Skip to content

PHYling pipeline for species tree building from annotated genomes (see https://github.com/stajichlab/AAFTF and https://github.com/nextgenusfs/funannotate for assembly and annotation steps)

License

Notifications You must be signed in to change notification settings

stajichlab/PHYling

Repository files navigation

Conda build Python build Python codecov License

PHYling tool

The unified PHYling pipeline for phylogenomic data collection from annotated genomes.

This is latest iteration of tool for using phylogenetically conserved markers to pull out informative gene or protein info from genomic and transcriptomic datasets in order to construct gene trees and species phylogenies.

The aligned markers can be extracted from protein sequences for phylogenetic analyses and also projected into coding sequence alignments for codon-based analyses for better resolution of recently diverged species.

The assumptions in this approach are that the markers are generally single copy in genomes and taking best hit is sufficient first approximation for identifying orthologs. A separate file is parsed and file best_multihits which lists all the hits above the cutoff threshold for a given marker which can be used to assess duplication or attempt to incorporate paralogs into the analysis down the road.

The marker sets developed for this approach in fungi are available as part of the 1KFG Phylogenomics_HMMs project resource and preferred use of the BUSCO marker sets.

Flow chart

PHYling flowchart

New features compared to the original version

  • Using pyhmmer to improve the multithread performance in hmmsearch and hmmalign.
  • Implement all stuff in python. The entire program will be more readable and maintainable.
  • Simplify some steps and reduce the intermediate files as much as possible.
  • Muscle is now available for alternative alignment method.
  • Use PhyKIT to remove uninformative orthologs.
  • FastTree, RAxML and IQTree are now available for tree construction.
  • ASTER, a C++ version of ASTRAL is now integrated to resolve consensus among trees built upon individual genes.

Usage

First of all, install the package following the instruction below.

PHYling is a package to extract phylogenomic markers and build a phylogenetic tree upon them. It comprises 3 modules - download, align and tree. Use phyling --help to see more details.

positional arguments:
  {download,align,tree}
    download            Download HMM markers
    align               Run multiple sequence alignments against orthologs found among samples
    tree                Build a phylogenetic tree based on multiple sequence alignment results

options:
  -h, --help            show this help message and exit
  -V, --version         show program's version number and exit

To test run on the example files, please cd into the folder example.

cd example

In general, PHYling takes fasta as input. The gzipped fasta is also valid.

The folder example/pep includes 5 example peptide fasta which can be used for test run.

In addition to the peptide sequences, PHYling can also takes DNA coding sequences as inputs to more accurately estimate the phylogeny of closely related species. When taking DNA coding sequences as inputs, DNA sequences will be translated into peptide sequences and all the hmmsearch/align are done on the peptide version. The final MSA results will be back-translated into DNA at the final stage. The example DNA sequences are placed under the folder example/cds.

Download HMM markerset

The download module is used to download the HMM markerset from BUSCO website. (Currently is updated to v5) See all options with phyling download --help.

positional arguments:
  HMM markerset or "list"
                        Name of the HMM markerset

options:
  -h, --help            show this help message and exit
  -v, --verbose         Verbose mode for debug

Firstly, use download list to show the available BUSCO markersets.

phyling download list

By default the downloaded markersets will be saved to the ~/.phyling/HMM. The Datasets available online section lists the markersets that available on the BUSCO website.

And the Datasets available on local section lists the markersets that have already been downloaded.

To download the markerset, copy the name from the list and paste it to the download module directly. Here we use fungi_odb10 as example.

phyling download fungi_odb10

The download module will automatically check for updates to the markerset online each time it runs. The local markersets which have available updates online will be marked as [Outdated]. You can rerun phyling download [markerset] to update the local files.

Find the orthologs and align them

The align module identify the orthologs among all the samples using hmmsearch. HMM profiles that have matches on more than 4 samples are considered orthologs.

Before conducting hmmsearch, the module will first search for the bitscore cutoff file within the root HMM folder. If the cutoff file is not found, the reporting threshold for hmmsearch will be determined based on the -E/-evalue (default is 1e-10).

Once the orthologs are identified, the sequences extracted from each sample undergo multiple sequence alignment. By default, the alignment is performed using the hmmalign method. However, users have the option to switch to muscle by specifying the -M/--method muscle flag.

