genome assembly: then and now (with notes) — v1.2
DESCRIPTION
This was a talk given on 2014-09-17 for the Genome Center’s Bioinformatics Core as part of a 1 week workshop. It concerns the Assemblathon projects as well as other aspects relating to genome assembly. A version of this talk is also available on Slideshare without notes. Note, this is an evolving talk. There are older and newer versions of the talk also available on slideshare.TRANSCRIPT
Genome assembly: then and nowKeith Bradnam
Image from Wellcome Trust
v1.2
Author: Keith Bradnam, Genome Center, UC Davis This work is licensed under a Creative Commons Attribution 4.0 International License. This was a talk given on 2014-09-17 for the Genome Center’s Bioinformatics Core as part of a 1 week workshop on learning the command-line. Other versions of this talk are probably available at slideshare.com
Image from flickr.com/photos/dougitdesign/5613967601/
Contents
Sequencing 101!! Genome assembly: then!! Genome assembly: now
Assemblathons!! Intermission!!Advice
Sequencing 101A, C, G, T...
Image from nlm.nih.gov
Fred Sanger, who invented the sequencing technology that helped sequence most of the good quality genomes that are out there. He was also a winner of two Nobel prizes.
Read
Most sequencing technologies start with a sequencing read. A read could be as short as 25 bp (Solexa sequencing from a few years ago), or >25,000 bp (e.g. with PacBio or Oxford Nanopore). The record read length is currently held by PacBio and is over 50,000 bp.
Read pair
Most sequencing is done with pairs of connected reads, separated by a short interval whose approximate length is known. Not all reads will have this exact ‘insert size’. There can be a LOT of variation. Read pairs can also overlap with each other.
Read pair
Mate pair
Mate pairs, also known as jumping pairs, have much larger inserts (thousands or tens of thousands of bp), but it is hard to make good mate pair libraries. Having very large inserts is very useful for the purposes of genome assembly. Again, there is a lot of variation in the actual size of inserts (as determined by mapping mate pairs back to a known reference).
If you sequence a lot of read pairs, hopefully they will overlap with each other and allow you to start making contiguous sequences...
Contigs
...which are better known as contigs.
Mate pairs — or other information — can hopefully be used to connect contigs together into scaffolds. The unknown gap between contigs is replaced with unknown bases (Ns). Some scaffold sequences can therefore end up containing a lot of Ns.
Mate pairs — or other information — can hopefully be used to connect contigs together into scaffolds. The unknown gap between contigs is replaced with unknown bases (Ns). Some scaffold sequences can therefore end up containing a lot of Ns.
ScaffoldNNNNNNNNNNNNNNNNNNN
Mate pairs — or other information — can hopefully be used to connect contigs together into scaffolds. The unknown gap between contigs is replaced with unknown bases (Ns). Some scaffold sequences can therefore end up containing a lot of Ns.
Assembly size
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
15
15
15
5
Assembly size is simply the sum of all scaffolds or contigs that are included in the final genome assembly. If you are calculating the assembly size from scaffolds, then some fraction of that final size will come from the Ns in scaffold sequences. !Here we have a toy genome assembly, with 12 scaffolds totaling 200 Mbp.
Assembly size
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
200 Mbp
15
15
15
5
Assembly size is simply the sum of all scaffolds or contigs that are included in the final genome assembly. If you are calculating the assembly size from scaffolds, then some fraction of that final size will come from the Ns in scaffold sequences. !Here we have a toy genome assembly, with 12 scaffolds totaling 200 Mbp.
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
200 Mbp
15
15
15
5
The most widely used measure to describe genome assemblies is the N50 length of scaffolds or contigs. This is essentially a weighted mean, designed to be more informative than a crude mean length (which is not very useful if you end up with thousands of very short scaffolds/contigs). To calculate the N50 scaffold length, start with the length of the longest scaffold...
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
200 Mbp
15
15
15
5
The most widely used measure to describe genome assemblies is the N50 length of scaffolds or contigs. This is essentially a weighted mean, designed to be more informative than a crude mean length (which is not very useful if you end up with thousands of very short scaffolds/contigs). To calculate the N50 scaffold length, start with the length of the longest scaffold...
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
200 Mbp
15
15
15
5
70
The most widely used measure to describe genome assemblies is the N50 length of scaffolds or contigs. This is essentially a weighted mean, designed to be more informative than a crude mean length (which is not very useful if you end up with thousands of very short scaffolds/contigs). To calculate the N50 scaffold length, start with the length of the longest scaffold...
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
15
15
15
5
200 Mbp
95
…if this length does not exceed 50% of the total assembly size (50% is why it is called N50), then proceed to the next longest scaffold, and add that length to a running total.
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
15
15
15
5
200 Mbp
95
…if this length does not exceed 50% of the total assembly size (50% is why it is called N50), then proceed to the next longest scaffold, and add that length to a running total.
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
15
15
15
5
200 Mbp
115
After looking at 3 scaffolds, we have exceeded 50% of the total assembly size.
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
15
15
15
5
200 Mbp
115
After looking at 3 scaffolds, we have exceeded 50% of the total assembly size.
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
55
15
15
15
5
200 Mbp
The length of the contig or scaffold that takes you past 50% is what is reported as the N50 length. So here, we have an N50 length of 20 Mbp.
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
5
15
15
15
5
5
N50 may be more robust than using a simple mean length, but it can still be easily manipulated. What if we excluded the two shortest scaffolds from our assembly?
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
5
15
15
15
5
5
N50 may be more robust than using a simple mean length, but it can still be easily manipulated. What if we excluded the two shortest scaffolds from our assembly?
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
5
15
15
15
N50 may be more robust than using a simple mean length, but it can still be easily manipulated. What if we excluded the two shortest scaffolds from our assembly?
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
5
15
15
15
Now the total assembly size is 10 Mbp smaller, which is only a 5% reduction, but the N50 increases to 25 Mbp...a 25% increase. If these were two different assemblies and you only saw an N50 of 25 Mbp vs N50 of 20 Mbp, you might think that the first assembly was much better.
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
5
15
15
15
190 Mbp
Now the total assembly size is 10 Mbp smaller, which is only a 5% reduction, but the N50 increases to 25 Mbp...a 25% increase. If these were two different assemblies and you only saw an N50 of 25 Mbp vs N50 of 20 Mbp, you might think that the first assembly was much better.
N50 length
NNNNNNNNNNNNNNNNNNN
NNNNNNNNNNN
NNNNNNNNNNN
70 25
20
10
10
5
5
15
15
15
190 Mbp
Now the total assembly size is 10 Mbp smaller, which is only a 5% reduction, but the N50 increases to 25 Mbp...a 25% increase. If these were two different assemblies and you only saw an N50 of 25 Mbp vs N50 of 20 Mbp, you might think that the first assembly was much better.
N50 for two assemblies
Here are another two fictional assemblies. The first assembly now has a lower N50 value, but this is purely because it contains more sequence (which come from very short scaffolds). Do you want more sequence in your assembly, or fewer but longer sequences?
N50 for two assemblies
208 Mbp 190 Mbp
Here are another two fictional assemblies. The first assembly now has a lower N50 value, but this is purely because it contains more sequence (which come from very short scaffolds). Do you want more sequence in your assembly, or fewer but longer sequences?
N50 for two assemblies
208 Mbp 190 Mbp
N50 = 15 Mbp N50 = 25 Mbp
Here are another two fictional assemblies. The first assembly now has a lower N50 value, but this is purely because it contains more sequence (which come from very short scaffolds). Do you want more sequence in your assembly, or fewer but longer sequences?
NG50 for two assemblies
208 Mbp 190 Mbp
We prefer a measure called NG50. This does not use the assembly size, but instead uses the known (or estimated) genome size (the 'G' in NG50 refers to the Genome). We first used this measure in the Assemblathon 1 paper and (thankfully) it has seen some adoption by the assembly community.
NG50 for two assemblies
We prefer a measure called NG50. This does not use the assembly size, but instead uses the known (or estimated) genome size (the 'G' in NG50 refers to the Genome). We first used this measure in the Assemblathon 1 paper and (thankfully) it has seen some adoption by the assembly community.
NG50 for two assemblies
Expected genome size = 250 Mbp
We prefer a measure called NG50. This does not use the assembly size, but instead uses the known (or estimated) genome size (the 'G' in NG50 refers to the Genome). We first used this measure in the Assemblathon 1 paper and (thankfully) it has seen some adoption by the assembly community.
Expected genome size = 250 Mbp
NG50 for two assemblies
The NG50 of these two assemblies is now the same. We think that NG50 is a fairer way of comparing genome assemblies that might differ in their total size.
NG50 = 15 Mbp NG50 = 15 Mbp
Expected genome size = 250 Mbp
NG50 for two assemblies
The NG50 of these two assemblies is now the same. We think that NG50 is a fairer way of comparing genome assemblies that might differ in their total size.
People always seem to want higher N50 values, so I recently published a tool called 'N50 Booster!!!' that can increase the N50 length of any genome assembly.
$ n50_booster.pl c_japonica.WS230.genomic.fa!!Before:!==============!Total assembly size = 166256191 bp!N50 length = 94149 bp!!Boosting N50...please wait!!After:!==============!Total assembly size = 166256191 bp!N50 length = 104766 bp!!Improvement in N50 length = 10617 bp!!See file c_japonica.WS230.genomic.fa.n50 for your new (and improved) assembly
This is some real output from my 'N50 Booster!!!' script. In this case, it increased the N50 length of the Caenorhabiditis japonica assembly from 94.1 Kbp to 104.8 Kbp. Note that, amazingly, the assembly size remains the same!
