Transcriptome assembly using 12GB RAM?

[ckidner]

Is it possible to assemble a eukaryotic transcriptome from 50M Illumina paired end reads (101bp) using a machine with only 12 GB of RAM or am I going to have to find a bigger server? I've tried using Trinity, which can't manage it, but is there a more computationaly economical program I could use?

[pbluescript]

12GB is a tiny amount of RAM for an assembly. I would recommend getting more if you can.
Otherwise, you could try Oases. I've had cases where assembly requires more RAM on Trinity than on Oases and vice versa.
Besides that you could try removing poor quality reads, duplicate reads, or overlapping reads. Just be sure that the remaining reads are still paired in the fastq files.

[ckidner]

Thanks, I'll look into more RAM. I was just hopeful as it had managed Newbler 2.5 assemblies of 454 data a year or so ago.

[Wallysb01]

With that few of reads you *might* be able to manage it will ABySS/Trans-ABySS. I'd suggest doing some quality trimming (Q=20 is usually reasonable) and excluding unique kmers to decrease your memory footprint. But with a modern desktop computer you should really be able to get up to 32 GB of RAM for pretty cheap. If you do this much its worth the extra $100-200.

[pbluescript]

With 454, you get longer reads, so assembly is generally a less memory intensive process. Do your Illumina reads overlap? If so, you could try a program to merge them. That would decrease your read number but maintain the information and could reduce the amount of RAM needed.

[ymc]

Is 64GB enough for illumina human transcriptome assembly with about 100mil 100bp reads?

[pbluescript]

Is 64GB enough for illumina human transcriptome assembly with about 100mil 100bp reads?

It could be. Try it and see.

[Mark]

You could also try limiting over abundant reads and removing very low frquency (often erroneous) reads via the Titus Brown's khmer package. This can greatly reduce the size of your dataset without greatly affecting your assembly