De Bruijn Graph Assembly

Contributors

Authors:

Simon Gladman

Questions

What are the factors that affect genome assembly?
How does Genome assembly work?

Objectives

Perform an optimised Velvet assembly with the Velvet Optimiser
Compare this assembly with those we did in the basic tutorial
Perform an assembly using the SPAdes assembler.

last_modification Published: May 24, 2017

last_modification Last Updated: Feb 7, 2022

.enlarge120[

De novo Genome Assembly

Part 2: De Bruijn Graph Assembly

]

With thanks to T Seemann, D Bulach, I Cooke and Simon Gladman

.enlarge120[

de Bruijn Graphs

] .pull-left[

Named after Nicolaas Govert de Bruijn
Directed graph representing overlaps between sequences of symbols
Sequences can be reconstructed by moving between nodes in graph ] .pull-right[ .image-50[] ] — .enlarge120[# de Bruijn Graphs
A directed graph of sequences of symbols
Nodes in the graph are k-mers
Edges represent consecutive k-mers (which overlap by k-1 symbols) ] Consider the 2 symbol alphabet (0 & 1) de Bruijn Graph for k =3 — .enlarge120[# Producing sequences
Sequences of symbols are produced by moving through the graph ]

.enlarge120[

K-mers? .pull-right[.image-25[]]

To be able to use de Bruijn graphs, we need reads of length L to overlap by L-1 bases.
Not all reads will overlap another read perfectly.
- Read errors
- Coverage “holes”
Not all reads are the same length (depending on technology and quality cleanup)

To help us get around these problems, we use all k-length subsequences of the reads, these are the k-mers. ] — .enlarge120[

What are K-mers? .pull-right[.image-25[]]]

k=12 kmers show three sequences with length 12 overlapping. For k=5 10 different shorter sequences are shown.

.enlarge120[

K-mers de Bruijn graph .pull-right[.image-25[]]]

An example of happiness is given with 4 fragments, happi, pine, iness, appin. — .enlarge120[

K-mers de Bruijn graph .pull-right[.image-25[]]]

all 4-mers are shown from those 5-mers, so happi becomes happ and appi. These are then uniqued. — .enlarge120[

K-mers de Bruijn graph .pull-right[.image-25[]]]

A graph is drawn based on overlaps between those 4-mers. — .enlarge120[

K-mers de Bruijn graph .pull-right[.image-25[]]]

The graph is collapsed into the word happiness, success.

.enlarge120[

The problem of repeats .pull-right[.image-25[]]]

A second example with the word mississippi is shown. The fragments are now missis, ssissi, ssippi. — .enlarge120[

The problem of repeats .pull-right[.image-25[]]]

All 4-mers are generated so missis becomes miss, issi, and ssis. These are then uniqued to get 7 unique 4-mers. — .enlarge120[

The problem of repeats .pull-right[.image-25[]]]

A graph is generated but this graph is more complex and has a loop. — .enlarge120[

The problem of repeats .pull-right[.image-25[]]]

This graph is walked and now the results are shown, mississippi, mississississippi, or more could be found by following the loop repeatedly. — .enlarge120[

Different k .pull-right[.image-25[]]]

The same mississippi example from the start. — .enlarge120[

Different k .pull-right[.image-25[]]]

Instead of 4-mers, we now produce 5-mers. — .enlarge120[

Different k .pull-right[.image-25[]]]

The graph is produced, and now there are two disconnected sets of nodes. — .enlarge120[

Different k .pull-right[.image-25[]]]

When collapsing this graph we see two results, MISSISSIS and SSIPPI.

.enlarge120[

Choose k wisely .pull-right[.image-25[]]]

.enlarge120[

Lower k
- More connections
- Less chance of resolving small repeats
- Higher k-mer coverage
Higher k
- Less connections
- More chance of resolving small repeats
- Lower k-mer coverage

Optimum value for k will balance these effects.

]

.enlarge120[

Read errors .pull-right[.image-25[]]]

.image-75[ Example 3, happiness again, with fragments happi, iness, aplin, pinet ] — .enlarge120[

Read errors .pull-right[.image-25[]]]

.image-75[ Happiness example 5-mers ] all 3-mers are produced — .enlarge120[

Read errors .pull-right[.image-25[]]]

.image-75[ Happiness example 5-mers ] The three mers are collapsed into an overlap graph — .enlarge120[

Read errors .pull-right[.image-25[]]]

.image-75[ Happiness example 5-mers ] The graph is coloured and six contigs are produced based on overlaps.

