MetaEuk - sensitive, high-throughput gene discovery and annotation for large-scale eukaryotic metagenomics
MetaEuk is a modular toolkit designed for large-scale gene discovery and annotation in eukaryotic metagenomic contigs. MetaEuk combines the fast and sensitive homology search capabilities of MMseqs2 with a dynamic programming procedure to recover optimal exons sets. It reduces redundancies in multiple discoveries of the same gene and resolves conflicting gene predictions on the same strand. MetaEuk is GPLv3-licensed open source software that is implemented in C++ and available for Linux and macOS. The software is designed to run efficiently on multiple cores.
MetaEuk can be used by compiling from source (see below) or downloading a statically compiled version. It requires a 64-bit system (check with
uname -a | grep x86_64) with at least the SSE4.1 instruction set (check by executing
cat /proc/cpuinfo | grep sse4_1 on Linux or
sysctl -a | grep machdep.cpu.features | grep SSE4.1 on MacOS).
# install via conda conda install -c conda-forge -c bioconda metaeuk # static Linux AVX2 build wget https://mmseqs.com/metaeuk/metaeuk-linux-avx2.tar.gz; tar xzvf metaeuk-linux-avx2.tar.gz; export PATH=$(pwd)/metaeuk/bin/:$PATH # static Linux SSE4.1 build wget https://mmseqs.com/metaeuk/metaeuk-linux-sse41.tar.gz; tar xzvf metaeuk-linux-sse41.tar.gz; export PATH=$(pwd)/metaeuk/bin/:$PATH # static macOS build (universal binary with SSE4.1/AVX2/M1 NEON) wget https://mmseqs.com/metaeuk/metaeuk-osx-universal.tar.gz; tar xzvf metaeuk-osx-universal.tar.gz; export PATH=$(pwd)/metaeuk/bin/:$PATH
Precompiled binaries for other architectures (ARM64, PPC64LE) and very old AMD/Intel CPUs (SSE2 only) are available at https://mmseqs.com/metaeuk.
MetaEuk will search for eukaryotic protein-coding genes in contigs based on similarity to a reference set of proteins or protein profiles. The starting point are Fasta files of sequences (you can use contigs.fna and proteins.faa from the tests/two_contigs directory as a small toy example).
You could either use the
easy-predict workflow directly on the Fasta files or convert them to databases by running the createdb command and later on specific MetaEuk modules.
Read here to learn more on how to create a protein profile database using MMseqs2. Once created, this database can be used as referenceDB in the commands below.
A gene call is an optimal set of exons predicted based on similarity to a specific target (T) in a specific contig (C) and strand (S). In the following it is referred to as a TCS or as a call. After redundancy reduction (see details below), the representative TCS is referred to as prediction.
easy-predict Predict proteins from contigs (fasta/db) based on similarities to targets (fasta/db) and return a fasta & GFF predictexons Call optimal exon sets based on protein similarity reduceredundancy Cluster metaeuk calls which share an exon and select representative unitesetstofasta Create fasta output from optimal exon sets (and (1) a TSV map between headers and internal identifiers, (2) GFF summary) groupstoacc Create a TSV output from representative to calls taxtocontig Assign taxonomic labels to MetaEuk predictions and contigs by majority voting
Using MMseqs2 commands within MetaEuk:
MMseqs2 commands are available through MetaEuk and no additional MMseqs2 installation is required.
For example, the MMseqs2 command
mmseqs createdb can be replaced with
mmseqs databases with
metaeuk databases, etc. Please see also the MMseqs2 Wiki for more info about MMseqs2 commands.
--min-length minimal number of codons in putative protein fragment -e maximal E-Value to retain a match between a putative protein fragment and a reference target --metaeuk-eval maximal combined E-Value to retain an optimal exon set --metaeuk-tcov minimal length ratio of combined set to target --exhaustive-search if referenceDB is a profile database, should be added (before version 4 called slice-search) --max-exon-sets maximal number of exon sets on each contig and strand for a given target (from version 6)
This workflow combines the following MetaEuk modules into a single step: predictexons, reduceredundancy and unitesetstofasta (each of which is detailed below). Its inputs are contigs (either as a Fasta file or a previously created database) and targets (either as a Fasta file of protein sequences or a previously created database of proteins or protein profiles). It will run the modules and output the predictions in Fasta format (as well as a GFF format).
metaeuk easy-predict contigsFasta/contigsDB proteinsFasta/referenceDB predsResults tempFolder
It will result in predsResults.fas (protein sequences), predsResults.codon.fas, predsResults.headersMap.tsv and predsResults.gff.
