DivA: detection of non-homologous and very divergent regions in protein sequence alignments

BACKGROUND: Sequence alignments are used to find evidence of homology but sometimes contain regions that are difficult to align which can interfere with the quality of the subsequent analyses. Although it is possible to remove problematic regions manually, this is non-practical in large genome scale studies, and the results suffer from irreproducibility arising from subjectivity. Some automated alignment trimming methods have been developed to remove problematic regions in alignments but these mostly act by removing complete columns or complete sequences from the MSA, discarding a lot of informative sites.

FINDINGS: Here we present a tool that identifies Divergent windows in protein sequence Alignments (DivA). DivA makes no assumptions on evolutionary models, and it is ideal for detecting incorrectly annotated segments within individual gene sequences. DivA works with a sliding-window approach to estimate four divergence-based parameters and their outlier values. It then classifies a window of a sequence of an alignment as very divergent (potentially non-homologous) if it presents a combination of outlier values for the four parameters it calculates. The windows classified as very divergent can optionally be masked in the alignment.

CONCLUSIONS: DivA automatically identifies very divergent and incorrectly annotated genic regions in MSAs avoiding the subjective and time-consuming problem of manual annotation. The output is clear to interpret and allows the user to take more informed decisions for reducing the amount of sequence discarded but still finding the potentially erroneous and non-homologous regions.