TransAnnot: a fast transcriptome annotation pipeline

TransAnnot is a toolkit designed to predict protein functions, identify orthologous relationships, and decipher biological pathways for newly sequenced transcriptomes. Utilizing MMseqs2's fast sequence-sequence and sequence-profile search, it identifies the closest homologs from reference databases to infer essential details such as protein function, structure, and orthologous groups.

Optionally, TransAnnot can use Plass for transcriptome assembly, enabling de novo assembly of raw sequence reads at the protein level.

TransAnnot is a free and open-source (GPLv3), modular toolkit developed in C++.

Now published in Bioinformatics Advances: https://doi.org/10.1093/bioadv/vbae152

Compile from source

Compiling TransAnnot from the source allows for system-specific optimization. For the compilation cmake, g++ and git are required. After the compilation, TransAnnot will be located in the build/bin directory.

git clone https://github.com/soedinglab/transannot.git
cd transannot && mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=RELEASE -DCMAKE_INSTALL_PREFIX=. ..
make -j 4
make install
export PATH=$(pwd)/bin/:$PATH

❗️ If you compile from source under macOS we recommend installing and using gcc instead of clang as a compiler. gcc can be installed with Homebrew. Force cmake to use gcc as a compiler by running:

CC="$(brew --prefix)/bin/gcc-13"
CCX="$(brew --prefix)/bin/g++-13"
cmake -DCMAKE_BUILD_TYPE=RELEASE -DCMAKE_INSTALL_PREFIX=. ..

Other dependencies for the compilation from the source are zlib and bzip.

Workflow dependencies

• Plass - should be installed separately, see corresponding repository. To perform de novo assembly, it is required to install Plass to the current working directory. Standard usage is running on the results of a nucleotide assembler such as Trinity. PLASS requires read lengths of at least 100 nt, so for shorter reads, a nucleotide assembler has to be used.

• Genome assembly is a dynamic field, so the software is being continously updated. That is why no non-inhouse assemblers (e.g Trinity) are included in the TransAnnot release package. One can install them separately on demand. Some of the tools which might be useful are:

• transdecoder - identifies coding regions within the transcript
• mrna-spades
• Trinity

• Another dependencies are UniProtKB/Swiss-Prot, eggNOG and Pfam databases provided in the MMseqs2 format.

Before starting

tmp folder

tmp folder keeps temporary files. By default, all the intermediate output files from different modules will be kept in this folder. To clear tmp pass --remove-tmp-files parameter.

Quick-ish start

For the fastest results, please consider assembling the data and translating it into amino acid sequences beforehand.

STEP 1: Download default databases

(THIS IS A ONE-TIME PROCESS THAT WILL ONLY HAVE TO BE EXECUTED THE FIRST TIME AFTER TRANSANNOT HAS BEEN DOWNLOADED.)

Download the default databases using transannot downloaddb: transannot downloaddb eggNOG <path_to_output>/<eggNOGDB_name> <eggNOGDB_tmpdir_name> [options] transannot downloaddb Pfam <path_to_output>/<PfamDB_name> <PfamDB_tmpdir_name> [options] transannot downloaddb SwissProt <path_to_output>/<SwissProtDB_name> <SwissProtDB_tmpdir_name> [options]

(The downloads can take quite long depending on the download server loads, so it is advisable to execute these commands inside a windows manager such as tmux or screen.)

STEP 2: Annotating the data

This can be done in one of the following three ways (we strongly recommend option C).

Option A: Starting with sequencing reads

The quickest way to run TransAnnot is by using the easytransannot module:

transannot easytransannot <inputReads.fastq> Pfam-A.full eggNOG UniProtKB/Swiss-Prot <resDB> <tmp> [options]

If (one of the) target databases is already downloaded in MMseqs2 format, directly provide the path to them, otherwise simply specify their names, and the databases will be downloaded automatically. easytransannot uses Plass assembler, for more details check the descriptions for assemblereads module below.

