When evaluating how constrained a gene is, it is essential to consider the CI when using O/E. In research and clinical interpretation of Mendelian cases, pLI > 0.9 has been widely used for filtering. Accordingly, the Gnomad team suggests using the upper bound of the O/E confidence interval LOEUF < 0.35 as a threshold if needed.
Please see the Methods section below for more information about how the scores were calculated.
The pLI and Z-scores of the deviation of observed variant counts relative to the expected number are intended to measure how constrained or intolerant a gene or transcript is to a specific type of variation. Genes or transcripts that are particularly depleted of a specific class of variation (as observed in the gnomAD data set) are considered intolerant of that specific type of variation. -Z-scores are available for the missense and synonynmous categories and pLI scores are available for +Z-scores are available for the missense and synonymous categories and pLI scores are available for the loss-of-function variation.
Missense and Synonymous: Positive Z-scores indicate more constraint (fewer observed variants than expected), and negative scores indicate less constraint (more observed variants than expected). A greater Z-score indicates more intolerance to the class of variation. Z-scores were generated by a sequence-context-based mutational model that predicted the number of expected rare (< 1% MAF) variants per transcript. The square root of the chi-squared value of the deviation of observed counts from expected counts was multiplied by -1 if the observed count was greater than the expected and vice versa. For the synonymous score, each Z-score was corrected by dividing by the standard deviation of all synonymous Z-scores between -5 and 5. For the missense scores, a mirrored distribution of all Z-scores between -5 and 0 was created, and then all missense Z-scores were corrected by dividing by the standard deviation of the Z-score of the mirror distribution.