Re: MetaPhlAn for Soil metagenomes

[Nicola Segata]

Hi Joe,

 the coverage of reference genomes for microbes present in the human body is very good (thanks to several recent projects that filled the holes in the sequenced human-associated taxonomy). As a consequence, MetaPhlAn performs very well on human metagenomes. For soil metagenomes, instead, the coverage is not comprehensive and, consequently, MetaPhlAn (and the other approaches) are not performing comparably to the human case.

Specifically answering to your question, we do have environmental clades in MetaPhlAn. In the next release (on which we are working right now) there will be many more species and thus the coverage of environmental microbes should be higher, but I suspect it will still be quite far from the human microbiome case. I would always couple 16S analysis (even using the reads from metagenomics) for non-human associated microbiomes, but MetaPhlAn should predict the abundance of specific species when possible, and the coverage of environmental species will definitely grow as we keep it updated with the increasing number of available reference genomes from environmental sources.

many thanks

Nicola

On Thursday, November 22, 2012 2:55:30 PM UTC+1, Joe wrote:

Hi Nicola,
I'm currently working on a project where we are considering the use of MetaPhlAn for analysis of soil samples. Is this possible using the current BLAST database, or is it adjusted for use with human metagenomic samples only? If this is the case, are there other BLAST databases or parameters which could be changed to optimize MetaPhlAn for environmental results?

Many thanks,
Joe    

[Elisa Catão]

Hi Nicola and Joe,
I am too working with soil metagenomics and since I heard Curtis Huttenhower talk in the last GRC-AEM I was excited to try the metaphlan. But I have just managed to make it work in my pc and unfortunately it gave back 100% unclassified.
It was so disappointing.
I think your idea of performing human microbiome classification with several clade-markers was so interesting.
I will keep updated to see if the next release will be able to classify any of my sequences.

But let me ask you something Nicola. I have 750bp sequences from 454 in my metagenomic data. Do you think that if I use as an input the "assembled" data it can work better?

thanks,
Elisa

[Kristjan Oopkaup]

Hi Nicola
Does the latest version of Metaphlan include higher number of environmental reference genomes? I ran it on our environmental data (assembled Illumina paired-end reads, mean length about 300 bp) and it found almost nothing.

Thanks
Kristjan

On Saturday, October 26, 2013 10:25:07 AM UTC+3, Nicola Segata wrote:
> Hi Elisa,
>  thanks for getting in touch. MetaPhlAn has been mostly applied on Illumina dataset (shorter reads but higher throughput compared to 454). As a consequence, the default for BowTie2 mapping is to perform an end-to-end alignment of each read. For 750bp-long reads I'd strongly recommend the local alignment option for BowTie2 or a the Blast alternative (if computationally feasible). For BowTie2 it should be sufficient to set "--bt2_ps very-sensitive-local". The total number of reads in your metagenome is also an important aspect, how large are your 454 datasets?
>
>
> thanks
> Nicola
>
> On Thursday, October 24, 2013 8:27:12 PM UTC+2, Elisa Catão wrote:Hi Nicola and Joe,

[Nicola Segata]

Hi Kristjan,

 yes, the MetaPhlAn version 2 that we are testing right now starts from 5x more genomes and 3x more microbial species. Also, for MetaPhlAn I'd recommend using unassebled reads, just considering the two ends independently.

thanks

Nicola

[Nicola Segata]

Hi Elisa,

 thanks for getting in touch. MetaPhlAn has been mostly applied on Illumina dataset (shorter reads but higher throughput compared to 454). As a consequence, the default for BowTie2 mapping is to perform an end-to-end alignment of each read. For 750bp-long reads I'd strongly recommend the local alignment option for BowTie2 or a the Blast alternative (if computationally feasible). For BowTie2 it should be sufficient to set "--bt2_ps very-sensitive-local". The total number of reads in your metagenome is also an important aspect, how large are your 454 datasets?

thanks

Nicola

On Thursday, October 24, 2013 8:27:12 PM UTC+2, Elisa Catão wrote: