ARSDA: A New Approach for Storing, Transmitting and Analyzing Transcriptomic Data

Abstract

Two major stumbling blocks exist in high-throughput sequencing (HTS) data analysis. The first is the sheer file size, typically in gigabytes when uncompressed, causing problems in storage, transmission, and analysis. However, these files do not need to be so large, and can be reduced without loss of information. Each HTS file, either in compressed .SRA or plain text .fastq format, contains numerous identical reads stored as separate entries. For example, among 44,603,541 forward reads in the SRR4011234.sra file (from a Bacillus subtilis transcriptomic study) deposited at NCBI's SRA database, one read has 497,027 identical copies. Instead of storing them as separate entries, one can and should store them as a single entry with the SeqID_NumCopy format (which I dub as FASTA+ format). The second is the proper allocation of reads that map equally well to paralogous genes. I illustrate in detail a new method for such allocation. I have developed ARSDA software that implement these new approaches. A number of HTS files for model species are in the process of being processed and deposited at http://coevol.rdc.uottawa.ca to demonstrate that this approach not only saves a huge amount of storage space and transmission bandwidth, but also dramatically reduces time in downstream data analysis. Instead of matching the 497,027 identical reads separately against the B. subtilis genome, one only needs to match it once. ARSDA includes functions to take advantage of HTS data in the new sequence format for downstream data analysis such as gene expression characterization. I contrasted gene expression results between ARSDA and Cufflinks so readers can better appreciate the strength of ARSDA. ARSDA is freely available for Windows, Linux. and Macintosh computers at http://dambe.bio.uottawa.ca/ARSDA/ARSDA.aspx.

Keywords: ARSDA; novel storage solution; quantifying expression of paralogous genes; sequence format; transcriptomics.

Figure 1

User interface in ARSDA. (A) The menu system, with database creation under the “Database” menu, gene expression characterization under the “Analysis” menu, etc. (B) Converting a FASTQ/FASTA file to a FASTQ+/FASTA+ file. (C) Site-specific read quality visualization. (D) Global read quality visualization.

Figure 2

Contrasting read quality between two transcriptomic data files (SRR5484239.sra from M. musculus and SRR922267.sra from E. coli. It does not imply that E. coli data are always better than mouse data as there are also poor-quality E. coli data and high-quality mouse data.

Figure 3

Allocation of shared reads in a gene family with three paralogous genes A, B, and C with three idealized segments with a conserved identical middle segment, strongly homologous first segment that is identical in B and C, and a diverged third segment. Reads and the gene segment they match to are of the same color.

Figure 4

Contrast in gene expression (RPKM) between ARSDA and Cufflinks output for the same transcriptomic data in file SRR1536586.sra for E. coli wild type.

Figure 5

Phylogenetic relationship among paralogous genes cspA to cspI in E. coli, based on coding sequences, with bootstrap values next to internal nodes. Sequences were aligned by MAFFT (Katoh and Toh 2008) with accurate L_INS-i option and a maximum of 16 iterations. Coding sequences were first translated in amino acid sequences, which are aligned with BLOSUM62 matrix. Nucleotide sequences were then aligned against aligned amino acid sequences. Phylogenetic analysis was done with PhyML (Guindon et al. 2010). All these analyses were automated in DAMBE (Xia 2013, 2017).