Option B: Starting with assembled nucleotide sequences (NOT RECOMMENDED)

Should a nucleotide assembly already be available (e.g., in <input.fasta>), it can be annotated as follows:

transannot createquerydb <input.fasta> <input_queryDB_name> <tmp> [options]
transannot annotate <input_queryDB_name> Pfam-A.full eggNOG UniProtKB/Swiss-Prot <resDB> <tmp> [options]

(<input_queryDB_name> is just a string providing either the name and additionally the path to the MMseqs2-formatted database for the input sequences.)

We recommend against starting with nucleotide sequences as of the current release because the translated search that TransAnnot relies upon is quite slow.

Option C: Starting with assembled, in silico translated amino acid sequences (RECOMMENDED)

It is far more preferable to translate the assembly with a tool such as TransDecoder prior to annotation with TransAnnot as the searches are very fast in this case. The workflow in this case is identical to the one described above; simply provide the input FASTA file containing the translated amino acid sequences to tranannot createquerydb and then supply the created query DB as input to transannot annotate.

Inputs

Possible inputs are assembled on the protein level:

• assembled transcriptomes (obtained e.g. using Trinity) or raw transcriptome reads, which will be de novo assembled on the protein level using plass
• metatranscriptomes
• single-organism transcriptomes
• TransAnnot can work with the long reads input too. Thus, to enable PLASS assembly, input should be provided as a concatenated single file for single-end reads

Execution

Modules

• assemblereads - de novo assembles raw sequencing reads to large genomic fragments (contigs).
• createquerydb - creates a database in a memory-efficient MMSeqs2 format for the query input sequence.
• downloaddb - downloads reference databases in MMSeqs2 format on which annotations for query sequences will be searched.
• annotate - performs clustering on input sequences to reduce redundancy and runs sequence-profile and sequence-sequence searches for the reference query sequences to obtain the closest homologs with the annotated function. In addition, it maps descriptions of orthologous groups and protein families to the query sequences.
• easytransannot - an easy one-line command module for the complete transannot workflow, starting from input assembly, downloading reference databases to an output of sequence annotations.
• annotatecustom - facilitates annotation against user-supplied databases instead of the default databases used by TransAnnot.

assemblereads

This module uses Plass to assemble input read sequences and obtain translated protein sequences. Plass requires read length of at least 100nt, therefore assemblereads or easytransannot module is applicable only for input reads of length $\ge$ 100nt. To run, Plass must be installed and located in the current working directory. Detailed information about installation can be found here.

In this step, reads will be assembled with Plass and afterwards an MMseqs2 database will be created, you may skip this step if the transcriptome is already assembled and translated. Usage:

transannot assemblereads <inputReads.fastq[.gz|bz]> ... <inputReads.fastq[.gz|bz]> <o: fastaFile with assembly> <o: seqDB> <tmp> [options]

Since Plass has not been benchmarked for transcriptome assembly, for standard usage, we recommend using nucleotide assembler such as [Trinity] (https://github.com/trinityrnaseq/trinityrnaseq/wiki) followed by protein translator (e.g TransDecoder) (https://github.com/TransDecoder/TransDecoder/wiki) before running TransAnnot for sequence annotation. If the input query sequence is obtained from external tools, just proceed directly with createquerydb module.

createquerydb

This module creates a database for input query sequences in memory-efficient MMSeqs2 format.

To create a query database, execute the following command:

transannot createquerydb <inputFastaFile> <sequenceDB> <tmpDir> [options]

If input fasta file is obtained from external tools without using assemblereads module, this createquerydb module must be used. Otherwise, assemblereads module already provides a query sequence database in MMSeqs2 format and hence this step can be skipped.

downloaddb

This module downloads sequence databases for homology searches.