Finally, each alignment result is output separately. You can decide whether you want to concatenate them or use consensus strategy in the tree module. See all the options with phyling align --help.

options:
  -h, --help            show this help message and exit
  -v, --verbose         Verbose mode for debug
  -i file [files ...], --inputs file [files ...]
                        Query pepetide/cds fasta or gzipped fasta
  -I directory, --input_dir directory
                        Directory containing query pepetide/cds fasta or gzipped fasta
  -o directory, --output directory
                        Output directory of the alignment results (default: phyling-align-20240423-173849-0700 [current timestamp])
  -m directory, --markerset directory
                        Directory of the HMM markerset
  -E float, --evalue float
                        Hmmsearch reporting threshold (default: 1e-10, only being used when bitscore cutoff file is not available)
  -M {hmmalign,muscle}, --method {hmmalign,muscle}
                        Program used for multiple sequence alignment (default: hmmalign)
  --non_trim            Report non-trimmed alignment results
  -t THREADS, --threads THREADS
                        Threads for hmmsearch and the number of parallelized jobs in MSA step. Better be multiple of 4 if using more than 8 threads (default: 1)

Run the align module with all the fasta files under folder pep.

phyling align -I pep -o align -m fungi_odb10

An equivalent way to send inputs.

phyling align -i pep/*.fasta.gz -o align -m fungi_odb10

Or if you're just interested in part of the fasta, you can specify the inputs one-by-one.

phyling align -i pep/Pilobolus_umbonatus_NRRL_6349.aa.fasta.gz \
  pep/Rhizopus_homothallicus_CBS_336.62.aa.fasta.gz \
  pep/Rhizopus_rhizopodiformis_NRRL_2570.aa.fasta.gz \
  pep/Zygorhynchus_heterogamous_NRRL_1489.aa.fasta.gz \
  -o align \
  -m fungi_odb10

Note: Required at least 4 samples to build a tree!

Accelerate by using 16 cpus.

phyling align -I pep -o align -m HMM/fungi_odb10/hmms -t 16

Multithreading strategy

According to pyhmmer benchmark, the acceleration benefits from multithreading drop significantly as more CPUs are utilized. When less then 8 cpus are given, the hmmsearch step will run on single-thread manner and all cpus will be used for each round of hmmsearch. When 8 or more cpus are given, the hmmsearch step will use 4 cpus for each parallel job. In the example above, 4 hmmsearch jobs will run parallelly and each job utilize 4 cpus. For the alignment step, 16 parallel jobs will be launched and each parallel job is running on single-thread manner.

Highly recommended if muscle is chosen for alignment. (muscle is much slower than hmmalign!!)

Use coding sequences instead of peptide sequences

In some circumstances, the highly shared peptide sequences make it difficult to resolve the relationship among closely related species. To address the issue, one can use DNA coding sequences (CDS), which contain more evolutionary traces, instead of peptide sequences for phylogeny analysis.

Run the align module with cds fasta files under folder cds.

phyling align -I cds -o align_cds -m HMM/fungi_odb10/hmms -t 16

The CDS inputs will be translated into peptide sequences in the first steps. The translated peptide sequences will be used for hmmsearch and the alignment steps. The peptide alignment results will then being back-translated according to the original CDS inputs. And the back-translated DNA version alignments will be output.

Checkpoint for quick rerun

Once the align module complete, a checkpoint file will be generated to the output folder. This checkpoint file stores the parameters, samples and identified orthologs, which will be loaded to the pipeline when rerun with the same output folder. Then the align module will determine whether to skip the hmmsearch on some of the samples that were already completed in the previous run.

For example, we run the align module by:

phyling align -i pep/Pilobolus_umbonatus_NRRL_6349.aa.fasta.gz \
  pep/Rhizopus_homothallicus_CBS_336.62.aa.fasta.gz \
  pep/Rhizopus_rhizopodiformis_NRRL_2570.aa.fasta.gz \
  pep/Zygorhynchus_heterogamous_NRRL_1489.aa.fasta.gz \
  -o align \
  -m fungi_odb10
  -t 16

And later we want to add Actinomucor elegans to the analysis but kick out Zygorhynchus heterogamous. We can start another run and specifying the same output folder:

phyling align -i pep/Actinomucor_elegans_CBS_100.09.aa.fasta.gz
  pep/Pilobolus_umbonatus_NRRL_6349.aa.fasta.gz \
  pep/Rhizopus_homothallicus_CBS_336.62.aa.fasta.gz \
  pep/Rhizopus_rhizopodiformis_NRRL_2570.aa.fasta.gz \
  -o align \
  -m fungi_odb10
  -t 16

In this case, Pilobolus umbonatus, Rhizopus homothallicus and Rhizopus rhizopodiformis will be skipped from the hmmsearch process since they have already been searched in the previous run. The Actinomucor elegans is the only sample need to be hmmsearched. On the other hand, the Zygorhynchus heterogamous will be removed from the current run.

