 New Assembly Statistics done With Daligner: #Seqs 5,058 Mean 562,695 Max 27,292,514 n50 5,265,098 Total 2,846,115,586 http://dazzlerblog.wordpress.com
- 5. What Can We Learn from High-contiguity De Novo Human Assemblies? 5
- 6. What Can We Learn from High-contiguity Human Assemblies? • Low-hanging Fruits – Calling SNPs (assembly not needed, but it helps) – Calling structure variants with whole-genome alignment approaches – Inferring repeats by coverage analysis • Assembly graph can provide information for understanding more complicated polymorphisms 6
- 7. Call SNPs / Example: HLA-B 7
- 8. Call Structure Variation By Whole-genome Alignment • Whole-genome alignments ( ~ 1 hr in a 32-core machine) – With multi-threaded Mummer – Clustering the hits with Mgaps and identified “gaps” in the alignments, convert to bed format for visualization 8 Structure Variants Called in Chromosome 1
- 9. Distribution of The Structure Variation Sizes • Number of insertions/deletions: 13796 SV calls (for insertion or deletion > 100 bp against hg19) 9
- 10. PacBio® vs. Short-read Alignment View for SV in the MHC region 10 318bp insertion
- 11. Assembly Graph 11 Each edge is associated with a sequence. Every path is a candidate of a model of part of the genome. From Gene Myers’ ISMB 2014 Keynote talk
- 12. Dissect a Contig from a String Graph The autonomy of a contig from a string graph layout 12 A contig: a linear non-branching path Each node: the begin (5’) or end (3’) of a read Each edge: a continuous sub- sequence from one read Ek: (V1, V2, Read, Range) = ( 00099576_1:B, 00101043_0:B, 00101043_0, 1991-‐0 ) Read 1: 00099576_1, Read 2: 00101043_0 In practice, we might just encode the paths in a contig rather than each single edge: C = (Ek, Ek+1, Ek+2, Ek+2) = (Pj Pj+1) V1 V2 V3 V4 V5 Ek Ek+1 Ek+2 Ek+3 V1 V3 V5 Pj Pj+1 C = =
- 13. Assembly String Graph of CHM1 Genome • Largest connect component: 31998 nodes, 39399 edges, ~36.5% (~1Gbp) of the human genome (total: 87572 nodes, 94530 edges) 13 Centromere? Casey Bergman: “it almost looks like an electron micrograph of the nucleus” #convergence
- 14. Polymorphism Structure vs. Local Assembly Graph Structure 14 SNPs SNPs SNPs SVsSVs Diploid Genome Segmental Duplication Similar String Graph
- 15. Identify Contigs: A New Proposal SNPs SNPs SNPs SVs SVs Associated contig 1 Associated contig 2 Primary contig 1 full length contig + 2 associated contigs Keep the long-range information while maintaining the relations of the alternative alleles.
- 16. Contig 4076 Alignment Around DPY19L2 Locus Same contig
- 17. Contig Graph and Segmental Duplication Contig 4076, one primary contig, 3 associate contigs, aligned to Chr7 and Chr12
- 18. Coting 4076 Alignment to Chr7 Same contig SV calls from CHM1 asm SV calls from GRC38
- 19. Local Neighborhood Subgraph of Contig 4076 19
- 20. Examining an Assembly Graph at Contig Level Around 1q21 • Contig graph, 1q21, contig 4108, another potential segmental duplication? 20
- 21. Another Intriguing Case 21 • Contig 4006 mapped to chr 9 The aligned region changes a lot in GRC38.
- 22. Contig Coverage Analysis 22 18.5 X 2 * 18.5 X 3 * 18.5 X High coverage long contigs 40 contigs > 100kbp > 2.5 * 18.5 X Poor assemblies, alignment artifacts, or sequence errors? High repeat elements
- 23. Checking the Complexity of the High-coverage Contigs 23 Contig 4006, 687kb, 53x coverage Contig 4235, 453k, 59x coverage Contig 3842, 235k, 54x coverage Warning: These contigs may not be 100% correctly assembled due to some nasty repeats. However, the local graphs give hints about the true genome structures.
- 24. How does the High-coverage Contig Look? 24 >2000X in this region
- 25. How does The High-coverage Contig Look? 25 High-coverage Region Alpha satellites?
- 26. For Research Use Only. Not for use in diagnostic procedures. Extreme Repeats 26
- 27. Identify Centromere Alpha-satellite Structure • Most of the nasty contig graphs are around the centromere. Currently, it remains hard to get long contigs around those very long tandem repeats. • However, we can still learn many useful things from long-read data • Tool In Development: α-Centauri for identifying different high-order repeat structures (https://github.com/volkansevim/alpha-CENTAURI, Volkan Sevim, Ali Bashir & Karen Miga ) 27
- 28. Centromere Alpha Satellites Have Non-trivial High-order Repeat Structure 28 Karen Miga
- 29. Example: A Read Reconstructs a 24-mer HOR 29 Align monomer to each other to identify near identical mon0mers Identify HOR with the monomer IDs and positions 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 171819 20 21 22 23 24
- 30. Many Other Open Topics • Low-coverage assembly: cost vs. quality analysis • Phasing for haplotypes • Crowd-sourcing infrastructure for examining / annotating / correcting genome assemblies • Evaluation about SNPs calling with short reads on better assembly • Large-scale comparative genomes with de novo assemblies • Assembly-graph data format • Visualization Techniques • Combining other data types, e.g. optical mapping 30 It is a very exciting time. We still need more tools to harvest information to generate new knowledge.
- 31. For Research Use Only. Not for use in diagnostic procedures. Pacific Biosciences, the Pacific Biosciences logo, PacBio, SMRT, SMRTbell and Iso-Seq are trademarks of Pacific Biosciences in the United States and/or other countries. All other trademarks are the sole property of their respective owners. 31