PhyloCSF can identify protein-coding regions in the genome based on multiple-sequence alignments. PhyloCSF++ is an implementation of the original methods PhyloCSF and PhyloCSF HMM. It allows you to easily create browser tracks for any genome to identify coding regions, score local alignments or annotate GFF/GTF files with PhyloCSF and its confidence scores.
If you find our implementation useful and use it in your work, please consider citing it:
Christopher Pockrandt, Martin Steinegger, Steven Salzberg. PhyloCSF++: A fast and user-friendly implementation of PhyloCSF with annotation tools. Bioinformatics, 2021.
Please also consider citing the original method papers:
Lin MF et al. PhyloCSF: a comparative genomics method to distinguish protein-coding and non-coding regions. Bioinformatics, 2011.
Mudge JM et al. Discovery of high-confidence human protein-coding genes and exons by whole-genome PhyloCSF helps elucidate 118 GWAS loci. Genome Research, 2019.
Download the latest release (static binary for Linux 64bit).
$ conda install -c conda-forge -c bioconda phylocsfpp
If you want to build it from source, we recommend cloning the git repository as shown below. You will need the GNU scientific library (gsl), which should be available in your package manager.
$ git clone https://github.com/cpockrandt/PhyloCSFpp.git PhyloCSF++ $ mkdir PhyloCSF++/build && cd PhyloCSF++/build $ cmake .. -DCMAKE_BUILD_TYPE=Release $ make
You can then install and run PhyloCSF++ as follows
$ sudo make install $ phylocsf++ --help
or run the binary directly:
$ ./phylocsf++ --help
Requirements
- Operating System
- GNU/Linux, Mac
- Compiler
- GCC ≥ 4.9, Clang ≥ 3.8
- Build system
- CMake ≥ 3.2
- Dependencies
- GNU scientific library (gsl), OpenMP, zlib
A detailed list of arguments and explanations can be retrieved with --help:
$ phylocsf++ --help $ phylocsf++ build-tracks --help
Please also check out the FAQ in our wiki.
To produce PhyloCSF tracks including the power track (confidence scores), you can run the following command.
Afterwards, you can use wigToBigWig to transform the outputted wig files to bw files and load them into a genome browser.
$ phylocsf++ build-tracks 58mammals hg38.100way.maf
We recommend to additionally smoothen the tracks (i.e., compute posterior probabilities) by setting --output-phylo 1.
This requires to pass the total genome length and a list of known coding regions as training data.
Here we extract the known coding regions from a GFF file using awk (for more information on the file format, please have a look at the wiki.
$ awk -F'\t' 'BEGIN { OFS="\t" } ($3 == "CDS") { print $1, $7, $8, $4, $5 }' gene_catalogue.gff > CodingExons.txt
$ phylocsf++ build-tracks --output-phylo 1 --genome-length 3272116950 --coding-exons CodingExons.txt 58mammals hg38.100way.maf
If not all of the species from the model are present in your alignments, reducing the model can speed up the computation significantly. For this pass all species from your alignment, e.g., --species Human,Chimp,Gorilla.
Here is a minimal example with all files at hand in the repository.
It includes a small set of alignments for chicken (galGal6).
The wig files are written to ./galgal6-tracks.
$ gunzip example/galGal6_chr22_25_28_each_30k_bases.maf.gz
$ phylocsf++ build-tracks \
--threads 4 --output ./galgal6-tracks \
--output-phylo 1 --genome-length 1065365434 --coding-exons example/galGal6_coding_exons.txt \
53birds example/galGal6_chr22_25_28_each_30k_bases.maf
Instead of using a pre-defined model such as 53birds, you can also pass an external model from the files, e.g., ./test/53_birds.
If you have tracks in bigWig computed or downloaded, PhyloCSF++ can annotate CDS with PhyloCSF and confidence scores:
$ phylocsf++ annotate-with-tracks /path/to/PhyloCSF+1.bw genes.gff
For this you need to have all six files in the same directory (PhyloCSF+1.bw, PhyloCSF+2.bw, etc.) as well as PhyloCSFpower.bw if you also want to compute confidence scores.
Here is a minimal example with all files at hand in the repository.
It includes a few transcripts from chicken (galGal6) and precomputed tracks.
The output is written into the same directory with the suffix .PhyloCSF++.gtf.
$ phylocsf++ annotate-with-tracks ./example/tracks/PhyloCSF+1.bw ./example/galGal6_chr22_25_28_subset_ensGene.gtf $ less ./example/galGal6_chr22_25_28_subset_ensGene.PhyloCSF++.gtf
For some species you can download complete track files either on the Broad institute server or here: ftp://ftp.ccb.jhu.edu/pub/software/phylocsfpp/
If you want to score alignments and not create tracks for an entire genome, simply run:
$ phylocsf++ score-msa 58mammals hg38.100way.maf
You can also specify the strategy (fixed, mle and omega; default: mle) and choose which scores to compute (PhyloCSF score, ancestral sequence composition score, branch length score).
NOTE: compared to the original implementation of PhyloCSF, PhyloCSF++ only scores the forward strand starting from the first base. For other frames and strands, you need to remove the first 1-2 bases and/or compute the reverse complement of the sequences.
To make these scores easier to interpret, we added the mode fixed_mean.
It scores every codon in the MSA, computes posterior probabilities and computes a mean over all codons.
Hence, the final score is in the interval [-15, +15] just as the tracks.
If you don't have tracks available for your genome of interest, PhyloCSF++ can annotate CDS with PhyloCSF and confidence scores by computing an alignment on the fly using MMseqs:
$ phylocsf++ annotate-with-mmseqs genomes.txt 58mammals genes.gff
genomes.txt has to contain the paths to genomes from the selected model to align to (one per line).
After all CDS lines were extracted and aligned with MMseqs, PhyloCSF++ scores each CDS alignment with the sub-tool score-msa.
We think that PhyloCSF is a very useful method for gene finding and annotation. Unfortunately no binaries are available and we think the outdated Ocaml code might be difficult to get running for inexperienced users. To build tracks the user also has to set up their own pipeline and do some coding. Hence, we thought it would be helpful to make an easy-to-use program that merges all necessary steps into a single step to quickly create tracks for entire genomes. As part of this project we computed tracks for more species and included them into the UCSC genome browser as well as offer them for download:
ftp://ftp.ccb.jhu.edu/pub/software/phylocsfpp
This is an implementation of the original methods (PhyloCSF and PhyloCSF HMM), which were released under the GNU AGPL v3 and Apache License v2. We have reimplemented the core algorithms (originally written in OCaml and Python) in C++, they were not changed except for running time improvements or where explicitly stated in the source code.