.enlarge120[

More coverage .pull-right[.image-50[]]]

.enlarge120[

Errors won’t be duplicated in every read
Most reads will be error free
We can count the frequency of each k-mer
Annotate the graph with the frequencies
Use the frequency data to clean the de Bruijn graph

More coverage depth will help overcome errors! ] — .enlarge120[

Read errors revisited .pull-right[.image-50[]]]

Same happiness example, all three mers are present, but now the counts are shown. One path has many more counts than the other. Text asks which path looks most valid? why?

.enlarge120[

Another parameter - coverage cutoff]

.enlarge120[

At what point is a low coverage indicative of an error?
Can we ignore low coverage nodes and paths?
This is a new assembly parameter

Coverage cutoff ]

.enlarge120[

de Bruijn graph assembly process]

.enlarge120[

Select a value for k
“Hash” the reads (make the kmers)
Count the kmers
Make the de Bruijn graph
Perform graph simplification steps - use cov cutoff
Read off contigs from simplified graph

]

.enlarge120[

Graph simplification]

Step 1: Chain merging

Looking at a doubly connected graph with forward/reverse sequences and piles of overlaps. Nodes are connected with lines. One node is labelled as already merged, and two nodes sharing a prefix are labelled as further merging possible.

.enlarge120[

Graph simplification]

Step 2: Tip clipping

a graph with every node doubled for forward/reverse strands is shown, clip tips if the length of the tip of the node is less than 2 times k.

.enlarge120[

Graph simplification]

.pull-left[ Three graphs are shown, B, C, and D. B is the most complex, C combines two nodes B and B prime, while D collapses even more nodes to further simplify the graph. ] .pull-right[

Step 3: Bubble collapsing

Detect redundant paths through graph
Compare the paths using sequence alignment
If similar, merge the paths

.reduce70[Image: Zerbino & Birney 2008] ]

.enlarge120[

Graph simplification

Step 4: Remove low coverage nodes

Remove erroneous nodes and connections using the “coverage cutoff”
Genuine short nodes will have a high coverage

]

.enlarge120[

Make contigs

Find an unbalanced node in the graph
Follow the chain of nodes and “read off” the bases to produce the contigs
Where there is an ambiguous divergence/convergence, stop the current contig and start a new one.
Re-trace the reads through the contigs to help with repeat resolution

]

.enlarge120[

Velvet

Velvet has two separate programs:

Velveth
- Makes the k-mers and
- Efficiently counts (hashes) them
- All in O(N) time
Velvetg
- Makes the graph - O(U) time. U = unique k-mers.
- Simplifies it
- Makes contigs - O(E) time. E = edges in graph

But: You need to choose k and c wisely!

]

.enlarge120[

Velvet - Paired end scaffolding

Breadcrumb algorithm ]

Two boxes A and B are at opposite ends, there are many nodes along a line between them, and some sitting outside of the graph or otherwise weirdly connected.

.enlarge120[

Extensions of the idea

] — .enlarge120[

SPAdes .pull-right[.image-50[]]]

.enlarge120[

de Bruijn graph assembler by Pavel Pevzner’s group out of St. Petersburg
Uses multiple k-mers to build the graph
- Graph has connectivity and specificity
- Usually use a low, medium and high k-mer size together.
Performs error correction on the reads first
Maps reads back to the contigs and scaffolds as a check
Under active development
Much slower than Velvet
Should be used in preference to Velvet now.

]

.enlarge120[

A move back to OLC]

.pull-left[ .enlarge120[

New long read technologies
- PacBio and MinIon
Assemblers: HGap, CANU
- Use overlap, layout consensus approach
CANU can perform hybrid assemblies with long and short reads ] ]

.pull-right[ image of a minion connected to a computer via a cable. Below a man stands next to a very large machine. ] — .enlarge120[# Bandage

Assembly graph viewer and manipulator
Written by Ryan Wick of Centre for Systems Genomics - Uni. Melbourne, Australia ]

Key Points

We learned about how the choice of k-mer size will affect assembly outcomes
We learned about the strategies that assemblers use to make reference genomes
We performed a number of assemblies with Velvet and SPAdes.
You should use SPAdes or another more modern assembler than Velvet for actual assemblies now.

Thank you!

This material is the result of a collaborative work. Thanks to the Galaxy Training Network and all the contributors!

Tutorial Content is licensed under Creative Commons Attribution 4.0 International License.