Notice the date? This was indeed an April Fool's prank, but the script is only being slightly dishonest. The first thing it does is simply discard the shortest 25% of all sequences. Then it adds an equivalent length of Ns to some of the remaining contigs (to preserve the assembly size). Manipulating assemblies like this is unscientific, but most genome assemblers will remove some of the shortest sequences. Which ones should you keep or remove?
Notice the date? This was indeed an April Fool's prank, but the script is only being slightly dishonest. The first thing it does is simply discard the shortest 25% of all sequences. Then it adds an equivalent length of Ns to some of the remaining contigs (to preserve the assembly size). Manipulating assemblies like this is unscientific, but most genome assemblers will remove some of the shortest sequences. Which ones should you keep or remove?
You should check that high N50 values!are not simply due to lots of Ns in the scaffolds!
You should always look at your assembly before you do anything with it!
Assembly 'x'
In this assembly, which I was asked to run CEGMA on, it turned out to be 91% N! This assembly is not going to be good for anything. There are lots of other ambiguity characters too (e.g. R for puRine).
Assembly 'x'
Size: 859 Mbp!!
Number of scaffolds: 28!!
N50 = 70.3 Mbp
In this assembly, which I was asked to run CEGMA on, it turned out to be 91% N! This assembly is not going to be good for anything. There are lots of other ambiguity characters too (e.g. R for puRine).
Assembly 'x'
Size: 859 Mbp!!
Number of scaffolds: 28!!
N50 = 70.3 Mbp
Ns = 90.6% !!!
In this assembly, which I was asked to run CEGMA on, it turned out to be 91% N! This assembly is not going to be good for anything. There are lots of other ambiguity characters too (e.g. R for puRine).
Assembly 'x'
Size: 859 Mbp!!
Number of scaffolds: 28!!
N50 = 70.3 Mbp
Ns = 90.6% !!!
In this assembly, which I was asked to run CEGMA on, it turned out to be 91% N! This assembly is not going to be good for anything. There are lots of other ambiguity characters too (e.g. R for puRine).
Basic assembly metrics
Apart from assembly size, and N50/NG50 length, there are many other ways to describe a genome assembly.
Basic assembly metrics
Metric Description
Assembly size With or without very short contigs?
N50 / NG50 For contigs and/or scaffolds
Coverage When compared to a reference sequence
Errors Base errors from alignment to reference sequence !and/or input read data
Number of genes From comparison to reference transcriptome !and/or set of known genes
Apart from assembly size, and N50/NG50 length, there are many other ways to describe a genome assembly.
Basic assembly metrics
Metric Description
Assembly size With or without very short contigs?
N50 / NG50 For contigs and/or scaffolds
Coverage When compared to a reference sequence
Errors Base errors from alignment to reference sequence !and/or input read data
Number of genes From comparison to reference transcriptome !and/or set of known genes
And many, many more...
Apart from assembly size, and N50/NG50 length, there are many other ways to describe a genome assembly.
Genome assemblyBack in the day...
How were genomes assembled back in the late 1990s when genome sequencing projects were starting to make the news?
Genome assemblyBack in the day...
1998
How were genomes assembled back in the late 1990s when genome sequencing projects were starting to make the news?
Genome assembly: then
Genome sequencing projects often had a fantastic amount of supporting material which helped put the genome together. They were further helped by targeting genomes which had low heterozygosity. And of course this was all done with Sanger sequencing which gave long, accurate reads.
Genetic maps ✓
Genome assembly: then
Genome sequencing projects often had a fantastic amount of supporting material which helped put the genome together. They were further helped by targeting genomes which had low heterozygosity. And of course this was all done with Sanger sequencing which gave long, accurate reads.
Genetic maps ✓ Physical maps ✓
Genome assembly: then
Genome sequencing projects often had a fantastic amount of supporting material which helped put the genome together. They were further helped by targeting genomes which had low heterozygosity. And of course this was all done with Sanger sequencing which gave long, accurate reads.
Genetic maps ✓ Physical maps ✓Understanding of target genome ✓
Genome assembly: then
Genome sequencing projects often had a fantastic amount of supporting material which helped put the genome together. They were further helped by targeting genomes which had low heterozygosity. And of course this was all done with Sanger sequencing which gave long, accurate reads.
Genetic maps ✓ Physical maps ✓Understanding of target genome ✓Haploid / low heterozygosity genome ✓
Genome assembly: then
Genome sequencing projects often had a fantastic amount of supporting material which helped put the genome together. They were further helped by targeting genomes which had low heterozygosity. And of course this was all done with Sanger sequencing which gave long, accurate reads.
Genetic maps ✓ Physical maps ✓Understanding of target genome ✓Haploid / low heterozygosity genome ✓Accurate & long reads ✓
Genome assembly: then
Genome sequencing projects often had a fantastic amount of supporting material which helped put the genome together. They were further helped by targeting genomes which had low heterozygosity. And of course this was all done with Sanger sequencing which gave long, accurate reads.
Genetic maps ✓ Physical maps ✓Understanding of target genome ✓Haploid / low heterozygosity genome ✓Accurate & long reads ✓Resources (time, money, people) ✓
Genome assembly: then
Genome sequencing projects often had a fantastic amount of supporting material which helped put the genome together. They were further helped by targeting genomes which had low heterozygosity. And of course this was all done with Sanger sequencing which gave long, accurate reads.
So what was the result of spending millions of dollars !to assemble genomes of well-characterized species,!with accurate long reads, and detailed maps???
So hopefully this gave us a useful set of finished genomes, right?
✤ 2000: published genome size = 125 Mbp
✤ 2007: genome size = 157 Mbp
✤ 2012: genome size = 135 Mbp
Arabidopsis thaliana
Many published genome sizes are sometimes based on estimates which can be wrong. As they sequenced more and more of the Arabidopsis genome, they had to revise how big it was. So between 2000 and 2007 they produced more sequence but paradoxically it became less complete because the estimate of the size went up. Now it has come back down again. But the genome remains unfinished.
✤ 2000: published genome size = 125 Mbp
✤ 2007: genome size = 157 Mbp
✤ 2012: genome size = 135 Mbp
✤ Amount sequenced = 119 Mbp
Arabidopsis thaliana
Many published genome sizes are sometimes based on estimates which can be wrong. As they sequenced more and more of the Arabidopsis genome, they had to revise how big it was. So between 2000 and 2007 they produced more sequence but paradoxically it became less complete because the estimate of the size went up. Now it has come back down again. But the genome remains unfinished.
✤ 2000: published genome size = 125 Mbp
✤ 2007: genome size = 157 Mbp
✤ 2012: genome size = 135 Mbp
✤ Amount sequenced = 119 Mbp
✤ Ns = 0.2% of genome
Arabidopsis thaliana
Many published genome sizes are sometimes based on estimates which can be wrong. As they sequenced more and more of the Arabidopsis genome, they had to revise how big it was. So between 2000 and 2007 they produced more sequence but paradoxically it became less complete because the estimate of the size went up. Now it has come back down again. But the genome remains unfinished.
Drosophila melanogaster
✤ Genome published 1998
✤ Heterochromatin finished 2007
The fly genome was 'finished' in 1998. But this was only really the easy-to-sequence portion of the genome (the euchromatin). The trickier heterochromatin was sequenced as a separate project that didn't finish until almost a decade later. The fly genome remains unfinished.
Drosophila melanogaster
✤ Genome published 1998
✤ Heterochromatin finished 2007
✤ Ns = 4% of genome
The fly genome was 'finished' in 1998. But this was only really the easy-to-sequence portion of the genome (the euchromatin). The trickier heterochromatin was sequenced as a separate project that didn't finish until almost a decade later. The fly genome remains unfinished.
Caenorhabditis elegans
✤ Genome published 1998
✤ 2004: last N removed
The worm genome has no unknown bases in it. However, since the publication of the genome sequence the genome has continued to be refined as errors are corrected. The last big batch of changes all occurred fairly recently (November 2012). So after almost 15 years of post-genome-publication, it was still possible to find over 1,400 errors in one of the best characterized genome sequences that exists.
Caenorhabditis elegans
✤ Genome published 1998
✤ 2004: last N removed
✤ 1998–2014: genome sequence changes
The worm genome has no unknown bases in it. However, since the publication of the genome sequence the genome has continued to be refined as errors are corrected. The last big batch of changes all occurred fairly recently (November 2012). So after almost 15 years of post-genome-publication, it was still possible to find over 1,400 errors in one of the best characterized genome sequences that exists.
Caenorhabditis elegans
✤ Genome published 1998
✤ 2004: last N removed
✤ 1998–2014: genome sequence changes
✤ 558 insertions
✤ 230 deletions
✤ 614 substitutions
The worm genome has no unknown bases in it. However, since the publication of the genome sequence the genome has continued to be refined as errors are corrected. The last big batch of changes all occurred fairly recently (November 2012). So after almost 15 years of post-genome-publication, it was still possible to find over 1,400 errors in one of the best characterized genome sequences that exists.
Caenorhabditis elegans
✤ Genome published 1998
✤ 2004: last N removed
✤ 1998–2014: genome sequence changes
✤ 558 insertions
✤ 230 deletions
✤ 614 substitutions
} Nov 2012
The worm genome has no unknown bases in it. However, since the publication of the genome sequence the genome has continued to be refined as errors are corrected. The last big batch of changes all occurred fairly recently (November 2012). So after almost 15 years of post-genome-publication, it was still possible to find over 1,400 errors in one of the best characterized genome sequences that exists.
