Computational approaches to interpreting genomic sequence variation

Abstract

Identifying sequence variants that play a mechanistic role in human disease and other phenotypes is a fundamental goal in human genetics and will be important in translating the results of variation studies. Experimental validation to confirm that a variant causes the biochemical changes responsible for a given disease or phenotype is considered the gold standard, but this cannot currently be applied to the 3 million or so variants expected in an individual genome. This has prompted the development of a wide variety of computational approaches that use several different sources of information to identify functional variation. Here, we review and assess the limitations of computational techniques for categorizing variants according to functional classes, prioritizing variants for experimental follow-up and generating hypotheses about the possible molecular mechanisms to inform downstream experiments. We discuss the main current bioinformatics approaches to identifying functional variation, including widely used algorithms for coding variation such as SIFT and PolyPhen and also novel techniques for interpreting variation across the genome.

Figure 1

A set of annotation terms used to describe the potential effects of sequence variants according to the genic regions they fall in and their allele sequences. The terms are drawn from the Sequence Ontology and are depicted on the molecules they are predicted to affect. Variants categorized as any of the terms 2, 4, 9 and 10 are often collectively referred to as ‘loss-of-function’ variants, and are typically expected to severely affect gene function [25].

Figure 2

A sequence logo for the transcriptional factor CTCF derived from binding site predictions from Ensembl on human chromosome 22. The height of the letters represents information content at each position. For example, if a particular nucleotide is always found at a given position, it will have the maximal height and information content, while if a position has all four nucleotides at equal frequencies, it will have a minimal height and no information content. One instance of a motif alignment is shown, which contains a variant at a high information position (boxed). The alternative allele at this position, A, results in a sequence more different from the motif represented by the PWM as measured by the motif score.

Figure 3

A protein multiple alignment for the human GALP gene built from the SIFT alignment pipeline. Color intensity corresponds to conservation in each column. Two variants that are predicted to alter the amino acid sequence (A/V and Y/H) are indicated by arrows and their SIFT scores are presented. Note that SIFT scores ≤0.05 are predicted to be deleterious and other scores are predicted to be tolerated.