Note that if the input files have changes, they will also being detected by align module and trigger the rerun. If the hmmsearch evalue/bitscore cutoff is changed, all the samples will need to rerun the hmmsearch step. Also, the changes on HMM markersets or input samples with different seqtype will terminate the align module. (since this case should be considered an entirely different analysis)

Build tree from multiple sequence alignment results

Finally, we can run the tree module, use the multiple sequence alignment results to build a phylogenetic tree. By default, it uses the consensus tree strategy (conclude the majority of trees which was built upon each single gene) But you can choose to use concatenated alignment strategy by specifying -c/--concat. Currently, 3 methods (FastTree, RAxML and IQTree) are available for tree building. You can choose your own preferred method by specifying -M/--method. (default is FastTree) We also use the treeness/RCV calculated by PhyKit and select the most informative markers for final tree building. You can adjust the number of selected markers by specifying -n/--top_n_toverr. See all the options with phyling tree --help.

options:
  -h, --help            show this help message and exit
  -v, --verbose         Verbose mode for debug
  -i file [files ...], --inputs file [files ...]
                        Multiple sequence alignment fasta of the markers
  -I directory, --input_dir directory
                        Directory containing multiple sequence alignment fasta of the markers
  -o directory, --output directory
                        Output directory of the newick treefile (default: phyling-tree-20240423-183138-0700 [current timestamp])
  -M {ft,raxml,iqtree}, --method {ft,raxml,iqtree}
                        Algorithm used for tree building. (default: ft)
                        Available options:
                        FastTree: ft
                        RAxML: raxml
                        IQTree: iqtree
  -n TOP_N_TOVERR, --top_n_toverr TOP_N_TOVERR
                        Select the top n markers based on their treeness/RCV for final tree building (default: 50. Specify 0 to use all markers)
  -c, --concat          Concatenated alignment results
  -p {seq,codon,seq+codon}, --partition {seq,codon,seq+codon}
                        Create a partition file by sequence or by codon position when --concat enabled. "codon" and "seq+codon" only work when inputs are DNA sequences
  -f, --figure          Generate a matplotlib tree figure
  -t THREADS, --threads THREADS
                        Threads for tree construction (default: 1)

Run the tree module with all the alignment results under folder align.

phyling tree -I align

You can also use only part of the alignment results to build tree.

phyling tree -i align/100957at4751.aa.mfa align/174653at4751.aa.mfa align/255412at4751.aa.mfa

Note: the tree module uses the checkpoint file .align.ckp in the input folder (or the parent folder of the input files) to determine the sample names and seqtype. If the checkpoint file is missing or corrupted it can also automatically detects these information but requires more time.

Use IQTree instead of the default FastTree method for tree building and running with 16 threads.

phyling tree -I align -m iqtree -t 16

Use matplotlib to generate a tree figure.

phyling tree -I align -f -t 16

Use top_n_toverr to filter the markers by treeness/RCV

The align module sometimes reports a lot of orthologs depending on the size of BUSCO dataset and the gene set homogeneity among samples. However, not every marker is equally informative in resolving phylogeny; some substitutions may not contribute significantly to the branches in the phylogenetic tree. To address this issue, PHYling incorporates PhyKIT to compute the treeness/RCV scores (toverr) and ranks the markers accordingly. Higher ranks indicate greater informativeness and lower susceptibility to composition bias and will be selected for final tree building (thru consensus or concatenate strategy).

By default, PHYling pick the top 50 markers for further analysis but user can adjust the number by specifying -n/--top_n_toverr. In the example below, we pick only the top 20 markers:

phyling tree -I align -n 20 -t 16

All markers will undergo tree building and compute their treeness/RCV scores, and the top 20 markers will be selected to reconstruct the final consensus tree. (PHYling uses consensus strategy by default)

Note that if the number of identified orthologs is less than the assigned -n/--top_n_toverr value, PHYling will use all markers instead. Users can also specify 0 to use all the markers directly.

phyling tree -I align -n 0 -t 16

After tree construction, a file top_toverr_trees.tsv recording the treeness/RCV of selected markers and a folder selected_MSAs containing the symlinks to the mfa of the selected markers will be output as well.