Saccharomyces cerevisiae
✤ Genome published 1997
✤ 12 Mbp genome
✤ 1,653 changes to genome since 1997
Likewise in yeast. The first eukaryotic genome sequence continues to receives fixes to correct the sequence. The last set of changes were made in 2011. These changes affected coding sequences, not just intergenic and intronic DNA.
Saccharomyces cerevisiae
✤ Genome published 1997
✤ 12 Mbp genome
✤ 1,653 changes to genome since 1997
✤ Last changes made in 2011
Likewise in yeast. The first eukaryotic genome sequence continues to receives fixes to correct the sequence. The last set of changes were made in 2011. These changes affected coding sequences, not just intergenic and intronic DNA.
Genetic maps ✓ Physical maps ✓Understanding of target genome ✓Haploid / low heterozygosity genome ✓Accurate & long reads ✓Resources (time, money, people) ✓
Genome assembly: then
And all of this was done in an era when we had all of these supporting materials.
Genetic maps ✗
Physical maps ✗
Understanding of target genome ✗
Haploid / low heterozygosity genome ✗
Accurate & long reads ✗
Resources (time, money, people) ✗
Genome assembly: now
We don't have these now! Genome sequencing no longer requires an international consortium, rather it could be a project for a Grad student.
Assembling & finishing!a genome is not easy!
It was never easy, even when we access to lots of resources to help us put together genomes. And it is not easy now. Don't be fooled into thinking that because there are many published genome sequences, that these sequences represent the absolute ideal genome sequence. !And don’t be fooled that just because you can afford to sequence a genome, that you will have the resources to make a useful assembly from that sequence data.
AssemblathonsA new idea is born
Image from flickr.com/photos/dullhunk/4422952630
The Assemblathon was born out of the Genome 10K project.
If you sequence 10,000 genomes...!...you need to assemble 10,000 genomes
The Assemblathon was born out of the Genome 10K project.
How many assembly tools are out there?
There are many, many tools out there for assembling, or helping to assemble, a genome sequence (there are 125 on this page). People may not have the time, patience, or expertise to try more than a handful of these. But…
bambus2
How many assembly tools are out there?
Ray
Celera
MIRA
ALLPATHS-LG
SGACurtain MetassemblerPhusion
ABySS
Amos
Arapan
CLC
Cortex
DNAnexus
DNA Dragon
EdenaForge
GeneiousIDBA
Newbler
PRICE
PADENA
PASHA
Phrap
TIGR
Sequencher
SeqMan NGen
SHARCGS
SOPRA
SSAKE
SPAdes
Taipan
VCAKE
Velvet
Arachne
PCAP
GAM
MonumentAtlas
ABBA
Anchor
ATAC
Contrail
DecGPU GenoMinerLasergene
PE-Assembler
Pipeline Pilot
QSRA
SeqPrep
SHORTY
fermiTelescoper
QuastSCARPA
Hapsembler
HapCompass
HaploMerger
SWiPS
GigAssembler
MSR-CA
MaSuRCA
GARM
Cerulean
TIGRA
ngsShoRT
PERGA
SOAPdenovo
REAPR
FRCBam
EULER-SR SSPACE
Opera
mip
gapfiller
image
PBJelly
HGAP
FALCON
Dazzler
GGAKE
A5
CABOG
SHRAPSR-ASM
SuccinctAssembly
SUTTARagout
Tedna
Trinity
SWAP-Assembler
SILP3
AutoAssemblyD
KGBAssembler
MetAMOS
iMetAMOS
MetaVelvet-SL
KmerGenie
Nesoni
Pilon
Platanus
CGAL
GAGM
Enly
BESST
Khmer
GRIT
IDBA-MTP
dipSPAdes
WhatsHap
SHEAR
ELOPER
OMACC Omega
GABenchToB
HiPGA
SAGE
HyDA-Vista
MHAP
Mapsembler 2
GAML
SAT-Assembler
RAMPART
VICUNA
There are many, many tools out there for assembling, or helping to assemble, a genome sequence (there are 125 on this page). People may not have the time, patience, or expertise to try more than a handful of these. But…
bambus2
How many assembly tools are out there?
Ray
Celera
MIRA
ALLPATHS-LG
SGACurtain MetassemblerPhusion
ABySS
Amos
Arapan
CLC
Cortex
DNAnexus
DNA Dragon
EdenaForge
GeneiousIDBA
Newbler
PRICE
PADENA
PASHA
Phrap
TIGR
Sequencher
SeqMan NGen
SHARCGS
SOPRA
SSAKE
SPAdes
Taipan
VCAKE
Velvet
Arachne
PCAP
GAM
MonumentAtlas
ABBA
Anchor
ATAC
Contrail
DecGPU GenoMinerLasergene
PE-Assembler
Pipeline Pilot
QSRA
SeqPrep
SHORTY
fermiTelescoper
QuastSCARPA
Hapsembler
HapCompass
HaploMerger
SWiPS
GigAssembler
MSR-CA
MaSuRCA
GARM
Cerulean
TIGRA
ngsShoRT
PERGA
SOAPdenovo
REAPR
FRCBam
EULER-SR SSPACE
Opera
mip
gapfiller
image
PBJelly
HGAP
FALCON
Dazzler
GGAKE
A5
CABOG
SHRAPSR-ASM
SuccinctAssembly
SUTTARagout
Tedna
Trinity
SWAP-Assembler
SILP3
AutoAssemblyD
KGBAssembler
MetAMOS
iMetAMOS
MetaVelvet-SL
KmerGenie
Nesoni
Pilon
Platanus
CGAL
GAGM
Enly
BESST
Khmer
GRIT
IDBA-MTP
dipSPAdes
WhatsHap
SHEAR
ELOPER
OMACC Omega
GABenchToB
HiPGA
SAGE
HyDA-Vista
MHAP
Mapsembler 2
GAML
SAT-Assembler
RAMPART
VICUNA
…people want to know which one is the best!
bambus2
How many assembly tools are out there?
Ray
Celera
MIRA
ALLPATHS-LG
SGACurtain MetassemblerPhusion
ABySS
Amos
Arapan
CLC
Cortex
DNAnexus
DNA Dragon
EdenaForge
GeneiousIDBA
Newbler
PRICE
PADENA
PASHA
Phrap
TIGR
Sequencher
SeqMan NGen
SHARCGS
SOPRA
SSAKE
SPAdes
Taipan
VCAKE
Velvet
Arachne
PCAP
GAM
MonumentAtlas
ABBA
Anchor
ATAC
Contrail
DecGPU GenoMinerLasergene
PE-Assembler
Pipeline Pilot
QSRA
SeqPrep
SHORTY
fermiTelescoper
QuastSCARPA
Hapsembler
HapCompass
HaploMerger
SWiPS
GigAssembler
MSR-CA
MaSuRCA
GARM
Cerulean
TIGRA
ngsShoRT
PERGA
SOAPdenovo
REAPR
FRCBam
EULER-SR SSPACE
Opera
mip
gapfiller
image
PBJelly
HGAP
FALCON
Dazzler
GGAKE
A5
CABOG
SHRAPSR-ASM
SuccinctAssembly
SUTTARagout
Tedna
Trinity
SWAP-Assembler
SILP3
AutoAssemblyD
KGBAssembler
MetAMOS
iMetAMOS
MetaVelvet-SL
KmerGenie
Nesoni
Pilon
Platanus
CGAL
GAGM
Enly
BESST
Khmer
GRIT
IDBA-MTP
dipSPAdes
WhatsHap
SHEAR
ELOPER
OMACC Omega
GABenchToB
HiPGA
SAGE
HyDA-Vista
MHAP
Mapsembler 2
GAML
SAT-Assembler
RAMPART
VICUNA
Which is the best?
…people want to know which one is the best!
At the time I presented this talk, these were six new papers that had recently been published, all of which describe new tools to help make genome assemblies. These papers were all published in about a month of each other. Genome assembly is a hard field to stay on top of!
All published since August 14th, 2014!
At the time I presented this talk, these were six new papers that had recently been published, all of which describe new tools to help make genome assemblies. These papers were all published in about a month of each other. Genome assembly is a hard field to stay on top of!
Comparing assemblers
✤ Can't fairly compare two assemblers if they:
It is not always straightforward to compare two tools if they were used on different species or on different datasets from the same species.
Comparing assemblers
✤ Can't fairly compare two assemblers if they:
✤ produced assemblies from different species
It is not always straightforward to compare two tools if they were used on different species or on different datasets from the same species.
Comparing assemblers
✤ Can't fairly compare two assemblers if they:
✤ produced assemblies from different species
✤ assembled same species, but used sequence data from different sequencing technologies
It is not always straightforward to compare two tools if they were used on different species or on different datasets from the same species.
Comparing assemblers
✤ Can't fairly compare two assemblers if they:
✤ produced assemblies from different species
✤ assembled same species, but used sequence data from different sequencing technologies
✤ used same sequencing technologies but have different sequence libraries
It is not always straightforward to compare two tools if they were used on different species or on different datasets from the same species.
Comparing assemblers
✤ Can't fairly compare two assemblers if they:
✤ produced assemblies from different species
✤ assembled same species, but used sequence data from different sequencing technologies
✤ used same sequencing technologies but have different sequence libraries
✤ Even using different options for the same assembler may produce very different assemblies!
