Computational evaluation of exome sequence data using human and model organism phenotypes improves diagnostic efficiency.

William P Bone,Nicole Vasilevsky,Michele Nehrebecky,Cynthia J Tifft,Damian Smedley,Lynne A Wolfe,Camilo Toro,Joie Davis,Thomas C Markello,Michael Brudno,Catherine Groden,Murat Sincan,Christopher J Mungall,David R Adams,Orion J Buske,Heather Trang,Gretchen Golas,Sebastian Köhler,William A Gahl,Elise D Flynn,Peter N Robinson,Rena A Godfrey ,Amanda Links ,Nicole Washington ,Julius O B Jacobsen ,David D Draper ,Marta Gı̂rdea ,Cornelius F Boerkoel ,Ariane Soldatos ,Melissa Haendel ,Colleen E Wahl

doi:10.1038/gim.2015.137

Abstract

Purpose:Medical diagnosis and molecular or biochemical confirmation typically rely on the knowledge of the clinician. Although this is very difficult in extremely rare diseases, we hypothesized that the recording of patient phenotypes in Human Phenotype Ontology (HPO) terms and computationally ranking putative disease-associated sequence variants improves diagnosis, particularly for patients with atypical clinical profiles.Genet Med 18 6, 608–617.Methods:Using simulated exomes and the National Institutes of Health Undiagnosed Diseases Program (UDP) patient cohort and associated exome sequence, we tested our hypothesis using Exomiser. Exomiser ranks candidate variants based on patient phenotype similarity to (i) known disease–gene phenotypes, (ii) model organism phenotypes of candidate orthologs, and (iii) phenotypes of protein–protein association neighbors.Genet Med 18 6, 608–617.Results:Benchmarking showed Exomiser ranked the causal variant as the top hit in 97% of known disease–gene associations and ranked the correct seeded variant in up to 87% when detectable disease–gene associations were unavailable. Using UDP data, Exomiser ranked the causative variant(s) within the top 10 variants for 11 previously diagnosed variants and achieved a diagnosis for 4 of 23 cases undiagnosed by clinical evaluation.Genet Med 18 6, 608–617.Conclusion:Structured phenotyping of patients and computational analysis are effective adjuncts for diagnosing patients with genetic disorders.Genet Med 18 6, 608–617.

Highlights

The National Institutes of Health (NIH) established the Undiagnosed Diseases Program (UDP) to provide explanations to individuals with elusive disorders and to improve understanding of rare and common disorders
The clinical phenotypes associated with the Human Gene Mutation Database (HGMD) variant were specified as Human Phenotype Ontology (HPO) terms
For imperfect phenotypic profiles, Exomiser ranked the causative variant as the top hit for 94–96% when using all available frequency data and for 90–92% when using only Exome Sequencing Project (ESP) frequency data

Summary

Introduction

The National Institutes of Health (NIH) established the Undiagnosed Diseases Program (UDP) to provide explanations to individuals with elusive disorders and to improve understanding of rare and common disorders. Exome sequencing identifies thousands of exomic variations relative to the human reference sequence, and this often precludes generation of a tractable number of diagnostic hypotheses. This problem remains after filtering for predicted deleteriousness, conservation, and Mendelian modes of inheritance.[4] Many variants may even be true loss-of-function mutations tolerated in the average human genome.[5,6] because the UDP often has DNA from only the proband or a few family members, or sometimes a few unrelated patients for each disease being studied, traditional linkage and cohort analyses are not possible

Methods

Results

Conclusion