Variant prioritization and selection of candidate genes relevant to pregnancy loss
We used Slivar19 to prioritize and filter variants based on modes of inheritance (e.g., compound heterozygous, de novo , autosomal dominant and x-linked recessive). Slivar integrates population allele frequencies from the Trans-Omics for Precision Medicine (TopMED)20 and spliceAI scores into a comprehensive variant filtering strategy to identify candidate genes.19 Details on variant prioritization and exploratory analyses of variants relevant to recurrent pregnancy loss are provided in Online Supplement and Table S2 . We evaluated SNVs across the families by modes of inheritance and highest impact on genes (in-frame deletion/insertion, missense [nonsynonymous], frameshift, stop gained, splice region).
Given the potential for identifying false positive germline SNVs due to DNA quality (e.g., prioritization of false positive autosomal dominant SNVs that differ by orders of magnitude from SNVs following other modes of inheritance19) and overwhelming majority of variants of unknown significance, we applied several approaches to interpret our main findings. First, we selected SNVs identified in any of the pregnancy losses but not live births within our data to interpret candidate genes relevant to recurrent pregnancy loss. Second, we interpreted rare (AF<.001) compound heterozygous SNVs, autosomal recessive variants, where both parents are heterozygous for the variant and the affected offspring receives two copies. We prioritized compound heterozygous SNVs that were identified in losses within our data but not observed in healthy controls (gnomeAD21) to highlight variants in haplosufficient genes relevant to embryonic/fetal lethality. Third, among SNVs that were identified in any of the pregnancy losses, we selected pathogenic SNVs (SNVs with pLI>0.90 and LOEUF<0.36) to highlight potentially damaging variants in candidate genes. Finally, we selected SNVs in genes that were involved in pregnancy loss-relevant phenotypes/diseases (e.g., embryonic/fetal death and developmental abnormalities22-24) to interpret candidate genes. Analyses were performed using Slivar and R, utilizing resources and support from the Center for High Performance Computing at the University of Utah.