It is not always straightforward to compare two tools if they were used on different species or on different datasets from the same species.
The PRICE genome assembler has 52 command-line options!!!
This assembler has 52 command-line options! Not all of these will affect the resulting assembly, but many of them will.
The PRICE genome assembler has 52 command-line options!!!
how many of them are you going to learn?
This assembler has 52 command-line options! Not all of these will affect the resulting assembly, but many of them will.
A genome assembly competition
That's where the Assemblathon came in.
An attempt to standardize some aspects !of the genome assembly process
Genome assembly contests
Others have been trying to do the same thing. E.g. GAGE, and dnGASP. If you can at least give difference assemblers the same input sequence data, you can start to take account of one of the biggest variables in genome assembly.
✤ 2010–2011!
✤ Used synthetic data!
✤ Small genome (~100 Mbp)!
✤ We knew the answer!
Assemblathon 1
It is easier to judge a tool when you know what the final answer should look like. However, many people that work on developing assemblers would prefer to work with real data…
…which is where Assemblathon 2 came in.
Published in GigaScience,!July 2013
The paper was formally published in the journal GigaScience in mid 2013…
First published !on arXiv.org!
Jan 2013
…but we first published the paper to arXiv.org and ensure that we uploaded updates as the paper changed.
Attracted lots of interest, and provoked lots of commentary
Many blogs commented on the paper.
The Altmetric site, which tracks the social media engagement of academic research, reveals how much interest there has been in the paper.
The Altmetric site, which tracks the social media engagement of academic research, reveals how much interest there has been in the paper.
The open nature by which we conducted the research was recognized with the 2013 BioMed Central award for Open Data. I strongly believe that trying to conduct this science in an open manner ended up making our research much more visible to the scientific community.
But what did the paper reveal?
Type of data Number of genomes
Size of genomes
Do we know the answer?
Assemblathon 1 Synthetic 1 Small ✓
Assemblathon 2 became a much bigger contest compared to Assemblathon 1.
Type of data Number of genomes
Size of genomes
Do we know the answer?
Assemblathon 1 Synthetic 1 Small ✓
Assemblathon 2 Real 3 Large ✗
Assemblathon 2 became a much bigger contest compared to Assemblathon 1.
Melopsittacus undulatus
Boa constrictor constrictorMaylandia zebra
These were the 3 species that were used: a budgie, a cichlid fish from Lake Mawali, and a reptile.
Bird
SnakeFish
Let's simplify the names for the rest of the presentation.
Why these three species?
There is no special reason why these species were used. People had a need to sequence the genomes, and some companies were willing to donate sequences.
Why these three species?
Because they were there
There is no special reason why these species were used. People had a need to sequence the genomes, and some companies were willing to donate sequences.
Species
Bird
Fish
Snake
Estimated genome size
1.2 Gbp
1.0 Gbp
1.6 Gbp
Assemble this!
Lots of sequence data was provided for the bird. Mate pair and read pair libraries were available for all Illumina datasets. !This probably doesn’t reflect a real world scenario. Not everyone can afford over 400x of sequence coverage!
Species
Bird
Fish
Snake
Estimated genome size
1.2 Gbp
1.0 Gbp
1.6 Gbp
Illumina
285x!(14 libraries)
192x!(8 libraries)
125x!(4 libraries)
Assemble this!
Lots of sequence data was provided for the bird. Mate pair and read pair libraries were available for all Illumina datasets. !This probably doesn’t reflect a real world scenario. Not everyone can afford over 400x of sequence coverage!
Species
Bird
Fish
Snake
Estimated genome size
1.2 Gbp
1.0 Gbp
1.6 Gbp
Illumina
285x!(14 libraries)
192x!(8 libraries)
125x!(4 libraries)
Roche 454
16x!(3 libraries)
Assemble this!
Lots of sequence data was provided for the bird. Mate pair and read pair libraries were available for all Illumina datasets. !This probably doesn’t reflect a real world scenario. Not everyone can afford over 400x of sequence coverage!
Species
Bird
Fish
Snake
Estimated genome size
1.2 Gbp
1.0 Gbp
1.6 Gbp
Illumina
285x!(14 libraries)
192x!(8 libraries)
125x!(4 libraries)
Roche 454
16x!(3 libraries)
PacBio
10x!(2 libraries)
Assemble this!
Lots of sequence data was provided for the bird. Mate pair and read pair libraries were available for all Illumina datasets. !This probably doesn’t reflect a real world scenario. Not everyone can afford over 400x of sequence coverage!
Who took part?
Lots of teams took part. Not just from the big sequencing/genome centers.
Who took part?
Lots of teams took part. Not just from the big sequencing/genome centers.
Who took part?
21 teams!43 assemblies!
52,013,623,777 bp of sequence
Lots of teams took part. Not just from the big sequencing/genome centers.
Species
Bird
Fish
Snake
Competitive entries
12
10
12
Entries
There were evaluation entries (not eligible to be declared the winner) allowed in addition to competition entries (only 1 per team).
Species
Bird
Fish
Snake
Competitive entries
12
10
12
Evaluation entries
3
6
0
Entries
There were evaluation entries (not eligible to be declared the winner) allowed in addition to competition entries (only 1 per team).
Goals
Defining quality was really the toughest part of organizing the Assemblathon competitions. Lots of people have lots of (different) ideas as to what contributes to assembly quality.
Goals
✤ Assess 'quality' of assemblies
Defining quality was really the toughest part of organizing the Assemblathon competitions. Lots of people have lots of (different) ideas as to what contributes to assembly quality.
Goals
✤ Assess 'quality' of assemblies
✤ Define quality!
Defining quality was really the toughest part of organizing the Assemblathon competitions. Lots of people have lots of (different) ideas as to what contributes to assembly quality.
Goals
✤ Assess 'quality' of assemblies
✤ Define quality!
✤ Produce ranking of assemblies for each species
Defining quality was really the toughest part of organizing the Assemblathon competitions. Lots of people have lots of (different) ideas as to what contributes to assembly quality.
Goals
✤ Assess 'quality' of assemblies
✤ Define quality!
✤ Produce ranking of assemblies for each species
✤ Produce ranking of assemblers across species?
Defining quality was really the toughest part of organizing the Assemblathon competitions. Lots of people have lots of (different) ideas as to what contributes to assembly quality.
Who did what?
Person/group Jobs
Me, Ian Korf, and Joseph Fass Perform various analyses of all assemblies
David Schwarz et al. Produce & evaluate optical maps
Jay Shendure et al. Produce Fosmid sequences !(bird & snake only)
Martin Hunt & Thomas Otto Performed REAPR analysis
Dent Earl & Benedict Paten Help with meta-analysis of final rankings
Lots of different groups were involved in the organization and assessment of the Assemblathon 2 entries.
91 co-authors!
flickr.com/photos/jamescridland/613445810
Hard to get agreement on how best to interpret the results. Some analyses and interpretations in the Assemblathon 2 paper end up being compromises.
Results!
Lots of results!
A screen grab of my master spreadsheet that contains all of the numerical results. Each row represents a submitted assembly, and each column represents a different assembly metric.
There were a lot of metrics. Many of these were not important or highly informative (e.g. %N).
102 different metrics!
There were a lot of metrics. Many of these were not important or highly informative (e.g. %N).
10 key metrics
We focused on 10 of 102 metrics that we thought were a) useful and b) captured different aspects of an assembly's quality.
Key Metric Description
1 NG50 scaffold length
2 NG50 contig length
3 Amount of assembly in 'gene-sized' scaffolds
4 Number of 'core genes' present
5 Fosmid coverage
6 Fosmid validity
7 Short-range scaffold accuracy
8 Optical map: level 1
9 Optical map: levels 1–3
10 REAPR summary score
The 10 key metrics.
Key Metric Description
1 NG50 scaffold length
2 NG50 contig length
3 Amount of assembly in 'gene-sized' scaffolds
4 Number of 'core genes' present
5 Fosmid coverage
6 Fosmid validity
7 Short-range scaffold accuracy
8 Optical map: level 1
9 Optical map: levels 1–3
10 REAPR summary score
In the remainder of this talk, I’ll just focus on a few of these metrics. See the Assemblathon 2 paper (or older version of this talk) for more details about the other metrics.
1) Scaffold NG50 lengths
✤ Can calculate NG50 length for each assembly!
✤ But also calculate NG60, NG70 etc.!
✤ Plot all results as a graph
An N50 (or NG50) value on its own doesn't tell you that much. Ideally you should always be aware of the total assembly size and the distribution of lengths when comparing assemblies. You can do this by not only calculating NG50, but NG1..NG100. NG1 would be the length of scaffold that captures 1% of the estimated genome size (when summing scaffolds from longest to shortest).
1) Scaffold NG50 lengths
Scaffold length is on a log axis and team identifiers are shown in the legend. !The black dashed line shows the NG50 value, but the point where each series starts on the left shows the lengths of the longest scaffolds. Also, if the NG100 value is greater than zero, then that assembly is bigger than the known/estimated genome size.
2) Contig vs scaffold NG50
We did the same thing for contig NG50 as well as scaffold NG50. The two measure are sometimes, but not always, correlated. The two highlighted data points show outliers for bird assemblies, reflecting assemblies that are good at making long contigs *or* good at making long scaffolds, but not both.
2) Contig vs scaffold NG50
We did the same thing for contig NG50 as well as scaffold NG50. The two measure are sometimes, but not always, correlated. The two highlighted data points show outliers for bird assemblies, reflecting assemblies that are good at making long contigs *or* good at making long scaffolds, but not both.
