Deep Phenotyping for Translational Research and Precision Medicine NIH Symposium: Linking Disease Model Phenotypes to Human Conditions
Total Page:16
File Type:pdf, Size:1020Kb
Deep Phenotyping for Translational Research and Precision Medicine NIH Symposium: Linking Disease Model Phenotypes to Human Conditions Peter Robinson Charit´e Universit¨atsmedizin Berlin September 10–11, 2015 Peter Robinson (Charite)´ Deep Phenotyping 1/50 September 10–11, 2015 1 / 50 Thanks! Matthew Brush Nathan Dunn Melissa Haendel Harry Hochheiser Sebastian K¨ohler Suzanna Lewis Julie McMurry Christopher Mungall Peter Robinson Damian Smedley Nicole Vasilevsky Kent Shefchek Nicole Washington Zhou Yuan 1 http://monarchinitiative.org Peter Robinson (Charite)´ Deep Phenotyping 2/50 September 10–11, 2015 2 / 50 Plan 1 Human Phenotype Ontology (HPO) 2 Ontology Algorithms: The Bare-Bones Basics 3 The Phenomizer 4 The HPO for translational research 5 PhenIX: Clinical Diagnostics in Medical Genetics 6 HPO: Semantic Unification of Common and Rare Disease 7 Pressing Needs and Goals for Future Impact Peter Robinson (Charite)´ Deep Phenotyping 3/50 September 10–11, 2015 3 / 50 Bioinformatics Since the beginnings of the field of Bioinformatics in the 1960s, a central theme has been the development of algorithms that calculate similarity scores between biological entities and use them to rank lists Margaret Dayhoff, originator of PAM matrices BLAST: Find and rank homologous sequences Peter Robinson (Charite)´ Deep Phenotyping 4/50 September 10–11, 2015 4 / 50 Bioinformatics for medicine? But how exactly do we calculate the similarity between diseases, symptoms, patients,:::? Peter Robinson (Charite)´ Deep Phenotyping 5/50 September 10–11, 2015 5 / 50 The Human Phenotype Ontology 11,030 terms 117,348 annotations for ∼ 7000 mainly monogenic diseases http://www.human-phenotype-ontology.org Widely used in rare disease community: UK 100,000 genomes; NIH Undiagnosed Diseases Network; DDD/DECIPHER, GA4GH, etc. Applications: I linking human diseases to animal models I inferring novel drug interactions I prioritizing gene-disease targets I describing rare clinical disorders Interoperable with model organism data and basic research standards A computable representation of human disease Peter Robinson (Charite)´ Deep Phenotyping 6/50 September 10–11, 2015 6 / 50 Why HPO? Substantially better coverage of phenotype concepts than any other terminology Winnenburg and Bodenreider, ISMB PhenoDay, 2014 Peter Robinson (Charite)´ Deep Phenotyping 7/50 September 10–11, 2015 7 / 50 Widely used in the community Databases & Bioinformatics Resources Using HPO DECIPHER (Sanger Institute) DDD (Sanger Institute) ECARUCA FORGE (Genome Canada) GWAS Central IRDiRC ISCA NCBI Genetic Testing Registry NIH Undiagnosed diseases program UK 100,000 Genomes Program UMLS Phenotips (Brudno Group, U Toronto) ::: Major credits go to OMIM and Orphanet Peter Robinson (Charite)´ Deep Phenotyping 8/50 September 10–11, 2015 8 / 50 Plan 1 Human Phenotype Ontology (HPO) 2 Ontology Algorithms: The Bare-Bones Basics 3 The Phenomizer 4 The HPO for translational research 5 PhenIX: Clinical Diagnostics in Medical Genetics 6 HPO: Semantic Unification of Common and Rare Disease 7 Pressing Needs and Goals for Future Impact Peter Robinson (Charite)´ Deep Phenotyping 9/50 September 10–11, 2015 9 / 50 What's The Problem? Phenotypic descriptions that are very evocative for humans but meaningless for computers: I myopathic electromyography I still walking 25 years after onset The following descriptions mean the same thing to you: “generalized amyotrophy”, “generalized muscle atrophy”, “muscular atrophy, generalized” (etc)1 Many publications have little2 information about the actual phenotypic features seen in patients with particular mutations Databases cannot talk to one another about phenotypes Peter Robinson (Charite)´ Deep Phenotyping 10/50 September 10–11, 2015 10 / 50 A tale of two footballs A football ::: A football ::: American Football = Football 6= Football = European Football = Soccer When you see “football”, your computer sees: 0100011001101111011011110111010001100010011000010110110001101100 Peter Robinson (Charite)´ Deep Phenotyping 11/50 September 10–11, 2015 11 / 50 A tale of two fibrillations fibrillation ::: fibrillation ::: muscle fibrillation = fibrillation 6= fibrillation = ventricular fibrillation When you see “fibrillation”, your computer sees: 01100110011010010110001001110010011010010110110001101100011000010111010001101001 0110111101101110 Peter Robinson (Charite)´ Deep Phenotyping 12/50 September 10–11, 2015 12 / 50 What is an Ontology? “An ontology is a specification of a conceptualization.” Tom Gruber, 1993 disjoint inverse part of value Catalog Thesaurus instances ::: restrictions logical Glossary subclassing properties constraints Peter Robinson (Charite)´ Deep Phenotyping 13/50 