Interpreting variants in recessive diseases is difficult because clinical severity depends on the combined function of both alleles. Deep mutational scanning (DMS) experiments can provide functional measurements at scale, but their scores often relate nonlinearly to true biochemical activity. Here, we describe a method for inferring enzymatic activities for thousands of variants by running two fitness assays at different expression levels and modeling the nonlinear activity-fitness relationship. These inferred activities allow the computation of a bi-allelic pathogenicity score that captures the joint effect of two alleles. We applied this approach to adenylosuccinate lyase (ADSL), quantifying the effects of >8,000 coding variants in a yeast-based DMS assay. The inferred activities separated pathogenic from benign alleles, and the bi-allelic scores correlated strongly with biochemical measurements and clinical outcomes, outperforming existing predictors. This framework provides a broadly applicable strategy for the mechanistic interpretation of variants in recessive enzymes.
Çubuk et al. (Wed,) studied this question.