2) Contig vs scaffold NG50
We did the same thing for contig NG50 as well as scaffold NG50. The two measure are sometimes, but not always, correlated. The two highlighted data points show outliers for bird assemblies, reflecting assemblies that are good at making long contigs *or* good at making long scaffolds, but not both.
3) Gene-sized scaffolds
It is great to have long scaffolds, but maybe for many questions that you might be interested in (e.g. studying codon usage bias), you only need to have scaffolds that have a good chance of capturing a full-length gene.
3) Gene-sized scaffolds
✤ Some assembly folks get a little obsessed by length!
It is great to have long scaffolds, but maybe for many questions that you might be interested in (e.g. studying codon usage bias), you only need to have scaffolds that have a good chance of capturing a full-length gene.
3) Gene-sized scaffolds
✤ Some assembly folks get a little obsessed by length!
✤ How long is 'long enough' for a scaffold?
It is great to have long scaffolds, but maybe for many questions that you might be interested in (e.g. studying codon usage bias), you only need to have scaffolds that have a good chance of capturing a full-length gene.
3) Gene-sized scaffolds
✤ Some assembly folks get a little obsessed by length!
✤ How long is 'long enough' for a scaffold?
✤ What if you just wanted to find genes?
It is great to have long scaffolds, but maybe for many questions that you might be interested in (e.g. studying codon usage bias), you only need to have scaffolds that have a good chance of capturing a full-length gene.
3) Gene-sized scaffolds
✤ Some assembly folks get a little obsessed by length!
✤ How long is 'long enough' for a scaffold?
✤ What if you just wanted to find genes?
✤ Average vertebrate gene = ~25 Kbp
It is great to have long scaffolds, but maybe for many questions that you might be interested in (e.g. studying codon usage bias), you only need to have scaffolds that have a good chance of capturing a full-length gene.
3) Gene-sized scaffolds
The red data series orders the bird assemblies in order of their NG50 scaffold length. The blue line shows the percentage of the estimated genome size that is present in scaffolds of 25 Kbp or longer. Most assemblies, even if they have a much shorter *average* scaffold length, may contain many scaffolds that are still long enough to contain a single gene.
4) Core genes
A previously developed tool (CEGMA) was used to see how many 'core genes' (extremely highly conserved) are present in each assembly. These genes have been identified in six different eukaryotes and are expected to be present in all eukaryotes. !Note that CEGMA finds genes where a full-length (or nearly full-length) gene is present within a single scaffold. Many core genes might be present, but split across scaffolds.
4) Core genes
✤ Used CEGMA (Core Eukaryotic Gene Mapping Approach)
A previously developed tool (CEGMA) was used to see how many 'core genes' (extremely highly conserved) are present in each assembly. These genes have been identified in six different eukaryotes and are expected to be present in all eukaryotes. !Note that CEGMA finds genes where a full-length (or nearly full-length) gene is present within a single scaffold. Many core genes might be present, but split across scaffolds.
4) Core genes
✤ Used CEGMA (Core Eukaryotic Gene Mapping Approach)
✤ CEGMA uses a set of 458 'Core Eukaryotic Genes' (CEGs)
A previously developed tool (CEGMA) was used to see how many 'core genes' (extremely highly conserved) are present in each assembly. These genes have been identified in six different eukaryotes and are expected to be present in all eukaryotes. !Note that CEGMA finds genes where a full-length (or nearly full-length) gene is present within a single scaffold. Many core genes might be present, but split across scaffolds.
4) Core genes
✤ Used CEGMA (Core Eukaryotic Gene Mapping Approach)
✤ CEGMA uses a set of 458 'Core Eukaryotic Genes' (CEGs)
✤ CEGs are conserved in: S. cerevisiae, S. pombe, A. thaliana, C. elegans, D. melanogaster, and H. sapiens
A previously developed tool (CEGMA) was used to see how many 'core genes' (extremely highly conserved) are present in each assembly. These genes have been identified in six different eukaryotes and are expected to be present in all eukaryotes. !Note that CEGMA finds genes where a full-length (or nearly full-length) gene is present within a single scaffold. Many core genes might be present, but split across scaffolds.
4) Core genes
✤ Used CEGMA (Core Eukaryotic Gene Mapping Approach)
✤ CEGMA uses a set of 458 'Core Eukaryotic Genes' (CEGs)
✤ CEGs are conserved in: S. cerevisiae, S. pombe, A. thaliana, C. elegans, D. melanogaster, and H. sapiens
✤ How many full-length CEGs are in each assembly?
A previously developed tool (CEGMA) was used to see how many 'core genes' (extremely highly conserved) are present in each assembly. These genes have been identified in six different eukaryotes and are expected to be present in all eukaryotes. !Note that CEGMA finds genes where a full-length (or nearly full-length) gene is present within a single scaffold. Many core genes might be present, but split across scaffolds.
4) Core genes
Species
Bird
Fish
Snake
Core genes (out of 458)
Best individual assembly
420
436
438
In the three species, most of the core genes were present across all assemblies, but individual assemblies typically lacked several core genes.
4) Core genes
Species
Bird
Fish
Snake
Core genes (out of 458)
Best individual assembly
420
436
438
Across all assemblies
442
455
454
In the three species, most of the core genes were present across all assemblies, but individual assemblies typically lacked several core genes.
4) Core genes
These results show the number of CEGMA genes that were present in any one assembly as a percentage of all possible CEGMA genes (i.e. those present across all assemblies for each species).
What does this all mean?
102 metrics!per assembly
10 key !metrics
1 final!ranking
Using the 10 key metrics, we combined the results to produce a single score for each assembly by which to rank them.
Assembly
CRACS
SYMB
PHUS
BCM
SGA
MERAC
ABYSS
SOAP
RAY
GAM
CURT
Number of !core genes
438
436
435
434
433
430
429
428
422
415
360
Although we did take an average rank from the 10 individual rankings, we preferred to use a Z-score approach. Each assembly was scored based on the total number of standard deviations from the average of each metric. This rewards/penalizes assemblies with very high/low scores in individual metrics. The above results are from the CEGMA metric in bird.
Assembly
CRACS
SYMB
PHUS
BCM
SGA
MERAC
ABYSS
SOAP
RAY
GAM
CURT
Number of !core genes
438
436
435
434
433
430
429
428
422
415
360
Rank
1
2
3
4
5
6
7
8
9
10
11
Although we did take an average rank from the 10 individual rankings, we preferred to use a Z-score approach. Each assembly was scored based on the total number of standard deviations from the average of each metric. This rewards/penalizes assemblies with very high/low scores in individual metrics. The above results are from the CEGMA metric in bird.
Assembly
CRACS
SYMB
PHUS
BCM
SGA
MERAC
ABYSS
SOAP
RAY
GAM
CURT
Number of !core genes
438
436
435
434
433
430
429
428
422
415
360
Rank
1
2
3
4
5
6
7
8
9
10
11
Z-score
+0.68
+0.59
+0.54
+0.49
+0.44
+0.30
+0.25
+0.21
–0.08
–0.41
–3.02
Although we did take an average rank from the 10 individual rankings, we preferred to use a Z-score approach. Each assembly was scored based on the total number of standard deviations from the average of each metric. This rewards/penalizes assemblies with very high/low scores in individual metrics. The above results are from the CEGMA metric in bird.
This graph shows the final rankings of bird assemblies based on their sum Z-scores. Assemblies in red are the evaluation entries. The error bars reflect what would be the highest and lowest sum Z-score if we had used any of the possible combinations of 9 key metrics rather than 10. Note that the highest ranked bird assembly was an evaluation assembly by Baylor College of Medicine (BCM), their competitive entry ranked number 2.
This graph shows the final rankings of bird assemblies based on their sum Z-scores. Assemblies in red are the evaluation entries. The error bars reflect what would be the highest and lowest sum Z-score if we had used any of the possible combinations of 9 key metrics rather than 10. Note that the highest ranked bird assembly was an evaluation assembly by Baylor College of Medicine (BCM), their competitive entry ranked number 2.
This graph shows the final rankings of bird assemblies based on their sum Z-scores. Assemblies in red are the evaluation entries. The error bars reflect what would be the highest and lowest sum Z-score if we had used any of the possible combinations of 9 key metrics rather than 10. Note that the highest ranked bird assembly was an evaluation assembly by Baylor College of Medicine (BCM), their competitive entry ranked number 2.
This graph shows the final rankings of bird assemblies based on their sum Z-scores. Assemblies in red are the evaluation entries. The error bars reflect what would be the highest and lowest sum Z-score if we had used any of the possible combinations of 9 key metrics rather than 10. Note that the highest ranked bird assembly was an evaluation assembly by Baylor College of Medicine (BCM), their competitive entry ranked number 2.
This graph shows the final rankings of bird assemblies based on their sum Z-scores. Assemblies in red are the evaluation entries. The error bars reflect what would be the highest and lowest sum Z-score if we had used any of the possible combinations of 9 key metrics rather than 10. Note that the highest ranked bird assembly was an evaluation assembly by Baylor College of Medicine (BCM), their competitive entry ranked number 2.
In fish, the BCM entry ranked 1st though the error bars suggest there is much variability. The lack of Fosmid data means that there is only 7 key metrics rather than 10.
Snake seemed to the only species where it looked like one assembler outperformed all others (SGA, in this case). We will return to this issue. Note that there were no evaluation entries for snake.
