New mutations, old statistical challenges

Abstract

New mutations, old statistical challengesBased on targeted sequencing of 208 genes in 11,730 neurodevelopmental disorder cases, Stessman et al. report the identification of 91 genes associated (at a False Discovery Rate [FDR] of 0.1) with autism spectrum disorders (ASD), intellectual disability (ID), and developmental delay (DD)-including what they characterize as 38 novel genes, not previously reported as connected with these diseases 1 .If true, this would represent a substantial step forward.Unfortunately, each of the two discovery analyses (1.De novo mutation analysis and, 2. a comparison of private mutations with public control data) contain critical statistical flaws.When one accounts for these problems, fewer than half of the genes--and very few, if any, of the novel findings--survive.These errors have implications for how future analyses should be conducted, for understanding the genetic basis of these disorders, and for genomic medicine.We discuss the two main analyses in turn and provide more detailed treatment of the issues in a supplementary technical note.1. Two-stage analysis of de novo mutations.The authors selected 208 genes, consisting of 130 with one or more de novo truncating mutations in prior published studies, along with 78 others belonging to related pathways or having related Mendelian disease association.Of these genes, 19 have an already documented 'genome-wide significant' excess of de novo mutations from 5-6,000 neurodevelopmental disorder patients used as the "discovery sample" by the authors.Moreover, a more recently published exome dataset 2 has convincingly elevated the number of formally genome-wide significant genes to 93.