Use concatenate strategy

The consensus strategy is used for tree construction by default but users can choose to concatenate the markers and generate a single tree on it.

phyling tree -I align -c -t 16

When concatenation mode is enabled, PHYling do the first round tree building on each marker with FastTree and compute their toverr. These highly ranked markers are then selected for concatenation and do the final tree building. (using the method specified by -M/--method) The concatnated fasta concat_alignments.mfa will also being output which allow users to perform tree building with tools not incorporated in PHYling. The example below shows to pick the top 20 markers and build a tree with concatenate strategy by specifying -n/--top_n_toverr.

phyling tree -I align -n 20 -c -t 16

Meanwhile, users can construct tree with a more sophisticated partition mode when using RAxML and IQTree. In general, the partition mode expects different sequence regions exhibit different evolutionary rates, which should be estimated with different models. Here we provide 3 different most commonly-used modes:

  • seq: partitioning by marker. (each marker evolves separately)
  • codon: partitioning by codon of 3 of concatenated sequence. (only available on CDS. Each codon evolves separately)
  • seq+codon: partitioning by codon of 3 marker-wisely. (only available on CDS. Each codon of each marker evolves separately)

The example below concatenate the top markers (50 by default) and run tree building with sequence partitioning thru iqtree.

phyling tree -I align -M iqtree -c -p seq -t 16

Note: Partition mode is not supported in FastTree.

Tune it yourselves

To adapt the most common needs, PHYling uses the very basic commands to run FastTree, RAxML and IQTree:

FastTree for peptide:

FastTree -gamma file.mfa -lg

FastTree for cds:

FastTree -gamma file.mfa -nt

RAxML for peptide:

raxml -s file.mfa -p 12345 -w [absolute_path_output] -n pep -m PROTGAMMAAUTO

RAxML for peptide + partition mode:

raxml -s file.mfa -p 12345 -w [absolute_path_output] -n pep -m PROTGAMMAAUTO -q file.partition

RAxML for cds:

raxml -s file.mfa -p 12345 -w [absolute_path_output] -n cds -m GTRCAT

RAxML for cds + partition mode:

raxml -s file.mfa -p 12345 -w [absolute_path_output] -n cds -m GTRCAT -q file.partition

IQTree for pep/cds:

iqtree2 -s file.mfa --prefix [output_path] -m MFP --mset raxml -T AUTO

IQTree partition mode:

iqtree2 -s file.mfa --prefix [output_path] -m MFP --mset raxml -T AUTO -p file.partition

You can use the tree module to prepare the required data (i.e. concat_alignment.mfa or toverr filtered mfas) and rerun the tree building step with your own preferred parameters.

Requirements

  • Python >= 3.9
  • Biopython
  • pyhmmer, a HMMER3 implementation on python3.
  • muscle for alternative method for multiple sequence alignment. (Optional)
  • ClipKIT for removing sites that are poor of phylogenetic signal.
  • PhyKIT for calculating treeness/RCV to filter uninformative orthologs.
  • FastTree, use approximately maximum-likelihood to build trees.
  • RAxML, a more sophisticated maximum-likelihood-based tree building tool. (Optional)
  • IQTree, a modern maximum-likelihood-based tool for tree building. (Optional)
  • ASTER, a C++ re-implementation of ASTRAL to resolve consensus among trees.

Install

Please download the source code from the latest release and decompress it or git clone the main branch.

Go into the PHYling folder. To avoid altering the base environment, it's advisable to install the software in a dedicated conda environment Please use the environment.yml to create environment and install all the required packages.

cd PHYling-2.0.0
conda env create -f environment.yml

Install the package through pip in the PHYling folder.

pip install .

Install additional package for developing (developer only)

Developer should clone the GitHub project directly instead of downloading from the releases. Some of the files for developing purpose only are not included in the releases.

In addition to the requirements listed above, the following packages are required for developing environment.

Developer can install it with conda by:

conda install pre-commit>=3.4.0

For convenience we also provide the additional conda env file. Please use the dev_additional_packages.yml to install the additional packages.

conda env update -f dev_additional_packages.yml