Another way of looking at all of this data is to plot the Z-scores for each metric as a heat map (red = higher Z-scores).
A parallel coordinates plot is another way of trying to show all of the information at once. Although you can try to show all of the results in a single figure, it doesn't always mean that you should. I.e. perhaps not easy to make sense of this.
What does this all mean?
No really, what does this all mean?
Still a bit hard to make sense of the overall rankings. What are the main findings from our paper?
Some conclusions
✤ Very hard to find assemblers that performed well across all 10 key metrics!
✤ Assemblers that perform well in one species, do not always perform as well in another!
✤ Bird & snake assemblies appear better than fish!
✤ No real 'winner' for bird and fish
This type of news is perhaps disappointing to many.
SGA — best assembler for snake?
Even if we had happened to use 9 key metrics rather than 10, and even if we threw out the metric where SGA performed the best, it would still probably rank 1st. So is that the end of the story?
SGA — best assembler for snake?
Even if we had happened to use 9 key metrics rather than 10, and even if we threw out the metric where SGA performed the best, it would still probably rank 1st. So is that the end of the story?
Description Rank of snake SGA assembly
NG50 scaffold length 2
NG50 contig length 5
Amount of assembly in 'gene-sized' scaffolds 7
Number of 'core genes' present 5
Fosmid coverage 2
Fosmid validity 2
Short-range scaffold accuracy 3
Optical map: level 1 2
Optical map: levels 1–3 1
REAPR summary score 2
SGA only ranked 1st in one of the ten key metrics and ranked 7th in another. So it is a good assembler *on average*. But if one of these metrics was highly important to you, you may want to use an assembler that ranked higher in that metric.
Description Rank of snake SGA assembly
NG50 scaffold length 2
NG50 contig length 5
Amount of assembly in 'gene-sized' scaffolds 7
Number of 'core genes' present 5
Fosmid coverage 2
Fosmid validity 2
Short-range scaffold accuracy 3
Optical map: level 1 2
Optical map: levels 1–3 1
REAPR summary score 2
SGA only ranked 1st in one of the ten key metrics and ranked 7th in another. So it is a good assembler *on average*. But if one of these metrics was highly important to you, you may want to use an assembler that ranked higher in that metric.
We found it interesting that the best bird assembly was the evaluation entry by Baylor College of Medicine. What is different about this entry compared to their competitive entry?
We found it interesting that the best bird assembly was the evaluation entry by Baylor College of Medicine. What is different about this entry compared to their competitive entry?
Assembler
BCM - evaluation
BCM - competitive
Final rank
1
2
NGS data used in
assembly
Illumina + 454
Illumina + 454 + PacBio
BCM bird assemblies
The only difference is that the BCM competitive entry included PacBio data, and somehow this led to the paradoxical situation where including more sequence in the assembly produced a lower measures for coverage and validity (from the Fosmids), though one key metric (NG50 contig length) did improve.
Assembler
BCM - evaluation
BCM - competitive
Final rank
1
2
NGS data used in
assembly
Illumina + 454
Illumina + 454 + PacBio
BCM bird assemblies
The only difference is that the BCM competitive entry included PacBio data, and somehow this led to the paradoxical situation where including more sequence in the assembly produced a lower measures for coverage and validity (from the Fosmids), though one key metric (NG50 contig length) did improve.
Assembler
BCM - evaluation
BCM - competitive
Final rank
1
2
NGS data used in
assembly
Illumina + 454
Illumina + 454 + PacBio
Coverage!Z-score
+2.0
–0.3
BCM bird assemblies
The only difference is that the BCM competitive entry included PacBio data, and somehow this led to the paradoxical situation where including more sequence in the assembly produced a lower measures for coverage and validity (from the Fosmids), though one key metric (NG50 contig length) did improve.
Assembler
BCM - evaluation
BCM - competitive
Final rank
1
2
NGS data used in
assembly
Illumina + 454
Illumina + 454 + PacBio
Coverage!Z-score
+2.0
–0.3
Validity!Z-score
+1.4
–0.8
BCM bird assemblies
The only difference is that the BCM competitive entry included PacBio data, and somehow this led to the paradoxical situation where including more sequence in the assembly produced a lower measures for coverage and validity (from the Fosmids), though one key metric (NG50 contig length) did improve.
Assembler
BCM - evaluation
BCM - competitive
Final rank
1
2
NGS data used in
assembly
Illumina + 454
Illumina + 454 + PacBio
Coverage!Z-score
+2.0
–0.3
Validity!Z-score
+1.4
–0.8
NG50 Contig Z-score
+1.5
+2.7
BCM bird assemblies
The only difference is that the BCM competitive entry included PacBio data, and somehow this led to the paradoxical situation where including more sequence in the assembly produced a lower measures for coverage and validity (from the Fosmids), though one key metric (NG50 contig length) did improve.
BCM evaluation scaffold
NNNNNNNNNNNNNNNNNNN
BCM used PacBio data in a targeted way…the PacBio reads were used to help fill in the gaps in their scaffolds.
BCM evaluation scaffold
NNNNNNNNNNNNNNNNNNN
BCM competition scaffold
NNNNNNNNNNNNNNNNNNN
BCM used PacBio data in a targeted way…the PacBio reads were used to help fill in the gaps in their scaffolds.
BCM evaluation scaffold
NNNNNNNNNNNNNNNNNNN
BCM competition scaffold
NNNNNNNNNNNNNNNNNNN
PacBio sequence
BCM used PacBio data in a targeted way…the PacBio reads were used to help fill in the gaps in their scaffolds.
BCM evaluation scaffold
NNNNNNNNNNNNNNNNNNN
BCM competition scaffold
CGTCGNNATCNNGGTTACG
Errors in the PacBio sequence were penalized by the choice of alignment program used to align Fosmid sequences to scaffolds.
BCM evaluation scaffold
NNNNNNNNNNNNNNNNNNN
BCM competition scaffold
CGTCGNNATCNNGGTTACG
Mismatches from PacBio sequence penalized alignment !score more than matching unknown bases
Errors in the PacBio sequence were penalized by the choice of alignment program used to align Fosmid sequences to scaffolds.
The choice of one command-line option,!used by one tool in the calculation of one key metric...
...probably made enough difference to drop!the PacBio-containing assembly to 2nd place.
This was actually down to the use of a single command-line option in the lastz alignment program. If we had not chosen this option, the PacBio-containing entry would have probably ranked 1st among all bird assemblies.
Other conclusions
✤ Different metrics tell different stories!
✤ Heterozygosity was a big issue for bird & fish assemblies!
✤ Final rankings very sensitive to changes in metrics!
✤ N50 is a semi-useful predictor of assembly quality
The last point may disappoint some. Despite looking at many different metrics, N50 scaffold length still does a reasonable job of predicting overall quality. However...
...the outliers in this relationship should be noted. The highlighted bird assembly had the second highest scaffold N50 length, but ranked 6th among bird assemblies.
...the outliers in this relationship should be noted. The highlighted bird assembly had the second highest scaffold N50 length, but ranked 6th among bird assemblies.
Inter-specific differences matter
Biological differences may account for differences in assembler performance between different species. However, the input data for each species was also very different and this may play a role as well (some assemblers perform prefer certain short-insert sizes).
Inter-specific differences matter
✤ The three species have genomes with different properties !
✤ repeats!
✤ heterozygosity
Biological differences may account for differences in assembler performance between different species. However, the input data for each species was also very different and this may play a role as well (some assemblers perform prefer certain short-insert sizes).
Inter-specific differences matter
✤ The three species have genomes with different properties !
✤ repeats!
✤ heterozygosity
✤ The three genomes had very different NGS data sets!
✤ Only bird had PacBio & 454 data!
✤ Different insert sizes in short-insert libraries
Biological differences may account for differences in assembler performance between different species. However, the input data for each species was also very different and this may play a role as well (some assemblers perform prefer certain short-insert sizes).
The Big Conclusion
People would like an assembler that consistently performs well across most (all?) metrics and across most species. We didn’t find such an assembler in the Assemblathon 2 contest.
The Big Conclusion
"You can't always get what you want"Sir Michael Jagger, 1969
People would like an assembler that consistently performs well across most (all?) metrics and across most species. We didn’t find such an assembler in the Assemblathon 2 contest.
What comes next?
What comes next?
There may one day be an Assemblathon 3 but there are no immediate plans (and no funding for us at UC Davis to do so).
What comes next?
3?
There may one day be an Assemblathon 3 but there are no immediate plans (and no funding for us at UC Davis to do so).
A wish list for Assemblathon 3
If there is to be an Assemblathon 3, here are some things that we have learned from Assemblathon 2.
A wish list for Assemblathon 3
✤ Only have 1 species
If there is to be an Assemblathon 3, here are some things that we have learned from Assemblathon 2.
A wish list for Assemblathon 3
✤ Only have 1 species
✤ Teams have to 'buy' resources using virtual budgets
If there is to be an Assemblathon 3, here are some things that we have learned from Assemblathon 2.
A wish list for Assemblathon 3
✤ Only have 1 species
✤ Teams have to 'buy' resources using virtual budgets
✤ Factor in CPU time/cost?
If there is to be an Assemblathon 3, here are some things that we have learned from Assemblathon 2.
A wish list for Assemblathon 3
✤ Only have 1 species
✤ Teams have to 'buy' resources using virtual budgets
✤ Factor in CPU time/cost?
✤ Agree on metrics before evaluating assemblies!
If there is to be an Assemblathon 3, here are some things that we have learned from Assemblathon 2.
