What is the best and RAM efficient pipeline for de novo assembly of...

[Diegodescarpates]

Hello everyone,

What is the best and RAM efficient pipeline for de novo assembly with two datasets including about 6 millions illumina paired-end reads of 36 bp and 44 millions illumina paired-end reads of 100 bp. It's a bacterian genome of 6.5 Mbp.

I tried Velvet, Abyss, SOAPdenovo... but it's seems impossible with 8G RAM.

Map Illumina reads to a backbone, is it a solution ? In this case, which pipeline ?

Thanks in advance for your help.

Diego

[winsettz]

What backbone would you have access to?

8 GB is going to be tough. My typical workflows to lower memory usage are to use longer kmer words and velvet's -create_binary (but I am using MiSeq), but I'm not sure if that will work with an 8G limit. What is your memory footprint assembling only the 36 bp paired ends?

Edit: C. Titus Brown has some workflows which may help. They are intended for very large metagenomic assemblies, but may be useful.

http://khmer.readthedocs.org/en/latest/introduction.html#introduction

[krobison]

In addition to digital normalization, you might try the Minia assembler, which is intended to be very memory efficient.

http://minia.genouest.org/

Also, Amazon EC2 is quite cheap as a source of compute power. You should be able to assemble this on EC2 with Ray for <10 euros -- one Quad Extra Large High Memory instance can devour much larger datasets in an hour or so.

[Diegodescarpates]

Thanks for your reply winsettz.

I have access to a working draft sequence (11x coverage), that's why I would prefer de novo assembly.
I am testing Gossamer so I can't answer now for memory footprint assembly but if I remember clearly he didn't exceed 8 GB RAM with 36 bp. With the second dataset the memory footprint was 8G (RAM) plus 5G of SWAP.

Thanks for the link, I am going to get into it.

Best regards,

Diego

[Diegodescarpates]

In addition to digital normalization, you might try the Minia assembler, which is intended to be very memory efficient.

http://minia.genouest.org/

Also, Amazon EC2 is quite cheap as a source of compute power. You should be able to assemble this on EC2 with Ray for <10 euros -- one Quad Extra Large High Memory instance can devour much larger datasets in an hour or so.

Thanks for information. I am trying Minia...

Best regards,

Diego