Calling optimal exons sets:
This module will extract all putative protein fragments from each contig and strand, query them against the reference targets and use dynamic programming to retain for each T the optimal compatible exon set from each C & S (thus creating TCS calls).
metaeuk predictexons contigsDB referenceDB callsResultDB tempFolder --metaeuk-eval 0.0001 -e 100 --min-length 40
Since this step involves a search, it is the most time-demanding of all analyses steps. Upon completion, it will output a database (contigs are keys), where each line contains information about a TCS and its exon (multi-exon TCSs will span several lines).
Optional calling of sub-optimal exon sets:
By default, MetaEuk calls a single and optimal compatible exon set from each C & S for each T. If you are interested in calling several matches to a certain T from each C & S (for example, to look for gene duplications), you can change the default value of
max-exon-sets to the number of sets to look for (from version 6). A few important notes:
max-exon-sets> 1, then it is no longer guaranteed that TCS is a unique identifier. Therefore, when parsing the output of such runs, it is recommended to use TCS together with low_contig as the identifier (see details about the MetaEuk header).
- If I run with
--max-exon-sets> 1, am I guaranteed to get ALL the predictions I get when running
--max-exon-sets 1? No! You most likely see all of them but this is not guaranteed because some complex cases can arise due to the redundancy reduction stage. You can see an example for such a case under tests/sub_opt/readme.txt.
- Running with
max-exon-sets> 1 is mainly useful in case your contigs are long enough to contain several genes (less common in metagenomic data)
If there are homologies in referenceDB (e.g., T1 is highly similar to T2), the same optimal exons set from a C & S combination will be called more than once. This module will group together TCSs that share an exon and will choose their representative prediction. By default, it will greedily obtain a subset of the predictions, such that there is no overlap of predictions on the same contig and strand (to allow same-strand overlaps, run with
metaeuk reduceredundancy callsResultDB predsResultDB predGroupsDB
Upon completion, it will output: predsResultDB and predGroupsDB. predsResultDB contains information about the predictions (same format as callsResultDB). Each line of predGroupsDB maps from a prediction to all TCSs that share an exon with it.
Converting to Fasta (and GFF):
The callsResultDB/predsResultDB produced by the modules above, can be used to extract the sequences of the predicted protein-coding genes.
metaeuk unitesetstofasta contigsDB referenceDB predsResultDB predsResults
It will result in predsResults.fas (protein sequences), predsResults.codon.fas, predsResults.headersMap.tsv and predsResults.gff
The MetaEuk header:
The header is composed of several sections, separated by pipes ('|'):
coord refers to the coordination on the contig (first base has coordinate 0). It is advisable to keep T_acc and C_acc short and without pipes. The exon_coords are of the structure: low[taken_low]:high[taken_high]:nucleotide_length[taken_nucleotide_length]
Since MetaEuk allows for a very short overlap on T of two putative exons (see P2 and P3 in the illustration below), when joining the sequences of the exons, one of them is shortened. The coordinates of the codons taken from this exon will be in square brackets ([taken_low], [taken_high] and [taken_nucleotide_length]). These refer to the orange section of P3 below, while the coordinates outside the brackets refer to the yellow+orange section of P3.
Example header (two exons on the minus strand):
Optionally, by setting the flag
--write-frag-coords 1, information about the position of stop codons will be added to the output. In this case the exon_coords will be given in the following structure:
In its initial stage, MetaEuk extracts putative coding fragments between stop codons. It later discovers exons within them by matching targets. The fragment coordinates in square brackets refer to the original fragment in which the exon was found. In addition to reporting these coordinates, MetaEuk will print the stop codon (
* in the protein output) right at the end of the last exon, if it exists.