A wish list for Assemblathon 3
✤ Only have 1 species
✤ Teams have to 'buy' resources using virtual budgets
✤ Factor in CPU time/cost?
✤ Agree on metrics before evaluating assemblies!
✤ Encourage experimental assemblies
If there is to be an Assemblathon 3, here are some things that we have learned from Assemblathon 2.
A wish list for Assemblathon 3
✤ Only have 1 species
✤ Teams have to 'buy' resources using virtual budgets
✤ Factor in CPU time/cost?
✤ Agree on metrics before evaluating assemblies!
✤ Encourage experimental assemblies
✤ Use FASTG or GFA genome assembly file format?
If there is to be an Assemblathon 3, here are some things that we have learned from Assemblathon 2.
A wish list for Assemblathon 3
✤ Only have 1 species
✤ Teams have to 'buy' resources using virtual budgets
✤ Factor in CPU time/cost?
✤ Agree on metrics before evaluating assemblies!
✤ Encourage experimental assemblies
✤ Use FASTG or GFA genome assembly file format?
✤ Get someone else to write the paper!
If there is to be an Assemblathon 3, here are some things that we have learned from Assemblathon 2.
But maybe we don't need an Assemblathon 3?
nucleotid.es is a new website that aims to a) provide a catalog of modern genome assemblers and b) evaluate their performance using some standardized sets of input read data.
It uses Docker containers to help make the software easy to run for others. And people are encouraged to upload 'dockerized' version of their assemblers.
The website also allows benchmarks for different versions of the same assembler, e.g. either using different parameter options, or different pre- and post-assembly filtering steps.
The website also allows benchmarks for different versions of the same assembler, e.g. either using different parameter options, or different pre- and post-assembly filtering steps.
These are some of the current 'winners' on the nucleotid.es site. Hopefully, more people will start using this site and maybe we won't ever need to have a dedicated Assemblathon 3 contest.
Intermission
And now a break in the scheduled program in order to let me vent a little steam.
NGS must die!
Next-generation sequencing (NGS) is heavily used as a convenient label for modern sequencing technologies. But those technologies have — in some cases — be in development since the mid 1990s. Do we refer to everything from the last 20 years as ‘next’ generation?
NGS must die!
‘NGS’ is used to refer to everything post-Sanger
Next-generation sequencing (NGS) is heavily used as a convenient label for modern sequencing technologies. But those technologies have — in some cases — be in development since the mid 1990s. Do we refer to everything from the last 20 years as ‘next’ generation?
NGS must die!
‘NGS’ is used to refer to everything post-Sanger
Pyrosequencing was developed ~1996
Next-generation sequencing (NGS) is heavily used as a convenient label for modern sequencing technologies. But those technologies have — in some cases — be in development since the mid 1990s. Do we refer to everything from the last 20 years as ‘next’ generation?
There are over 5,000 papers in Google Scholar which feature ‘Next-generation sequencing’ or ‘NGS’ in the title of the article. These do not help you if were trying to find papers that focus on pyrosequencing or nanopore sequencing. How could we improve these titles?
In many cases, including ‘next-generation’ adds nothing to the description of the paper. Here are the same paper titles with the words 'next-generation' removed.
NGS madness
Next generation sequencing
aka second generation sequencing
Some people have tried alternative names. These are all descriptions that have been used in published papers.
NGS madness
Next generation sequencing
aka second generation sequencing
but there’s also:
Some people have tried alternative names. These are all descriptions that have been used in published papers.
NGS madness
Next generation sequencing
aka second generation sequencing
but there’s also: third generation sequencing
Some people have tried alternative names. These are all descriptions that have been used in published papers.
NGS madness
Next generation sequencing
aka second generation sequencing
but there’s also: third generation sequencing
fourth generation sequencing
Some people have tried alternative names. These are all descriptions that have been used in published papers.
NGS madness
Next generation sequencing
aka second generation sequencing
but there’s also: third generation sequencing
fourth generation sequencing
next-next generation sequencing
Some people have tried alternative names. These are all descriptions that have been used in published papers.
NGS madness
Next generation sequencing
aka second generation sequencing
but there’s also: third generation sequencing
fourth generation sequencing
next-next generation sequencing
next-next-next generation sequencing
Some people have tried alternative names. These are all descriptions that have been used in published papers.
NGS madness
Technology
Complete Genomics
Ion Torrent
PacBio
Oxford Nanopore
According to some papers…
2nd generation
2nd generation
2nd generation
3rd generation
And of course, not everyone agrees on what is 2nd, 3rd, or 4th generation!
NGS madness
Technology
Complete Genomics
Ion Torrent
PacBio
Oxford Nanopore
According to some papers…
2nd generation
2nd generation
2nd generation
3rd generation
According to other papers…
3rd generation
3rd generation
3rd generation
4th generation
And of course, not everyone agrees on what is 2nd, 3rd, or 4th generation!
NGS madness
“PacBio is a 2.5th generation”
“Helicos lies between the transition of next-generation to third generation”
And of course, someone also has to be different!
NGS madness
There are different sequencing methodologies, !and there are different sequencing platforms.
I would suggest that it is more helpful to refer to different sequencing technologies by their methodology (sequencing by synthesis, pyrosequencing, nanopore sequencing etc), or by the company developing the product (PacBio, Illumina etc.).
NGS madness
There are different sequencing methodologies, !and there are different sequencing platforms.
Use one or the other.
I would suggest that it is more helpful to refer to different sequencing technologies by their methodology (sequencing by synthesis, pyrosequencing, nanopore sequencing etc), or by the company developing the product (PacBio, Illumina etc.).
NGS madness
There are different sequencing methodologies, !and there are different sequencing platforms.
Use one or the other.
Or just say ‘current sequencing technologies’.
I would suggest that it is more helpful to refer to different sequencing technologies by their methodology (sequencing by synthesis, pyrosequencing, nanopore sequencing etc), or by the company developing the product (PacBio, Illumina etc.).
Intermission
And now back to our scheduled programming.
My #1 piece!of advice
flickr.com/julia_manzerova
If you ever have to work with genome assemblies, here is my top piece of advice.
flickr.com/thomashawk
Look at your *input* data (what goes into the assembler) and *output* data (what comes out of the assembler). And really look at it (in a Unix terminal).
flickr.com/thomashawk
Look at your data!
Look at your *input* data (what goes into the assembler) and *output* data (what comes out of the assembler). And really look at it (in a Unix terminal).
I am frequently asked to run CEGMA for people (to assess the completeness of their genome assembly). I track the CEGMA results (using a narrower set of 248 of the most conserved core genes) and also record the N50 scaffold length. Even with just two metrics, there is a lot of variation.
I am frequently asked to run CEGMA for people (to assess the completeness of their genome assembly). I track the CEGMA results (using a narrower set of 248 of the most conserved core genes) and also record the N50 scaffold length. Even with just two metrics, there is a lot of variation.
I am frequently asked to run CEGMA for people (to assess the completeness of their genome assembly). I track the CEGMA results (using a narrower set of 248 of the most conserved core genes) and also record the N50 scaffold length. Even with just two metrics, there is a lot of variation.
From a vertebrate genome assembly with 72,214 sequences…
In one particular assembly, nearly all of the sequence was represented by incredibly short scaffolds. The shortest sequence in the assembly was 3 bp. Assemblies like this are not likely to be useful for anything. Unsurprisingly, this assembly didn’t contain any core genes.
From a vertebrate genome assembly with 72,214 sequences…
In one particular assembly, nearly all of the sequence was represented by incredibly short scaffolds. The shortest sequence in the assembly was 3 bp. Assemblies like this are not likely to be useful for anything. Unsurprisingly, this assembly didn’t contain any core genes.
From a vertebrate genome assembly with 72,214 sequences…
In one particular assembly, nearly all of the sequence was represented by incredibly short scaffolds. The shortest sequence in the assembly was 3 bp. Assemblies like this are not likely to be useful for anything. Unsurprisingly, this assembly didn’t contain any core genes.
From a vertebrate genome assembly with 72,214 sequences…
In one particular assembly, nearly all of the sequence was represented by incredibly short scaffolds. The shortest sequence in the assembly was 3 bp. Assemblies like this are not likely to be useful for anything. Unsurprisingly, this assembly didn’t contain any core genes.
From a vertebrate genome assembly with 72,214 sequences…
In one particular assembly, nearly all of the sequence was represented by incredibly short scaffolds. The shortest sequence in the assembly was 3 bp. Assemblies like this are not likely to be useful for anything. Unsurprisingly, this assembly didn’t contain any core genes.
From a vertebrate genome assembly with 72,214 sequences…
Length of 10 shortest sequences: !100, 100, 99, 88, 87, 76, 73, 63, 12, and 3 bp!
In one particular assembly, nearly all of the sequence was represented by incredibly short scaffolds. The shortest sequence in the assembly was 3 bp. Assemblies like this are not likely to be useful for anything. Unsurprisingly, this assembly didn’t contain any core genes.
For some of the CEGMA runs that I have made, I’ve noted which assemblers was used…
These results show that any assembler can be used to make a bad genome assembly. There is no one assembler which consistently performs well (as assessed by these two metrics). Note that these assemblies were generated from many different species.
Reasons to be cheerful
flickr.com/danielygo
After sounding quite pessimistic so far, here are some more positive reasons why genome assembly might be getting better.
