CoChat
Sign Up
Cited by 106Genomics and Rare DiseasesGenetic Associations and EpidemiologyArtificial Intelligence in Healthcare and Education

The DOI matches its Crossref record; 33 of the 34 references it cites were found in a registry or an index.

Xrare: a machine learning method jointly modeling phenotypes and genetic evidence for rare disease diagnosis

Open access
Li, Qigang, Zhao, Keyan, Bustamante, Carlos D., Ma, Xin, Wong, Wing H.·2019·Genetics in Medicine
DOI 10.1038/s41436-019-0439-8PMC PMC6752318
Open this paper in CoChat — free accountOpen PDF

Abstract

PURPOSE: Despite the successful progress next-generation sequencing technologies has achieved in diagnosing the genetic cause of rare Mendelian diseases, the current diagnostic rate is still far from satisfactory because of heterogeneity, imprecision, and noise in disease phenotype descriptions and insufficient utilization of expert knowledge in clinical genetics. To overcome these difficulties, we present a novel method called Xrare for the prioritization of causative gene variants in rare disease diagnosis. METHODS: We propose a new phenotype similarity scoring method called Emission-Reception Information Content (ERIC), which is highly tolerant of noise and imprecision in clinical phenotypes. We utilize medical genetic domain knowledge by designing genetic features implementing American College of Medical Genetics and Genomics (ACMG) guidelines. RESULTS: ERIC score ranked consistently higher for disease genes than other phenotypic similarity scores in the presence of imprecise and noisy phenotypes. Extensive simulations and real clinical data demonstrated that Xrare outperforms existing alternative methods by 10-40% at various genetic diagnosis scenarios. CONCLUSION: The Xrare model is learned from a large database of clinical variants, and derives its strength from the tight integration of medical genetics features and phenotypic features similarity scores. Xrare provides the clinical community with a robust and powerful tool for variant prioritization.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41436-019-0439-8