The MetaEuk GFF:
In addition to writing a Fasta file, MetaEuk writes a GFF file. Please note that GFF is not perfectly suitable for MetaEuk because MetaEuk doesn't predict non-coding regions. This means that the MetaEuk gene starts and ends where the first and last codons could be matched. The gene and mRNA categories are the same in the MetaEuk GFF. The exon and CDS coordinates will be the same unless a small target overlap was allowed, due to which, the MetaEuk exon was shortened (see above). In this case, the CDS will report the shortening. In the sixth column you can find their individual bitsocres. The contig index starts at 1 and the start coordinate is always smaller than the end coordinate, as required by GFF. The last column contains the TCS identifier. Here is an example where a MetaEuk header of two exons is reported in GFF format:
contig_acc MetaEuk gene 101 912 508 - . Target_ID=protein_acc;TCS_ID=protein_acc|contig_acc|-|low_coord contig_acc MetaEuk mRNA 101 912 508 - . Target_ID=protein_acc;TCS_ID=protein_acc|contig_acc|-|low_coord_mRNA;Parent=protein_acc|contig_acc|-|low_coord contig_acc MetaEuk exon 583 912 234 - . Target_ID=protein_acc;TCS_ID=protein_acc|contig_acc|-|low_coord_exon_0;Parent=protein_acc|contig_acc|-|low_coord_mRNA contig_acc MetaEuk CDS 583 912 234 - . Target_ID=protein_acc;TCS_ID=protein_acc|contig_acc|-|low_coord_CDS_0;Parent=protein_acc|contig_acc|-|low_coord_exon_0 contig_acc MetaEuk exon 101 502 273 - . Target_ID=protein_acc;TCS_ID=protein_acc|contig_acc|-|low_coord_exon_1;Parent=protein_acc|contig_acc|-|low_coord_mRNA contig_acc MetaEuk CDS 101 502 273 - . Target_ID=protein_acc;TCS_ID=protein_acc|contig_acc|-|low_coord_CDS_1;Parent=protein_acc|contig_acc|-|low_coord_exon_1
Creating a TSV map of predictions to their TCS group members:
A TSV file, of lines of the format (low_coord information added in version 6):
can help mapping from each representative prediction after the redundancy reduction stage to all its TCS group members. Since redundancy reduction is performed per contig and strand combination, there will always be agreement in these fields. Note, a representative also maps to itself.
metaeuk groupstoacc contigsDB referenceDB predGroupsDB predGroups.tsv
Taxonomic assignment with taxtocontig:
After obtaining MetaEuk predictions, the taxtocontig workflow allows assigning taxonomic labels to the predicted MetaEuk proteins and confer these predictions to their contigs. This workflow internally runs taxonomy on the MetaEuk predictions, using any
--lca-mode. It then performs majority voting among the taxonomically labeled predictions on a given contig to select a label for the contig. The parameter
--majority indicates the minimal fraction of labeled predictions that agree in their taxonomic assignment (1.0 - consensus, 0.5 - at least 50%, etc.). The contig's label will be the last common ancestor (LCA) of the fraction of labeled predictions in agreement. Please note that MMseqs2 commands are avaialble through MetaEuk.
predictions' taxonomic labels: Ostreococcus tauri, Ostreococcus mediterraneus, unclassified, Bathycoccus prasinos
- contig label (
--majority 0.5): Ostreococcus (genus), the LCA of 2 out of 3 labels
- contig label (
--majority 1): Bathycoccaceae (family), the LCA of 3 out of 3 labels
- The output of a MetaEuk run: contigsDB (if you run MetaEuk with easy-predict you will find it at
<tmpDir>/latest/contigs), predsResults.fas and predsResults.headersMap.tsv, which are produced by the unitesetstofasta module (called by easy-predict).
- A protein sequence database annotated with taxonomic information (seqTaxDb). See details here. You could download such a resource with >88M entries here.
metaeuk taxtocontig <i:contigsDB> <i:predsResults.fas> <i:predsResults.headersMap.tsv> <i:taxAnnotTargetDb> <o:taxResult> <tmpDir> --majority 0.5 --tax-lineage 1 --lca-mode 2
The run ends with two files: taxResult_per_pred.tsv and taxResult_per_contig.tsv, each of which is in taxonomy result TSV format
Compile from source
Compiling MetaEuk from source has the advantage that it will be optimized to the specific system, which should improve its performance. To compile MetaEuk
g++ (4.9 or higher) and
cmake (3.0 or higher) are required. Afterwards, the MetaEuk binary will be located in the
git clone https://github.com/soedinglab/metaeuk.git . mkdir build cd build cmake -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=. .. make -j make install export PATH="$(pwd)/bin/:$PATH"
gcc from Homebrew. The default macOS
clang compiler does not support OpenMP and MetaEuk will not be able to run multithreaded. Use the following cmake call:
CXX="$(brew --prefix)/bin/g++-10" cmake -DCMAKE_BUILD_TYPE=RELEASE -DCMAKE_INSTALL_PREFIX=. ..
MetaEuk will scale its memory consumption based on the available main memory of the machine. MetaEuk needs a CPU with at least the SSE4.1 instruction set to run.