Genome assembly of mixed E. coli samples

[agmcarthur]

I have a question about assembling mixed samples, where the organisms sampled are very similar. We usually sequence pure isolates but in a particular study ended up Illumina sequencing (paired ends) a sample that included a number of E. coli strains. These are all commensal E. coli but we know from separate work they have different phenotype and PFGE results. If each bacteria in the sample was a different organisms (e.g. E. coli, Salmonella, Acinetobacter) I know we could attempt to assemble the data and pull out separate genomes. But in this case, much of the genomes are identical or very similar. Yet other regions will include gene additions / deletions or genome re-arrangments. I was thinking that this would be analogous to assembling a plant or animal polyploid genome, where parts of the chromosome could be assembled to a haploid consensus but others would have multiple solutions, reflective of difference in chromosome copies. Can anyone suggest an assembly approach for our data or perhaps a polyploid assembler that may applied? Thanks!

[HESmith]

I don't know if the polyploid assemblers will work with your sample. Ignoring repetitive elements (which are difficult enough in haploids), polyploid genomes have (approximately) integer multiples of the haploid genome for variant representation. The data corresponding to one haploid genome, as well as the ploidy, can be estimated from kmer frequencies. For example, a tetraploid genome will typically have the tallest kmer peak at 4X the frequency of the first observable peak (the former represents the non-variant sequences, while the latter represents variants unique to one haploid genome), with peaks at 2X and 3X as well. Your sample has an unknown number of strains with non-integer representations, and you expect the majority of their genomes to be identical. Your predicted distribution of kmer frequencies would be one large peak corresponding to your depth of coverage, with no obvious peaks at lower coverage.

Perhaps the best approach would be to align the data to an E. coli reference with stringent parameters (e.g., minimal mismatches and gaps) to remove the bulk of the data, then try to assemble the unaligned reads with a metagenome assembler. Note that I have not attempted such an approach, so it may not work.

[maxsalm]

Hi there! You could try the CORTEX assembler: http://www.ncbi.nlm.nih.gov/pubmed/23172865

[Brian Bushnell]

The best approach would probably be to sequence and assemble with very long reads (PacBio) in order to separate the strains. Short reads are not very good for phasing.

[rhinoceros]

Give some metagenome assembler like IDBA-UD a try. It takes into account read coverage. Chances are high that your strains were not present in the isolated DNA in equal abundance. Also set very strict threshold for similarity, e.g. 99%.