sequencing QC and metagenomics

[JQL]

Hi,

I just started my first metagenomics shot-gun sequencing analysis. I hope I can get some advice here.

1. QC process: I am looking at removing duplicate sequences and trimming low quality bases, according to this HMP's SOP. Is this adequate? What is considered a good QC practice?

2. Data analysis: I am first trying MetaPhlAn by Huttenhower's group. Is there any 'standard' procedure/workflow that this community recommends?

Thanks for your help,
John

[fanli]

Originally Posted by JQL

1. QC process: I am looking at removing duplicate sequences and trimming low quality bases, according to this HMP's SOP. Is this adequate? What is considered a good QC practice?

You may want to remove human/host DNA as well. kneaddata from Curtis Huttonhower's group does this, but uses bowtie2's --un-conc output to filter reads, which misses repetitive DNA that doesn't align concordantly. We've taken to manually filtering any read pair that aligns in any fashion to the human reference.

Originally Posted by JQL

2. Data analysis: I am first trying MetaPhlAn by Huttenhower's group. Is there any 'standard' procedure/workflow that this community recommends?

There are also k-mer based approaches (e.g. kraken and CLARK) that were recently shown to outperform MetaPhlAn: see here. Generally speaking, I don't think shotgun is as well figured out as 16S, so there is still some exploring to do. You may also be interested in PanPhlAn, ConStrains, etc. that do strain-level profiling.

We've just gotten into shotgun too, so happy to share any advice.

[JQL]

Thank you very much for your reply. At this point, any thing will help me get started.

have a nice weekend,
John

Originally Posted by fanli

You may want to remove human/host DNA as well. kneaddata from Curtis Huttonhower's group does this, but uses bowtie2's --un-conc output to filter reads, which misses repetitive DNA that doesn't align concordantly. We've taken to manually filtering any read pair that aligns in any fashion to the human reference.



There are also k-mer based approaches (e.g. kraken and CLARK) that were recently shown to outperform MetaPhlAn: see here. Generally speaking, I don't think shotgun is as well figured out as 16S, so there is still some exploring to do. You may also be interested in PanPhlAn, ConStrains, etc. that do strain-level profiling.

We've just gotten into shotgun too, so happy to share any advice.

[bastianwur]

Other things you'll need to do:
- assembly, best cross-assembly if you have multiple samples
- genomic binning (e.g. maxBin)
- gene prediction (prodigal has a setting for meta, but only bacteria; ncRNAs can be predicted with rnammer, trnascan-SE and rfam)
- function prediction (via KO, InterproScan, PRIAM, dbCAN, etc)