Phylogenetic tree building from proGenomes database for shotgun metagenomics

Question

For some weeks I'm fighting with an issue about phylogenetic tree building to use in a phyloseq object in order to calculate beta-diversity metrics that takes into account phylogenetic distance.

I’m trying to create a phylogenetic tree from the representative genomes of proGenomes v.2.1 database. However, I do not have a representative genome for each taxa id. In contrast, I have a multifasta files for more than 100 contigs for each taxa ID (file freeze12.contigs.representatives.fasta.gz).

For each taxaID (the first numbers in each header of the multifasta file is NCBI taxon ID), I would like to bin the contigs into the biggest scaffold possible, and then perform an alignment (through MUSCLE as examples) followed by RaXML-NG to construct a Newick tree.

However, I cannot find any software for binning contigs with only the contigs as inputs. MetaBAT needs bam files from shotgun read mappings to the contigs in question (not present in the proGenomes database). I found another software (Scaffold_builder), but it needs a representative reference genome. And in microbiology I have more than 100 complete genomes for each NCBI tax id...

For this reason, I ask here an open question if exists some software that can perform of contigs taking into account only a multifasta file of contigs?

Answer 1

I would not recommend the approach that you describe. I am not sure that I completely understand it. For example, metagenomic binning by e.g. MetaBAT is not a scaffolding operation, which returns a single sequence. Rather, they are methods to separate a pool of contigs into groups with similar sequence characteristics indicating they come from the same genome.

It sounds like you already have a group of contigs from each genome, you just want them to all be part of a single sequence in a FASTA file, as would be suggested by your interest in the reference-guided scaffolding approach scaffold_builder.

I would propose that you instead use a tool like GTDBtk to place each of your genomes (each represented by a multifasta of contigs, which is the expected input) into an existing taxonomic tree, and then use their placements in that tree as a model of the taxonomy of your genomes, rather than trying to infer it de novo. I assure you that it is a high quality phylogeny that no one will criticize, and it avoids the possibility of introducing errors into your phylogeny through problematic scaffolding operations.

I would suggest reading GTDBtk documentation for more information.

A narrower answer to the original question

I realized that there is an answer to the question of binning contigs without any additional information. I do not recommend it as I think it would give you positively wrong answers because you are starting with your genomes already binned, and it would not provide you with scaffolds, which I think is what you actually want. But there is a way to perform this task.

You can compute tetranucleotide frequencies from contigs and use them to cluster arbitrary sets of contigs into genomes or other groupings based on similarities. This has a significant rate of error, but it does more or less work. It's an older approach.

Answer 2

Thank you another time @Maximilian Press. Yes, GTDB-Tk it's a robust software but although the merging of archaea and bacteria taxa is still a current research topic, I have succeed with another tool that permits me obtaining an unique tree (including archea and bacteria).

If I work with GTDB-Tk I've obtained one tree for archaea and other for bacteria. Then to use it in phyloseq I must work with two independent phyloseq objects which is nice also but my purpose of building a tree from shotgun is only to permits me to calculate diversities based on phylogeny.

As alternative that works it was the following: to construct a phylogenetic tree from Kaiju I used the 40 marker proteins that is based on the proGenomes database used by kaiju (https://progenomes.embl.de/data/proGenomes2.1_markerGenes.tar.gz) and then I've used the ETE-Toolkit(http://etetoolkit.org/) software to construct the tree in Newick format that I can integrate into a phyloseq object.

I post here if someone can help my strategy. It has been no easy (unless for a dummie as me) to construct a phylogenetic tree from Kaiju's taxonomic classifier.