Annocript: a flexible pipeline for the annotation of transcriptomes able to identify putative long noncoding RNAs

Abstract

Summary: The eukaryotic transcriptome is composed of thousands of coding and long non-coding RNAs (lncRNAs). However, we lack a software platform to identify both RNA classes in a given transcriptome. Here we introduce Annocript, a pipeline that combines the annotation of protein coding transcripts with the prediction of putative lncRNAs in whole transcriptomes. It downloads and indexes the needed databases, runs the analysis and produces human readable and standard outputs together with summary statistics of the whole analysis.

Availability and implementation: Annocript is distributed under the GNU General Public License (version 3 or later) and is freely available at https://github.com/frankMusacchia/Annocript.

Contact:  

1 Introduction

RNA-Seq has effectively portrayed the transcriptional complexity in eukaryotes demonstrating the widespread transcription of lncRNAs in a diverse group of organisms. However, annotation of denovo generated transcriptomes, remains a complex task requiring efficient management of large datasets. Current tools for annotation identify coding sequences by similarity searches (Conesa et al., 2005; Koski et al., 2005; Philipp et al., 2012; Schmid and Blaxter, 2008) while lncRNAs are typically predicted relying on sequence composition, lack of conservation and lack of ORFs (Kong et al., 2007; Lin et al., 2011; Liu et al., 2006; Wang et al., 2013). To fulfill the need of a single platform, we developed Annocript, a pipeline linking together these processes. It uses multiple programs to annotate sequences based on similarity and composition. It is capable of assigning protein names, domains, metabolic pathways (Morgat et al., 2012), Enzyme classes (Bairoch, 2000), GO terms (Ashburner et al., 2000) and other annotations. Further, it can identify putative lncRNAs based upon lack of any protein/domain similarity, lack of long ORFs and high non-coding potential. By default, a transcript is classified as lncRNA if it satisfies all the following conditions: (i) length ≥200 nucleotides, (ii) lack of similarity with any protein, domain and other short ncRNA from Rfam and rRNAs, (iii) longest open reading frame (ORF) < 100 amino acids and (iv) non-coding potential score ≥0.95.

2 Implementation

Annocript will annotate any FASTA containing a transcriptome given as input. There is no limit in the number of sequences it can analyze. The pipeline runs on Linux systems and requires the presence of (tested versions): Perl (5.10), BioPerl (1.6) (Stajich et al., 2002), Python (2.7.3), R (3.1.0) and MySQL (5.5). A single configuration file allows the user to alter programs parameters or run only specific analysis steps. Annocript performs the following analysis steps.

2.1 Database creation

This step downloads the sequence and annotation databases required for similarity searches: UniRef or TrEMBL coupled with SwissProt (UniProt Consortium, 2013), Rfam (Burge et al., 2012) and Conserved Domains Database (CDD) (Marchler-Bauer et al., 2013). The UniRef/TrEMBL and SwissProt FASTA headers are used to build a MySQL database containing information on taxonomy, protein id, name and description. The downloaded FASTA are indexed for BLAST searches (Camacho et al., 2009). Annocript also downloads ID mappings to associate UniProt and Pfam IDs to Enzymes (Bairoch 2000), GO and pathways into the MySQL database. The database construction requires a few hours on high memory machines. A pure Perl hash table is created to build relations between the tables before to store them into the database. This usage of Perl hash tables is a memory-intensive task that permits to build the database quickly (Table 1). Nevertheless, the pipeline can be run on low-memory machines with as little as 4 GB RAM using a different approach and longer computational time. By default Annocript requires 20 GB to build the hash table and this would be sufficient to build a database using UniRef90 + SwissProt. In our tests the peak amount of RAM used by the hash when loading the entire UniProt knowledge-base (TrEMBL + SwissProt, August 2014), is 44 GB. The database creation step is needed only for the first run of Annocript or when the user needs to update it.

Table 1.

