E. coli 36 bp

Escherichia coli K12 MG1655. The E. coli MG1655 consists of a circular chromosome of 4,639,675 bp in length.

Read source

The illuminia read data of E. coli (Paired-end 36 bp sequencing library with 200 bp inserts) were downloaded from Sequence Read Archive (SRA). More than 20.8 M reads

Sequence assembly

• Set1 (Different Assemblers)

Merged File: Set1_Contig

The above assemblers together with the parameter setting have been selected for de novo assembling of E. coli. After assembly, we discarded contigs with less than 100bp and evaluated the accuracy of the assemblies based on the Mauve Assembly Matrices (the results are shown below). In this set of data, we have different sequence assemblies, each was generated by the different assembler. Because the same assembler performs differently over varying parameter settings, such as kmer, we have tried different parameters for Abyss and SOAPdenovo in the following sets.

• Set2 (Different parameters for Abyss - the assembler provides the lowest number of contigs in Set1)

Merged File: Set2_Contig

• Set3 (Different parameters for SOAPdenovo - the assembler provides the largest number of contigs in Set1)

Merged File: Set3_Contig

• Scaffold the contigs using SSPACE

Contig integrator

• CISA

The input file for CISA is an assembled set of contigs, e.g., the set1 contains the contigs obtained from Abyss, CLC, Edena, SOAPdenovo, and Velvet.

The integrated contigs generated by CISA can be directly downloaded via the Download link.

• MAIA

  maia('./assembly_list.txt','./NC_000913.fna')
  

assembly_list.txt:

Edena	./Edena_contigs.fa	2
SOAP    ./SOAP_contigs.fa   2
Velvet  ./Velvet.fa 2
CLC ./CLC.fa    2
Abyss	./Abyss_contigs.fa	3

A file named maia_assembly.fa was generated, containing a draft genome of 4640055 bp with 126761 uncalled bases (Ns). We split the genome into contigs (at >10Ns) and got the set of contigs maia.fa.

In using MAIA for other examples, we found that Matlab crashed while the recursion reached its limitation. We therefore split the reference genome into several parts, and then performed MAIA.

 maia('./assembly_list.txt','./NC_000913_1.fna')
 maia('./assembly_list.txt','./NC_000913_2.fna')
 maia('./assembly_list.txt','./NC_000913_3.fna')

The split references and the integrated results can be downloaded maia_ecoli_36.

• minimus2 and GAA

All_M2_GAA
Beacuase minimus2 and GAA merge two assemblies at a time, we iteratively integrate the five assemblies in random order.
minimus2: C_S_V_A_E, A_C_V_S_E, S_V_E_C_A, A_E_C_V_S, A_E_S_C_V, E_V_S_C_A, E_V_C_A_S, V_E_S_A_C, V_S_E_A_C, A_E_V_C_S
GAA: E_S_C_V_A, V_S_E_C_A, V_S_C_E_A, S_V_C_E_A, V_E_A_S_C, E_C_V_S_A, V_A_S_C_E, V_C_E_S_A, A_S_C_V_E, S_C_V_E_A

Evaluation

• Benchmark genome

Eshcherichia coli K12 MG1655

• Evaluated by Mauve Assembly Metrics

How to score genome assemblies using the Mauve system

• Evaluated by Blast with Features
• Evaluated by Gage

Gage

• Score with Mauve metrics:

Set1

[^#] Please note that GAA and minimus2 were designed to merge two assemblies at a time, we thus performed all runs and took the average scores.

[*] Please note that the scores of minimus2 and GAA were taken from the average of ten random combinations (details).

CISA integrates the five assemblies (Abyss, CLC, Edena, SOAPdenovo and Velvet from the outer to the inner) and sequentially generates the processed contigs after phase (1), (2), (3) and (4):

We have visually inspected the assemblies against the reference genome (NC_000913) by using graphic representations, e.g. dot plots. Therefore, we knew that the largest contig generated by minimus2(A,C,E,S,V) was misassembled. Moreover, we found that the merging order dramatically influences the result of minimus2.

Minimus2 firstly merges SOAPdenovo (the inner) with CLC (the outer), then merges the output (the inner) with Velvet (the outer) in the second run, merges the output (the inner) with Edena (the outer) in the third run, finally merges the output (the inner) with Abyss (the outer) in the fourth run and generates the integrated contigs.

Set2-Set3

As can be seen here, CISA can successfully integrate the sets of contigs and reduce the number of contigs. However, in comparison with integrating assemblies generated from different assemblers (Set 1), varying assemble parameters is less efficient (in the case of Set 2 or Set3) in completing the genome.

• Scaffold the contigs using SSPACE

Since we have the paired-end reads of E. coli, it is possible to assess the order, distance and orientation of contigs and combine them into scaffolds. We, therefore, used SSPACE to scaffold the contigs and quantified the scaffolds by Mauve assembly metrics.

The results show:

1. The integrated contigs output by CISA can be scaffolded by SSPACE to a limited extent (from 72 to 61), which suggests that our CISA can indeed integrate the sequence information from different assemblies.
2. To introduce the paired-end reads to the pre-assembled contigs generated by Abyss using SSPACE (Abyss+SSPACE) can only reduce the number of contigs from 133 to 97, smaller than the effect made by CISA (from 133 to 72), which suggests that contig integration prior to scaffolding can further enhance the result.
3. The problem of misassembled contigs generated by minimus2 is not yet solved by SSPACE, which suggests that we should integrate contigs with caution.