Cites

33 of 33 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Schematic overview of the Xrare model and most important features for predicting variant pathogenicity. a Schematic overview of the Xrare model. The collected 49,021 pathogenic ClinVar variants were divided into three parts in terms of their publishing years: 41,590 variants identified by 2011 used for implementing American College of Medical Genetics and Genomics and the Association for Molecular Pathology (ACMG/AMP) guideline evidence; 6576 variants spiked in synthetic genomes for model training; the remaining 855 variants identified since 2016 used for model evaluation. The preexisting in silico computation scores of variants, population-level scores, and other ACMG evidence scores were used as features for machine learning. Phenotype-related features came from gene–phenotype similarity Emission-Reception Information Content (ERIC) score and predicted gene–phenotype associations. b The top 15 most important features from Xrare model. See Supplementary Figure S1 for all features.
Fig. 1 Schematic overview of the Xrare model and most important features for predicting variant pathogenicity. a Schematic overview of the Xrare model. The collected 49,021 pathogenic ClinVar variants were divided into three parts in terms of their publishing years: 41,590 variants identified by 2011 used for implementing American College of Medical Genetics and Genomics and the Association for Molecular Pathology (ACMG/AMP) guideline evidence; 6576 variants spiked in synthetic genomes for model training; the remaining 855 variants identified since 2016 used for model evaluation. The preexisting in silico computation scores of variants, population-level scores, and other ACMG evidence scores were used as features for machine learning. Phenotype-related features came from gene–phenotype similarity Emission-Reception Information Content (ERIC) score and predicted gene–phenotype associations. b The top 15 most important features from Xrare model. See Supplementary Figure S1 for all features.CC BY-NC-ND · PubMed Central
Fig. 2 Case illustrations and performance evaluation of Emission-Reception Information Content (ERIC) score. a Case illustrations demonstrating ideas of ERIC score. Four different phenotype cases (I to IV) are shown as examples. A–E are phenotype terms in the phenotype ontology tree, and R is the root. A and B are phenotype terms in comparison. C is the most informative common ancestor (MICA) of phenotypes A and B in cases I, II, and III. In case IV, both phenotypes A and E are far from the outside of the circle with the center of D (the MICA of A and E) and the radius of IC(D), thus E is considered as a noisy phenotype of A by ERIC score. b Performance evaluation of phenotype similarity measures for prioritizing target genes. Resnik, JC, Lin, and ERIC scores were compared. We simulated phenotypes with various imprecision and noise levels, e.g., 2|2|4 represents two true precise phenotypes in the Human Phenotype Ontology (HPO) database, two imprecise phenotypes, and four random noise phenotypes. The top panel represents phenotype sets simulated with imprecision. The middle panel represents phenotype sets simulated with noise. The bottom panel represents phenotype sets with both imprecision and noise. The x-axis is the rank percentile of target genes among 8643 OMIM genes. The y-axis is the cumulative distribution function (CDF) of the rank percentile.
Fig. 2 Case illustrations and performance evaluation of Emission-Reception Information Content (ERIC) score. a Case illustrations demonstrating ideas of ERIC score. Four different phenotype cases (I to IV) are shown as examples. A–E are phenotype terms in the phenotype ontology tree, and R is the root. A and B are phenotype terms in comparison. C is the most informative common ancestor (MICA) of phenotypes A and B in cases I, II, and III. In case IV, both phenotypes A and E are far from the outside of the circle with the center of D (the MICA of A and E) and the radius of IC(D), thus E is considered as a noisy phenotype of A by ERIC score. b Performance evaluation of phenotype similarity measures for prioritizing target genes. Resnik, JC, Lin, and ERIC scores were compared. We simulated phenotypes with various imprecision and noise levels, e.g., 2|2|4 represents two true precise phenotypes in the Human Phenotype Ontology (HPO) database, two imprecise phenotypes, and four random noise phenotypes. The top panel represents phenotype sets simulated with imprecision. The middle panel represents phenotype sets simulated with noise. The bottom panel represents phenotype sets with both imprecision and noise. The x-axis is the rank percentile of target genes among 8643 OMIM genes. The y-axis is the cumulative distribution function (CDF) of the rank percentile.CC BY-NC-ND · PubMed Central
Fig. 3 Performance evaluation on synthetic genomes of patients from various simulation scenarios. a, b Without specifying inheritance mode; (c, d) specifying inheritance mode; (a, c) for known genes; (b, d) for novel genes. Phen-Gen was not included in (a) and (b) for comparison because it required specifying inheritance mode. Lines with triangle dots represent methods utilizing both phenotype and genotype information (Xrare, Xrare_ERIC_RP, Exomiser PhenIX, Exomiser_hiPHIVE, eXtasy, and Phen-Gen), while lines with round dots represent prediction methods using only genotype information (Xrare_ACMG, MutationTaster, CADD, and REVEL). Xrare_ERIC_RP and Xrare_ACMG represent the Xrare model using only the phenotype feature ERIC_RP and only the genotype feature ACMG_score, respectively. CDF cumulative distribution function.
Fig. 3 Performance evaluation on synthetic genomes of patients from various simulation scenarios. a, b Without specifying inheritance mode; (c, d) specifying inheritance mode; (a, c) for known genes; (b, d) for novel genes. Phen-Gen was not included in (a) and (b) for comparison because it required specifying inheritance mode. Lines with triangle dots represent methods utilizing both phenotype and genotype information (Xrare, Xrare_ERIC_RP, Exomiser PhenIX, Exomiser_hiPHIVE, eXtasy, and Phen-Gen), while lines with round dots represent prediction methods using only genotype information (Xrare_ACMG, MutationTaster, CADD, and REVEL). Xrare_ERIC_RP and Xrare_ACMG represent the Xrare model using only the phenotype feature ERIC_RP and only the genotype feature ACMG_score, respectively. CDF cumulative distribution function.CC BY-NC-ND · PubMed Central
Fig. 4 Performance evaluation on two real exome sequencing clinical data sets. a Hard-to-solve clinical cases14 with known genes on the left panel and novel genes on the right panel. b Data used in performance evaluation of clinical expert–driven pipeline.15 Lines with triangle dots represent methods utilizing both phenotype and genotype information, while lines with round dots represent prediction methods using only genotype information. CDF cumulative distribution function.
Fig. 4 Performance evaluation on two real exome sequencing clinical data sets. a Hard-to-solve clinical cases14 with known genes on the left panel and novel genes on the right panel. b Data used in performance evaluation of clinical expert–driven pipeline.15 Lines with triangle dots represent methods utilizing both phenotype and genotype information, while lines with round dots represent prediction methods using only genotype information. CDF cumulative distribution function.CC BY-NC-ND · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 34 references

Cited by

Sorted by citations

Items 101–103 of 103

Related

See all

Metadata

Authors
Li, Q., Zhao, K., Bustamante, C. D., Ma, X., & Wong, W. H.
Affiliations
Stanford University (US)
Corresponding author
  • Ma, Xin
  • Wong, Wing H.
ORCID
Venue
Genetics in Medicine
Publisher
Elsevier BV
Published
2019-09
Volume
21
Issue
9
Pages
2126-2134
ISSN
1098-3600 (ISSN-L)
DOI
10.1038/s41436-019-0439-8
PMC
PMC6752318
OpenAlex
W2911737122
Cited by
106
References
34
Citations by year
2026: 10 · 2025: 13 · 2024: 27 · 2023: 12 · 2022: 21 · 2021: 8 · 2020: 13 · 2019: 2
FWCI
6.63
Type
journal article
Language
en
Open access
Licence
Locations
  • Genetics in Medicine (publishedVersion, cc-by-nc-nd)
  • PubMed Central (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
Funders
  • National Institutes of Health (P50-HG007735)
  • National Human Genome Research Institute (P50HG007735, R01HG007834)
  • National Cancer Institute (P20CA096470)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Genomics and Rare Diseases
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Genetics › Genetic Associations and Epidemiology
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
Keywords
rare disease diagnosis, monogenic disorders, variant prioritization, ACMG guidelines, genetic characteristics, medical genetics
MeSH terms
Machine Learning, Genetic Testing, Genotype, Humans, Mutation, Phenotype, Software, Genetic Variation, Computational Biology, Genomics, Rare Diseases, High-Throughput Nucleotide Sequencing, Exome

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review