Time required to build the internal MySQL database

	L	M	S	XS
TrEMBL + SwissProt	9 h	24 h	NA	NA
UniRef90 + SwissProt	5 h	11 h	40 h	120 h

Note: Time to create the MySQL database from TrEMBL + SwissProt or UniRef90 + SwissProt (April 2014) using machines with different number of cores and RAM: 24 cores/98GB (L), 24 cores/24GB (M), 8 cores/8GB (S), 2 cores/4GB (XS).

2.2 Programs execution

This step is responsible for the execution of the programs. It performs the following similarity searches: BLASTX against TrEMBL/UniRef and SwissProt, RPSBLAST against CDD profiles, BLASTN against Rfam and rRNAs. By default, Annocript uses speed-optimized BLASTX and BLASTP parameters and a custom parallelization of RPSBLAST to achieve a faster execution. The dna2pep (Wernersson et al., 2006) and Portrait softwares (Arrial et al., 2009) are used to identify, respectively, the longest ORF and the non-coding potential of each query sequence.

2.3 Results parsing and statistics

This step integrates all the results generated in previous steps building a comprehensive tab delimited text file and other standard outputs. The pipeline collects the best-hit and related annotations from each search. The output table contains specific columns for every search executed. In this way, the user can decide to choose a specific classification for cases of uncertainty based on scores or lengths or any other criteria. The columns of the file represent assigned proteins, domains, GO terms, Enzymes, pathways, short and ribosomal RNAs, longest ORF size and non-coding potential. Information about the score, position, coverages and combination of domains are also provided. The table provides a binary classification for each transcript to be a protein coding or a lncRNA. Annocript also generates FASTA files with the predicted lncRNAs, protein sequences from the longest ORFs and GFF3 files of complete BLAST outputs. Finally, a summary of the annotation is given as a HTML-formatted document. Examples of whole transcriptome annotations made by Annocript can be downloaded from http://bit.ly/15vnALW.

3 Optimization

To reduce the computational time, Annocript uses the following parameters in BLASTX and BLASTP searches: threshold = 18 and word_size = 4 (Korf 2003). We show that this customization does not affect the results on two sample datasets: transcripts from a species present in UniProt and from a species not present in UniProt at the time of analysis. The two datasets contain, respectively, 1500 coding sequences from Homo sapiens and 1500 transcripts from Acipenser naccarii (Vidotto et al., 2013). Annocript was executed for each species with BLAST default and Annocript optimized BLAST parameters. The results for human were identical while for the sturgeon, using optimized parameters, out of 899 transcripts annotated in the BLAST default run, 891 (99%) were identical, 3 (0.003%) were missing and 5 matched on different, but homolog hits. At a negligible cost in sensitivity, the computational time was reduced more than 5-folds (Table 2).

Table 2.

Time required to run searches with default and optimized parameters

Organism	BLASTX SP def	BLASTX SP cust	BLASTX Uf def	BLASTX Uf cust	RPSBLAST CDD def	RPSBLAST CDD cust
Human	50.2 min	10.9 min	33.4 h	5.2 h	9.4 h	67.5 min
Sturgeon	12.8 min	2.07 min	8.2 h	54.3 min	2.7 h	14.4 min

Note: Analysis performed on human coding transcripts (avg. length 4200 bp) and A. naccarii transcripts (avg. length 1200 bp) using 24 cores (2.67 GHz) and 24 GB RAM and default BLAST parameters (def) or Annocript-optimized parameters (cust).

4 Discussion

Annocript is a pipeline for the annotation of transcriptomes. The pipeline is optimized to run quickly without bargaining on accuracy. It is the first platform with the ability to annotate coding genes along with the prediction of putative lncRNAs without the need for a reference genome or comparative supporting data.

Acknowledgements

The authors acknowledge the laboratories of Luigia Santella and Jong Tai Chun, Graziano Fiorito, Mariella Ferrante and Gabriele Procaccini for their feedback. S.B. has been funded by the Stazione Zoologica under the ARC PhD program of The Open University, UK.

Funding

RITMARE Flagship Project, Progetto Premiale StarTrEgg.

Conflict of Interest: none declared.

Author notes

^†The authors wish it to be known that, in their opinion, the first two authors should be regarded as Joint First Authors.

Associate Editor: John Hancock