September 10–11, 2015 13 / 50 Information content Information Content of Clinical Features carnivore n = 1000 IC=0 IC canine feline −log(p) n = 450 n = 550 IC=0.799 IC=0.598 dog wolf fox cat n = 350 n = 60 n = 40 n = 550 IC=1.050 IC=2.813 IC=3.219 IC=0.598 0 1 2 3 4 5 6 7 beagle terrier husky big cat wildcat 1.0 0.8 0.6 0.4 0.2 0.0 n = 40 n = 5 n = 15 n = 250 n = 80 IC=3.219 IC=5.298 IC=4.200 IC=1.386 ic=2.526 p=Frequency of clinical feature cheetah tiger lion n = 100 n = 40 n = 90 IC=2.303 IC=3.219 IC=2.408 IC(t) = − log p(t); Information content of common ancestor: Similarity between ontology terms Average similarity between terms can be used to compare two diseases Peter Robinson (Charite)´ Deep Phenotyping 14/50 September 10–11, 2015 14 / 50 The Human Phenome: Network of Human Diseases and Disease Genes a) Muscular Disorder class Skeletal/Bone/ Bone Connective tissue/ Cancer Development Cardiovascular Neuro Connective Tissue Dermatological Developmental Ear, Nose, Throat Endocrine Gastrointestinal Hematological Heme/ Immuno Immunological Endo/Renal Metabolic Metab Muscular Neurological Nutritional CV Ophtamalogical Psychiatric Renal Respiratory Ophth Skeletal Multiple Cancer Derma b) c) Annotated terms 2 3 Mapped 2to parents 3 Added 50% random terms X X sim(d1; d2) = 0:5 · avg 4 max sim(s; t)5 + 0:5 · avg 4 max sim(s; t)5 Density t2d2 t2d1 s2d s2d 1 Rank of OMIM entry 2 0 10 20 30 40 Peter Robinson (Charite)´ Deep Phenotyping0 10 20 30 4015/50 50 60 September 10–11, 2015 15 / 50 0.0 0.2 0.4 0.6 0.8 1 2 3 4 5 6 Network score Number of terms used FOL: Klingon Opera (8x)(Klingon(x) ) OperaLover(x)) Klingon(W orf) If Klingon(W orf) is true, we can infer that Worf is an opera lover. OperaLover(W orf) Analogous algorithms are the basis for human , model organism comparisons Peter Robinson (Charite)´ Deep Phenotyping 16/50 September 10–11, 2015 16 / 50 What is a phenotype ontology? Precise language (and thinking), interoperability, improved database models to reliably capture and interpret phenotype information. A medical phenotype ontology describes the individual manifestations of diseases: 1 signs 2 symptoms 3 laboratory findings 4 imaging studies 5 etc. Deep phenotype: The precise and comprehensive analysis of phenotypic abnormalities Individual components of disease rather than ”gestalt” Robinson PN, Webber C (2014) Phenotype ontologies and cross-species analysis for translational research. PLoS Genet 10:e1004268. PN Robinson (2012) Deep phenotyping for precision medicine. Hum Mutat 33: 777–780 (Special Issue of Human Mutation on Deep Phenotyping) Peter Robinson (Charite)´ Deep Phenotyping 17/50 September 10–11, 2015 17 / 50 Plan 1 Human Phenotype Ontology (HPO) 2 Ontology Algorithms: The Bare-Bones Basics 3 The Phenomizer 4 The HPO for translational research 5 PhenIX: Clinical Diagnostics in Medical Genetics 6 HPO: Semantic Unification of Common and Rare Disease 7 Pressing Needs and Goals for Future Impact Peter Robinson (Charite)´ Deep Phenotyping 18/50 September 10–11, 2015 18 / 50 Ontological diagnostics Noonan Syndrome Opitz Syndrome a) b) abn. of abn. of abn. of the abn. of the the eye the eye ocular region ocular region abn. of the abn. of the abn. of globe eyelid abn. of globe eyelid localization or size localization or size telecanthus abn. of the abn. of the hypertelorism hypertelorism palpebral fissures palpebral fissures Syndrome term downward slanting downward slanting palpebral fissures palpebral fissures Query term Overlap between query and disease Noonan Syndrome c) 3.78 downward slanting palpebral fissures sim(Q,Noonan) = 3.78 + 3.05 2 = 3.42 Query (Q) 3.05 hypertelorism downward slanting palpebral fissures hypertelorism Opitz Syndrome 3.05 sim(Q,Opitz) =2.45 + 3.05 hypertelorism 2 = 2.75 2.45 (IC of abn. of the eyelid) telecanthus 2 3 X sim(Q ! d) = avg 4 max sim(s; t)5 t2d s2Q Peter Robinson (Charite)´ Deep Phenotyping 19/50 September 10–11, 2015 19 / 50 Q: Query terms d: Disease terms Basic idea of ontological search: Do not need exact match! But semantically similar diseases score well Peter Robinson (Charite)´ Deep Phenotyping 20/50 September 10–11, 2015 20 / 50 The Phenomizer Sebastian K¨ohler et al. (2009) Clinical Diagnostics with Semantic Similarity Searches in Ontologies. Am J Hum Genet, 85:457–64. http://compbio.charite.de/Phenomizer Peter Robinson (Charite)´ Deep Phenotyping 21/50 September 10–11, 2015 21 / 50 Plan 1 Human Phenotype Ontology (HPO) 2 Ontology Algorithms: The Bare-Bones Basics 3 The Phenomizer 4 The HPO for translational research 5 PhenIX: Clinical Diagnostics in Medical Genetics 6 HPO: Semantic Unification of Common and Rare Disease 7 Pressing Needs and Goals for Future Impact Peter Robinson (Charite)´ Deep Phenotyping 22/50 September 10–11, 2015 22 / 50 HPO for translational research Translation Translation Basic Science mechanisms Clinical strategy drugs guidelines