We are having discussions locally about Missense constraint use in gnomAD v4.1. We are aware that the Missense constraint in v4.1 is thought to still be experimental from previous posts however I have a specific question about Synonymous Z-scores and how to use them generally with TTN as a specific example.
In gnomAD v2.1.1 TTN has a Missense constraint Z score of -1.1 with a Synonymous Z score of -0.17. I would interpret this as the calculation of expected variants working as intended as the o/e in both Synonymous and Missense variants is approximately 1.
In gnomAD v4.1 TTN has a Missense contraint Z score of 6.98 with a Synonymous Z score of 6.94. I would interpret this as the calculation of expected variants NOT working as intended as the o/e in both Synonymous and Missense variants is approximately 0.9.
If I was to hazard a guess at what is happening here I would think that there are coverage issues, sequencing of pseudogenes or related sequences or issues with the calculations of expected variants here that affects both the Synonymous and Missense variants equally. I would therefore think it likely that TTN is not now actually missense constrained.
Comparing this to a genuinely missense constrained gene such as SCN1A, the Synonymous and Missense metrics are hugely different (o/e 0.89 and 0.56 respectively).
Can I confirm that this interpretation is correct and that you should be wary of using Missense constraint metrics when the Synonymous constraint metrics are also significantly different?
Thank you for your help with this matter, apologies if this is covered in some other resources.
This is a complex question, and I’ll respond to your inquiry in pieces.
First, for this point:
In gnomAD v4.1 TTN has a Missense contraint Z score of 6.98 with a Synonymous Z score of 6.94.
Z-scores are sensitive to sample size. For more information, please see our blog post on gene constraint.
To elaborate a bit here, part of the reason why the Z-scores for TTN in gnomAD v4.1 are so large is likely due to the number of variants included in these calculations. For example, the TTN missense observed/expected (oe) ratio was computed using 40245 observed and 44309.4 expected variants in gnomAD v4.1. In comparison, the gnomAD v2 TTN missense oe calculation included 18060 observed and 17647.99 expected variants. The much larger counts in v4.1 means that slight differences between the observed and expected counts look much more significant.
I would therefore think it likely that TTN is not now actually missense constrained.
TTN had a missense oe of 1.02 in gnomAD v2 and 0.91 in gnomAD v4.1. Neither of these numbers indicates that TTN is particularly constrained against missense variation.
Can I confirm that this interpretation is correct and that you should be wary of using Missense constraint metrics when the Synonymous constraint metrics are also significantly different?
In general, for either version of the constraint metric (within either v2 or v4.1 but not comparing between v2 and v4.1), we caution against overinterpretation of missense or pLoF constraint metrics for genes that have unusual synonymous constraint metrics.