1. High-Throughput Sequencing
Advanced Microarray Analysis
BIOS , 2008
Dr. Mark Reimers, VCU

2. Quantitative HTS - Outline
Technology
Preprocessing
Quantitative analysis
Applications
ChIP-Seq
RNA-Seq
Methyl-Seq

3. The Technology
Most sequencing proceeds by addition of fluor-labeled bases
Do this in parallel on a flat surface
Capture each stage with good camera
Align images

4. Roche - 454
Parallel Pyrosequencing on beads

5. Mardis, Trends in Genetics

6. 454 Sequencing Operation

7. Illumina - Solexa

8. ABI SOLiD Resquencing each fragment with different primers
Reconstruct each fragment separately

10. Paired-End Reads

11. Issues Pre-processing Quantitative analysis Base calling Mapping readsQA
Quantitative analysis
Variation and noise
Biases
Models
Accuracy and validation

12. Pre-processing – Base Calling
Not all steps completed properly
Sequence can lag behind or skip ahead
Hence most light spots a mixture of different colors
Simple rule: use brightest signal

13. Types of mismatches in uniquely mapped tags with a single mismatch are profoundly asymmetric and biased
Courtesy Thierry-Mieg

14. Typical Errors in Base-Calling

15. Position of single mismatch in uniquely mapped tags
Courtesy Thierry-Mieg

16. Improving Base-Calling with SVM

17. Pre-processing – Mapping Reads
Huge numbers (10M – 70M)
BLAT (2002 high-speed method)
Eland (proprietary Illumina)
Other new methods: MAQ, SOAP

18. Quality Assessment Fraction of reads mapping to targets
Typically 5-10M reads per lane and 60-80% map to targets
Some repetitive sequence

19. Comparing Samples - A Simple Normalization
Different numbers of counts per lane
Divide counts in a region of interest (a genomic region or a gene or an exon) by all counts (total per million reads -TPM)
For comparing genomic regions of different lengths divide also by length of region TPKM (total per kilobase per million)

20. Quant. Analysis - Variation
Poisson model often used for random variation
Most HTS data ‘over-dispersed’ relative to Poisson
Negative Binomial often used
Parameter fitted

21. Quantitative Analysis - Biases
Not all regions represented equally
GC rich regions represented more
Independent of GC some chromosome regions represented more
Euchromatin bias
Sequence initiation site biases
‘Mapability’ biases – some regions won’t have any uniquely mapped tags

22. GC Bias Density of reads depends strongly on GC content of regions

23. Genomic Position Biases
Count tags from randomly sheared DNA in red with GC content in blue

24. Start Position Bias

25. Consistent Start Position Bias
Counts per start site in lane 1 vs lane 2

26. RNA-Seq

27. RNA-Seq Data Gene Model Kidney Reads Liver Reads
From Marioni et al 2008

28. Accuracy of Illumina RNA-Seq

29. Comparing RNA-Seq & Affy
Issues
How replicable is RNA-Seq?
How consistent are the two technologies?
Which is better?
Marioni et al, Genome Research, 2008

30. Comparing Fold-Changes
D.E. by ILM
Red >250
Green <250
Black Not DE by ILM

31. Model for Variation Poisson counts hypergeometric comparison
Make uniform p-values
by adding random term
Use lower tails only

32. False Positive Rates
QQ-plots of p-values between tech. reps

33. Different Concentrations are NOT Comparable!
QQ-plots of p-values between 3pM and 1.5 pM

34. Normalization of RNA-Seq
Robinson et al noticed that most genes appeared less expressed in liver
Fig 1 from Robinson & Oshlak, Genome Biology 2010

35. A Better Normalization for RNA-Seq - TMM
Drop extremes of ratios
Drop very high count genes
Compute trimmed means of samples
Center log-ratios between samples

36. New Things to do with RNA-Seq
Allele-specific expression
Splice variation
Between tissues
In disease
Alternate initiation sites
Select 5’ capped RNA fragments
Alternate termination

37. Allelic Comparison
It is possible to compare allele-specific expression counts
Sample from VCU
Replicate samples
P-values for binomial tests of equality
About half show differential expression!

38. Detecting Splice Variation
Deep sequencing shows up clear variation in exon usage
Wang et al Nature 2008

39. Tissue Map of Splice Variation
From Wang et al
Brain is most distinctive
Individuals seem to differ
Cell lines seem to have distinct splice patterns

40. Splicing is Complex Many different splice operations exist
Only some of these characterized by counting exon reads

41. Issues in Detecting Splice Variants
Counts in exons reflect biases (as yet uncharacterized) as well as actual abundance
Reads that bridge splice junctions would be definitive but mapping is very dubious with short (<40 base) reads
All possible splice junctions are not known
Hard to even search through the known ones

42. Methodology for Splice Variants
Count reads mapped to exons and and compare ratios across samples
Wang et al, and most others
Count reads that cross splice junctions

43. Methodology for Finding Junctions

44. ChIP-Seq

45. Chromatin Immuno-precipitation

46. ChIP-Seq Workflow Cross-link proteins to DNA Fragment DNA
Extract with antibody
Reverse cross links
Sequence fragments
DO CONTROLS!

