The Genome Reverse Compiler: an explorative annotation tool

Abstract

Background: As sequencing costs have decreased, whole genome sequencing has become a viable and integral part of biological laboratory research. However, the tools with which genes can be found and functionally characterized have not been readily adapted to be part of the everyday biological sciences toolkit. Most annotation pipelines remain as a service provided by large institutions or come as an unwieldy conglomerate of independent components, each requiring their own setup and maintenance.

Results: To address this issue we have created the Genome Reverse Compiler, an easy-to-use, open-source, automated annotation tool. The GRC is independent of third party software installs and only requires a Linux operating system. This stands in contrast to most annotation packages, which typically require installation of relational databases, sequence similarity software, and a number of other programming language modules. We provide details on the methodology used by GRC and evaluate its performance on several groups of prokaryotes using GRC's built in comparison module.

Conclusion: Traditionally, to perform whole genome annotation a user would either set up a pipeline or take advantage of an online service. With GRC the user need only provide the genome he or she wants to annotate and the function resource files to use. The result is high usability and a very minimal learning curve for the intended audience of life science researchers and bioinformaticians. We believe that the GRC fills a valuable niche in allowing users to perform explorative, whole-genome annotation.

Figure 1

Procedure for gene calls. Gene calling procedure for GRC. Starting with all ORFs (set M), BLAST information and the generic EDPs are used to make an initial evaluation of coding and non-coding. ORFs determined to be coding go into set C and those ORFs that overlap them go into set L. The EDPs are retrained to be organism specific and are used to remove the low-scoring ORFs from M to create M'. Overlaps in M' are then resolved to create gene calls.

Figure 2

Start site determination. Support can come from different alignments for various start sites. Alignment 1 supports s1, s2, and s3; Alignment 2: s1 and s2; Alignment 3: s1, s2, s3, and s4. Higher start scores are given to start sites that: occur closer to a supporting alignment, occur at a higher frequency, and are supported by a higher scoring alignment.

Figure 3

Evaluate functional assignment using GO. Here the term "intracellular part" represents a reference function assigned to the reference gene. The terms "intracellular", "membrane", and "DNA helicase complex" represent possible GRC GO term assignments and their evaluation with respect to the reference term.

Figure 4

GRC pipeline. Internal pipeline for GRC. Maximal ORFs are found and translated. FSA-BLAST is run using the user specified database and the resulting alignments are used to call and annotate protein coding genes.

Figure 5

Gene finding performance. Performance of gene finding at increasing minimum gene length in comparison to Glimmer. Precision and Sensitivity tends to increase for each organism as the minimum gene length is increased. Lengths are 100, 150, 200, 250, and 300 bp. Panels a, b, c show the data with identical scaling. Panels d, e, f show details of Glimmer comparison for the E. coli, Pseudo, and Gamma groups. Note: The symbols for panels a, b, c match the symbols and legends of their detailed counterparts.

Figure 6

Start site determination. Performance of gene finding at increasing minimum gene length with respect to the fraction of TP with correct start sites (Start Precision) and Sensitivity. Panels a, b, c show the data with identical scaling. Panels d, e, f show details of Glimmer comparison for the E. coli, Pseudo, and Gamma groups. Note: The symbols for panels a, b, c match the symbols and legends of their detailed counterparts.

Figure 7

Performance on functional assignment. Columns show the average fraction of true positive ORFs with confirmed, compatible, and incompatible term assignments. These fractions are not additive since a TP can have a confirmed, compatible, and incompatible term assignment.

Figure 8

Running time. Total running time of GRC versus the total search space (product of total query and DB length). The main bottleneck for GRC is BLAST.