On the necessity of dissecting sequence similarity scores into segment-specific contributions for inferring protein homology, function prediction and annotation
1 Bioinformatics Institute (BII), Agency for Science, Technology and Research (A*STAR), 30 Biopolis Street, #07-01, Matrix, Singapore 138671, Singapore
2 School of Biological Sciences (SBS), Nanyang Technological University (NTU), 60 Nanyang Drive, Singapore 637551, Singapore
3 Department of Biological Sciences (DBS), National University of Singapore (NUS), 8 Medical Drive, Singapore 117597, Singapore
4 School of Computer Engineering (SCE), Nanyang Technological University (NTU), 50 Nanyang Drive, Singapore 637553, Singapore
BMC Bioinformatics 2014, 15:166 doi:10.1186/1471-2105-15-166Published: 2 June 2014
Protein sequence similarities to any types of non-globular segments (coiled coils, low complexity regions, transmembrane regions, long loops, etc. where either positional sequence conservation is the result of a very simple, physically induced pattern or rather integral sequence properties are critical) are pertinent sources for mistaken homologies. Regretfully, these considerations regularly escape attention in large-scale annotation studies since, often, there is no substitute to manual handling of these cases. Quantitative criteria are required to suppress events of function annotation transfer as a result of false homology assignments.
The sequence homology concept is based on the similarity comparison between the structural elements, the basic building blocks for conferring the overall fold of a protein. We propose to dissect the total similarity score into fold-critical and other, remaining contributions and suggest that, for a valid homology statement, the fold-relevant score contribution should at least be significant on its own. As part of the article, we provide the DissectHMMER software program for dissecting HMMER2/3 scores into segment-specific contributions. We show that DissectHMMER reproduces HMMER2/3 scores with sufficient accuracy and that it is useful in automated decisions about homology for instructive sequence examples. To generalize the dissection concept for cases without 3D structural information, we find that a dissection based on alignment quality is an appropriate surrogate. The approach was applied to a large-scale study of SMART and PFAM domains in the space of seed sequences and in the space of UniProt/SwissProt.
Sequence similarity core dissection with regard to fold-critical and other contributions systematically suppresses false hits and, additionally, recovers previously obscured homology relationships such as the one between aquaporins and formate/nitrite transporters that, so far, was only supported by structure comparison.