47. ChIP-Seq Data
From Rozowsky et al, Nature Biotech 2009

48. ChIP-Seq vs ChIP-chip

49. Peak-Finding - Simple Extend tags and count overlap
How much to extend?

50. Peak Finding – Better
Tags starting on opposite strands are likely to start at opposite ends
Identifying the cross-over point leads to improved accuracy

51. The Value of Controls: ChIP vs. Control Reads
Red dots are windows containing
ChIP peaks and black dots are windows
containing control peaks used for
FDR calculation

52. Cause of Variation in Read Density
In study of FoxA1 binding, even control reads enriched near FoxA1 binding site!
Probably due to open chromatin near FoxA1 binding site
Density of Control Channel
reads around FoxA1 site
Courtesy Shirley Liu

53. ChIP-Seq – MACS Key Ideas
Smart peak imputation estimate
Uses read directions
Empirical estimate of fragment length
Local frequency estimate
Using control, if available
Using wide estimate, otherwise
Not using sequence

55. Read Lengths and Directions
Some clear clusters – even before stats
Reads on opposite sides of peak map to opposite strands
Hence fragments have opposite directions
Can estimate apparent fragment length

56. Fragment Lengths
Puzzle: Fragments from sonication expected to be between 200 – 500 bp
Estimated fragment size ~ 100 bp
Shirley Liu’s explanation: preferential cutting near to TF ??

57. Comparison to ChIP-chip
Broad correlation
Not dramatic improve-ment in precision !

58. Methyl-Seq

59. Methylation Assays
Affinity purification: e.g. MeDIP-Seq (methylated dinucleotide immunoprecipitation)
Methylation-specific cleavage by endonucleases
e.g. Methyl-Seq: Cleaves with HPA2 to identify
Bisulphite conversion
WGBS (Whole-Genome Bisulphite Sequencing)
RRBS (Reduced Representation Bisulphite Sequencing)
Cleaves with MSPI to reduce complexity

60. Affinity: MeDIP-Seq & MBD-Seq

61. Issues with Affinity Methods
Analysis essentially like ChIP-Seq
BUT: Sequence count reflects both density of CpG’s and proportions of methylation
No individual CpG-level information
Advantages: no conversion so sequence tags are easily mappable

62. Methyl-Seq Use HPAII to cleave only at unmethylated CCGG sites
Size-select fragments (50-300)
Sequence fragment ends
Always starting at a CCGG
Easy to map – few possible loci (<1M)
Paired ends give actual fragment

63. Schematic Here

64. Issues for Methyl Seq
Computational problem to re-assemble actual proportions of methylation at each locus from counts
Prone to false positives because of incomplete digestion (for reasons other than methylation of CCGG site)
e.g. insufficient time …
rates vary by 50-fold depending on sequence context

65. WGBS Bisulphite conversion, fragmentation and shotgun sequencing
Requires very many reads!
Use of capture arrays reduces work… BUT different sequences have different capture efficiencies!

66. WGBS Data (from capture array)
top, CHP-SKN-1; bottom, MDA-MB-231
NB. Inconsistent tag numbers

67. Issues with WGBS Lose many C’s Hard to map to genome
Strategy depends on less penalty for mapping T to C
Too many loci!

68. RRBS Too many methylation sites in genome
Cleave with MSPI and size select in order to reduce number of fragments
Convert C to T with bisulphite (not mC)
Then sequence fragments
1.4 M fragments

69. Issues with RRBS
Fairly broad but not complete coverage of ‘interesting’ regions of genome
Bisulphite conversion of limited regions means mapping is fairly easy
Bisulphite conversion not always complete

70. Meta-Genomics

71. What is Meta-genomics?
Sequencing random fragments of DNA from all microbial denizens of a community (and traces of a few others)
Sometimes broadly used for surveys of microbial diversity based on sequencing all 16SrRNA genes present

72. Kinds of Questions What is out there?
Most microbial species not known
What metabolic fluxes in any environment?
What microbes associated with specific conditions?
Including disease or health
Human Microbiome Project

73. Environmental Meta-Genomics

74. Human Microbiome Project

75. Data Analysis Issues – 16S rRNA
Identification of microbes – most are unknown and un-culturable
Distinguishing errors in sequencing from novel microbes
Biases in sequencing

76. Data Analysis Issues - Metagenomics
Mapping and characterizing unknown protein sequences
Usually assume conservation
Full-coverage allows assembly of genomes
Counting
Biases probably smaller (Bork)