To see detailed information about databases, please use the following command:

mmseqs databases -h

and execute the below command to download the databases (Ensure the same keyword as given in mmseqs database -h):

transannot downloaddb <selection> <outDB> <tmp> [options]  

By default, transannot runs 3 searches in the subsequent annotate module against the following databases: (i) Pfam-A.full (profile database), (ii) eggNOG (profile database) and (iii) UniProtKB/SwissProt (sequence database). Hence, use the above command separately for each database to download them, for more information check MMseqs2 user guide.We use the abovementioned databases for the default annotation workflow to ensure comprehensive set of annotations that include hand-reviewed homologs (SwissProt), fine-grained orthologs (eggNOG), and domains (Pfam-A). On demand, one can use annotatecustom to perform annotation against user-defined database.

downloaddb allows resuming the download of the DB if it's detected in the provided directory path.

annotate

This module extracts representative sequences from the query database using clustering (redundancy-free set) and uses them as search input for 3 transcriptome annotation searches (one sequence-sequence and two sequence-profile). To ensure deep coverage of the transcriptome, TransAnnot only retains non-overlapping hits for each query.

To run the annotate module, execute the following command:

transannot annotate <assembledQueryDB> <path to Pfam profileTargetDB> <path to eggNOG profileTargetDB> <path to SwissProt sequenceTargetDB> <o:resTsvFile> <tmp> [options]

annotatecustom

Annotates against user-specified databases. Run as follows:

transannot annotatecustom <assembledQueryDB> <user-defined DB>

Output

TransAnnot's output is, in general, a tab-separated .tsv file containing the following columns:

queryID targetID qstart qend description E-value sequenceIdentity bitScore typeOfSearch nameOfDatabase 

Where,

• queryID is the sequence identifier for the annotated transcript.
• targetID is the identifier for the sequence/profile from which the annotation is sourced.
• qstart start index of the annotated query.
• qend end index of the annotated query.
• description is the string (e.g., FASTA header from the matched Swiss-Prot sequence) supplying the human-readable annotation (e.g., "XYZ protein").
• E-value is the expectation value for this particular match (combination of query and target sequences) which indicates, loosely, the confidence one can have that the match is real (i.e., due to shared evolutionary history) and not due to chance (lower the E-value the better).
• sequenceIdentity is the number of positions in both sequences that are identical to one another, represented as a fraction of the total (sum) of the sequence length(s). This is a percentage value.
• 'bitScoreis the normalized size (in bits) of the hypothetical database one would have to search against to obtain the match at hand purely by chance (higher thebitScore` the better).
• typeOfSearch is a filed that indicates whether the database searched is a sequence or a profile database.
• nameOfDatabase is a filed indicating the name of the database searched (e.g., "Pfam).

Important options of the annotate module

--simple-output parameter allows users to obtain simplified output for each query sequence, which only includes query and target IDs, a header of the target database and E-value. Whereas standard output contains sequence identity and bit score in addition to details provided in the --simple-output. Usage:

transannot annotate $1 $2 $3 $4 $5 $6 --simple-output 

When no tag is used, standard output will be provided.

--min-seq-id is a parameter to adjust the minimum sequence identity for the searches. The default value is set to 0.3.

--no-run-clust performs annotation without clustering. All the input query sequences will undergo annotation searches.

Use of custom database for annotation

If one is interested in annotation against a user-defined database, annotatecustom module provides such an opportunity. To run the custom annotate module execute the following command:

transannot annotatecustom <assembledQueryDB> <user-defined DB>

The user-provided database will be converted to the MMseqs2 format within the module, but it is also possible to initially provide a MMseqs2-formatted database. A limitation is that unless ID descriptors are included in the database, no mapping can be performed and no group descriptors will be retrieved.

tmp folder

tmp folder keeps temporary files. By default, all the intermediate output files from different modules will be kept in this folder. To clear tmp pass --remove-tmp-files parameter [bool], applicable for all modules except createquerydb and downloaddb.

TransAnnot wiki

The TransAnnot wiki can be found here: https://github.com/soedinglab/transannot/wiki.

Citing TransAnnot

Please cite our publication (https://doi.org/10.1093/bioadv/vbae152) in Bioinformatics Advances.