Read on Twitter
We made an exome diagnosis in a rare disease patient this month using a total of 2 PE sequencing reads, several hundred bps beyond the nearest coding region. Interesting anecdote but highlighted a couple of concepts i thought worth sharing. 1/n
We talk a lot these days about MDTs in genomics in the UK, mostly from the perspective of making sense of analytical output. The value of expert clinical INPUT, however, can’t be overstated in terms of contribution to analytical sensitivity of tests. 2/n
Expert clinical (and, in this case of a paediatric leukodystrophy, radiological) input can focus granular analysis and facilitate challenging diagnoses missed by standard large panel and gene-agnostic approaches. Here, we had a strong suspicion of HEMS https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4479525/ 3/n
This focused suspicion on one single large intron in PLP1. Most pipelines do some sort of hard or soft filtering of intronic variants, ours effectively doesn’t go beyond 100bp of NC material. However, with a genome alignment, deeper intronic variants are still aligned. 4/n
In this case, one of the previously reported pathogenic variants was visible in both reads of a locus covered to 2X. 2 reads seems intuitively completely unreliable, however, when we expanded our ROI, it was actually called by our GATK-based pipe. 5/n
The standard 20 or 30X threshold inferred to give “reliable” haplotype SNV calling in most short-read DNA experiments is actually phenomenally conservative in consanguineous and (as in this case) X-linked contexts. 6/n
As @AMeynert et al showed on what was presumably a now fairly obsolete pipe, sensitivity for homozygous and homozygous SNVs is >80% with just 2 reads, and >95% with 3. Specificity reaches its asymptote at 5 or 6. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3695811/figure/F1/ 7/n
We also know from experience that the vast majority of diagnoses in consanguineous patients and families with XL inheritance will be of this variety - yet I’ve never really seen anyone re-calibrating their read-depth/quality assumptions in this context 8/n
For these cases, the critical threshold for coverage might actually be >1X. Soft-filtering calls aggressively to reduce signal-noise at very low coverage may be essential in some circumstances but risks missing achievable diagnoses. /end
(Addendum) Also, to illustrate the point about great clinical input sometimes trumping technical power in terms of sensitivity, this diagnosis was missed by a large high profile WGS project, presumably at >>depth. More of a needle in a WGS haystack without the phenotyping.