Improvements in sequencing technology !will lead to improvements in genome assembly
Data from Lex Nederbragt’s blog, June 2014
Sequencing technologies continue to improve. 10,000 bp is sort of a ‘breakthrough’ length that would greatly assist genome assembly. Producing many reads that are >10,000 bp means that you can sequence all the way through most eukaryotic repeats (which are one of the two major scourges for genome assemblers).
Data from Lex Nederbragt’s blog, June 2014
Sequencing technologies continue to improve. 10,000 bp is sort of a ‘breakthrough’ length that would greatly assist genome assembly. Producing many reads that are >10,000 bp means that you can sequence all the way through most eukaryotic repeats (which are one of the two major scourges for genome assemblers).
Long-read technology
Moleculo read data from Illumina BaseSpace, July 2013
Moleculo (now owned by Illumina) can take Illumina reads and somehow (not sure anyone knows the science behind how it works) combine them to make much longer reads.
Long-read technology
From https://flxlexblog.wordpress.com (Lex Nederbragt's blog)
PacBio!data
Library preparation is a hugely important part of the genome assembly process. The Blue Pippin library prep greatly improves the number of super long PacBio reads.
Long-read technology
MinIon from Oxford Nanopore
Oxford Nanopore burst on to the scene and excited everyone. But it has been a wait before people had the chance to use their MinION devices for themselves. The UC Davis Genome Center recently received 3 MinIONs as part of the early access program.
Long-read technology
MinIon from Oxford Nanopore
Oxford Nanopore burst on to the scene and excited everyone. But it has been a wait before people had the chance to use their MinION devices for themselves. The UC Davis Genome Center recently received 3 MinIONs as part of the early access program.
Where is the data?
Nick Loman was the first person to publish a ‘real world’ read from these devices.
Where is the data?
Nick Loman was the first person to publish a ‘real world’ read from these devices.
Where is the data?
Nick Loman published the first real-world data on June 10th
Nick Loman was the first person to publish a ‘real world’ read from these devices.
He also shared some of the statistics from his entire run. This nanopore sequencing technology seems limited by how large your DNA fragments are. It may be possible to generated much longer reads.
An E. coli dataset was released to the GigaDB database (http://gigadb.org)
Nick also released the first MinION dataset on September 10th
An E. coli dataset was released to the GigaDB database (http://gigadb.org)
Although Illumina holds such a strong position in the world of sequencing, other companies continue to work on new sequencing technologies.
Base4 are developing a 'microdroplet sequencing' approach. All new technologies seem keen to target the world of 'single molecule' sequencing, with very long reads, and real-time (or 'near' real-time in this case) results.
PicoSeq have developed a technology called SIMDEQ (Single-molecule Magnetic Detection and Quantification) which has the potential to be used to generate long-reads from single molecules. !Maybe companies like Base4 and PicoSeq will never usurp Illumina, but it is good to see people trying to develop new technologies. Competition will help drive down the price of sequencing.
Some other ways to tackle the problems !inherent in genome assembly
Single chromosome assembly?
Breaking the problem up into smaller chunks may be one other way of tackling the genome assembly problem (though many single chromosomes in eukaryotes are still very long).
Tackling heterozygosity
1000 Genomes project plans to sequence 15 'trios' in high-depth
The second major problem for genome assemblers is that of heterozygosity that is present in most (diploid) genomes. The 1,000 Genomes project is trying to tackle this by sequencing ‘trios’, an individual plus their parents and will try to use the combination of datasets to resolve the heterozygosity.
Hi-C
✤ Nature Biotechnology, 31, 2013 !
✤ Burton et al.!
✤ Selvaraj et al.!
✤ Kaplan & Dekker
Hi-C is another new technology that might be able to improve the scaffolding step of genome assembly.
The future of genome assembly
Maybe one day, genome assembly will be as simple as downloading a sequence to your iPhone and clicking ‘assemble’. That day is still some time away.
Kwik-E-Assembler
acgtaacacaancac gggaacnnnacatta acnactagcataata nnnnnnnnnnaacac actttaaattatatc
The future of genome assembly
Maybe one day, genome assembly will be as simple as downloading a sequence to your iPhone and clicking ‘assemble’. That day is still some time away.
The future of genome assembly
Currently a lot of effort is spent generating huge datasets in order to produce a final genome assembly. There are hundreds of genome assemblies out there which are very poor and incomplete. In many cases we don’t know just how good or bad they are. The pace of change in this field means it will often be easier to simply resequence and reassemble a genome rather than attempt to work with a previous genome assembly. !Even if assembly improves, there will be lots of data to manage in future. And people will have their genomes sequenced at different time points throughout their lives (part of your ‘genome checkup’ at the doctors?).
The future of genome assembly
✤ At some point we will look back with embarrassment at this era.
Currently a lot of effort is spent generating huge datasets in order to produce a final genome assembly. There are hundreds of genome assemblies out there which are very poor and incomplete. In many cases we don’t know just how good or bad they are. The pace of change in this field means it will often be easier to simply resequence and reassemble a genome rather than attempt to work with a previous genome assembly. !Even if assembly improves, there will be lots of data to manage in future. And people will have their genomes sequenced at different time points throughout their lives (part of your ‘genome checkup’ at the doctors?).
The future of genome assembly
✤ At some point we will look back with embarrassment at this era.
✤ Assembly must, and will, get better, but...
Currently a lot of effort is spent generating huge datasets in order to produce a final genome assembly. There are hundreds of genome assemblies out there which are very poor and incomplete. In many cases we don’t know just how good or bad they are. The pace of change in this field means it will often be easier to simply resequence and reassemble a genome rather than attempt to work with a previous genome assembly. !Even if assembly improves, there will be lots of data to manage in future. And people will have their genomes sequenced at different time points throughout their lives (part of your ‘genome checkup’ at the doctors?).
The future of genome assembly
✤ At some point we will look back with embarrassment at this era.
✤ Assembly must, and will, get better, but...
✤ ...'perfect' genomes may remain elusive.
Currently a lot of effort is spent generating huge datasets in order to produce a final genome assembly. There are hundreds of genome assemblies out there which are very poor and incomplete. In many cases we don’t know just how good or bad they are. The pace of change in this field means it will often be easier to simply resequence and reassemble a genome rather than attempt to work with a previous genome assembly. !Even if assembly improves, there will be lots of data to manage in future. And people will have their genomes sequenced at different time points throughout their lives (part of your ‘genome checkup’ at the doctors?).
The future of genome assembly
✤ At some point we will look back with embarrassment at this era.
✤ Assembly must, and will, get better, but...
✤ ...'perfect' genomes may remain elusive.
✤ Data management will remain an issue:
Currently a lot of effort is spent generating huge datasets in order to produce a final genome assembly. There are hundreds of genome assemblies out there which are very poor and incomplete. In many cases we don’t know just how good or bad they are. The pace of change in this field means it will often be easier to simply resequence and reassemble a genome rather than attempt to work with a previous genome assembly. !Even if assembly improves, there will be lots of data to manage in future. And people will have their genomes sequenced at different time points throughout their lives (part of your ‘genome checkup’ at the doctors?).
The future of genome assembly
✤ At some point we will look back with embarrassment at this era.
✤ Assembly must, and will, get better, but...
✤ ...'perfect' genomes may remain elusive.
✤ Data management will remain an issue:
✤ the human genome -> human genomes -> tissue-specific genomes
Currently a lot of effort is spent generating huge datasets in order to produce a final genome assembly. There are hundreds of genome assemblies out there which are very poor and incomplete. In many cases we don’t know just how good or bad they are. The pace of change in this field means it will often be easier to simply resequence and reassemble a genome rather than attempt to work with a previous genome assembly. !Even if assembly improves, there will be lots of data to manage in future. And people will have their genomes sequenced at different time points throughout their lives (part of your ‘genome checkup’ at the doctors?).
Summary
The last point on this slide is something that I repeat every 5 years!
Summary
✤ There is no real consensus on how to make a good genome assembly
The last point on this slide is something that I repeat every 5 years!
Summary
✤ There is no real consensus on how to make a good genome assembly
✤ Try different assemblers, try different command-line options
The last point on this slide is something that I repeat every 5 years!
Summary
✤ There is no real consensus on how to make a good genome assembly
✤ Try different assemblers, try different command-line options
✤ Decide what it is you want to get out of a genome assembly
The last point on this slide is something that I repeat every 5 years!
Summary
✤ There is no real consensus on how to make a good genome assembly
✤ Try different assemblers, try different command-line options
✤ Decide what it is you want to get out of a genome assembly
✤ Look at your input and output data
The last point on this slide is something that I repeat every 5 years!
Summary
✤ There is no real consensus on how to make a good genome assembly
✤ Try different assemblers, try different command-line options
✤ Decide what it is you want to get out of a genome assembly
✤ Look at your input and output data
✤ Wait 5 years and come back, we’ll (probably) have solved everything!
The last point on this slide is something that I repeat every 5 years!
Useful blogs/tweeps to follow
Lex Nederbragt!@lexnederbragt!
flxlexblog.wordpress.com
Nick Loan!@pathogenomenick!
pathogenomic.bham.ac.uk/blog
Mick Watson!@BioMickWatson!
biomickwatson.wordpress.com
These people use their blogs to write about latest and greatest news in the worlds of sequencing and genome assembly. Their twitter accounts are also worth following.
Thank you for listening!
@kbradnam @assemblathon
My blog: http://acgt.me
And here are some of the ways to follow what I do. My ACGT blog is a source for many of my frustrations about the world of genomics and bioinformatics :-)
Any questions???