Phenotypic similarity-based approach for variant prioritization for unsolved rare disease: a preliminary methodological report

doi:10.21203/rs.3.rs-2948814/v1

Download PDF

Article

Phenotypic similarity-based approach for variant prioritization for unsolved rare disease: a preliminary methodological report

https://doi.org/10.21203/rs.3.rs-2948814/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 06 Nov, 2023

Read the published version in European Journal of Human Genetics →

You are reading this latest preprint version

Rare diseases (RD) have a prevalence of not more than 1/2000 in the European population, and are characterised by the difficulty of obtaining a correct and timely diagnosis. According to Orphanet, 72,5% of RD have a genetic origin although 35% of them do not yet have an identified causative gene. A significant proportion of patients suspected to have a genetic RD receive an inconclusive exome/genome sequencing. Working towards the International Rare Diseases Research Consortium (IRDiRC)’s goal for 2027 to ensure that all people living with a RD receive a diagnosis within one year of coming to medical attention, the Solve-RD project aims to identify the molecular causes underlying undiagnosed RD. As part of this strategy, we developed a phenotypic similarity-based variant prioritization methodology comparing submitted cases amongst them and with known RD in Orphanet. A 3-step programmatic cascade of phenotypic similarity calculations using The Human Phenotype Ontology (HPO), the Orphanet Rare Diseases Ontology (ORDO) and the HPO-ORDO Ontological Module (HOOM) was developed; genomics data reanalysis was performed by the RD-Connect Genome-Phenome Analysis Platform (GPAP). The methodology was tested in 4 exemplar cases, discussed with experts from European Reference Networks. Variants of interest (pathogenic or likely pathogenic) were detected in 8.8% of the 725 cases clustered by similarity algorithms, formulating diagnostic hypotheses that were validated in 42.1% of them and need further explorations in another 10.9%. Based on the promising results, we are devising an automated standardized phenotypic-based re-analysis pipeline to be applied to the entire unsolved cases cohort in Solve-RD.

Rare diseases (RD) are defined in Europe as those affecting not more than 5 per 10,000 inhabitants in the European population¹. Data from January 2021 show that 4,480 out of 6,171 RD² are genetic RD (72,5%), and 1,588 of them (35%) do not yet have a clearly identified causative gene.³ Rare diseases are characterised by the difficulty in obtaining a correct and timely diagnosis, because of their rarity, scarcity of patients and inequalities in access to expertise.⁴ During their diagnostic journey, patients can receive a clinical diagnosis⁵ (i.e. a name for their disease) or remain undiagnosed. In addition, for genetic disorders, identifying and characterising the underlying molecular basis is crucial for establishing a specific diagnosis and implementing an optimal therapeutic approach.

Solve-RD (Solving the unsolved rare diseases)^6,7 is a European-funded research project which aims to solve unsolved cases defined as those without a molecular diagnosis after undergoing WES (whole exome sequencing), by using different data and re-analysis approaches⁸. Solve-RD builds upon a core group of four European Reference Networks (ERNs: ERN-ITHACA, ERN-RND, ERN-Euro NMD, ERN-GENTURIS) which annually see more than 270,000 RD patients, and that contribute unsolved cases data.

Amongst the approaches explored within this project, we developed a methodology based on phenotypic similarity calculations among unsolved cases and known RD. Indeed, some studies^9–11 have demonstrated that using a phenotypic approach relying on a phenotypic annotation comparison is useful in the diagnosis research. We collect phenomic and genomic data from unsolved RD cases in a standardized and machine-readable format, and run a cascade of one-to-all comparisons to known RD in order to raise diagnostic hypotheses based on re-analysis of candidate causative variants. These are then submitted for further investigation and validation. Ultimately, this leads to the identification of a formerly undescribed disease, or to the identification of unreported manifestations of known RD. The final goal is to return a diagnosis to the patients, by providing them with a name for their disease. This article aims to describe the methodology used in our phenotype similarity-based pipeline by providing illustrative examples of its application.

Patient cohorts

Solve-RD collects phenomics and genomics data from patient cohorts⁷ and releases them as data freezes. The present study employs data selected from the 2020 Data Freeze 1 containing data of 8,730 cases and their unaffected relatives, a case being defined as an index case or a relative case affected by a RD, submitted by ERNs to the RD-Connect Genome-Phenome Analysis Platform (GPAP)¹². This initial dataset contains 1,101 “solved cases” (patients annotated as “solved” in the GPAP platform by the ERN data submitter, the “Solve-RD solved cohort”) and 7,269 “unsolved cases” (patients annotated as “unsolved” by the ERN data submitter, therefore with an inconclusive whole exome/genome sequencing WES/WGS result, the “Solve-RD Cohort 1”).

Ontologies used

To ensure data standardization, Human Phenotype Ontology (HPO)¹³, Orphanet Rare Disease Ontology (ORDO)¹⁴ and HOOM (HPO-ORDO Ontological Module)¹⁵ were used.

HPO provides a standardized vocabulary of phenotypic abnormalities encountered in human disease. For this study, we employed the HPO 2020-10-12 release (including 15,656 unique HPO terms).

ORDO is a structured vocabulary for rare diseases derived from the Orphanet knowledge base including relationships between diseases, genes amongst other features, enabling for computational analysis of rare diseases. This study uses ORDO v3.1, Dec 2020 including 9,338 active clinical entities. Clinical entities in ORDO are designated in this paper by their unique identifier in Orphanet, the ORPHAcode.

HOOM is an ORDO module that qualifies the association between a clinical entity and its HPO-based phenotypic abnormalities according to their frequency. For this study, we used the version 1.5 (Dec 2020) including 1,867,364 ORDO-HPO associations.

Data in ORDO and in HOOM are manually curated and expert-validated.

Data management

The overall Solve-RD data management is described elsewhere^7,8. Briefly, after data submission to the RD-Connect GPAP it is shared within the Solve-RD consortium via the European Genome-Phenome Archive (EGA)¹⁶. We retrieved data for our analysis through a cloud-based central database RD3 (rare disease data about data) using the MOLGENIS open-source data platform¹⁷. RD-Connect GPAP ensures data standardisation, pseudonymisation and harmonisation according to GA4GH-approved standards (Global Alliance for Genomics and Health)¹⁸, in a computer-readable format (Phenopacket)¹⁹ which enables exchange of phenotypic and family information.

Data and models employed

Figure 1

a. Study population:

From the 2020 Freeze 1 (8,730 cases submitted by the four core Solve-RD partnering ERNs) we built our study population by filtering the cases as depicted in Fig. 1A: (a) 644 cases with no causative gene were removed from the solved cases pool and added to the unsolved cases sub-cohort, as cases without a causative gene are considered unsolved in this analysis; (b) cases without phenotype annotations were removed from both populations (72 solved and 3,289 unsolved cases) corresponding to (1) records of patients’ relatives for which there were no phenotypic data (therefore not considered as “cases” in this analysis) and (2) non-annotated cases that are not exploitable for a phenotypic comparison. Cases bearing at least 1 HPO annotation, either patients or affected patients’ relatives, were considered as cases for the purpose of this study. The resulting final population used therefore includes 385 solved cases (4,6% of the initial cohort) and 4,624 unsolved cases (55,2% of the initial cohort).

Reference data:

Reference datasets were obtained after applying the following filtering to the ORDO v3.1 dataset (Fig. 1B). “Groups of disorders” and “inactive clinical entities” were removed. Of the resulting active entities (6,171 disorders and 993 subtypes of disorder) only entities associated with (a) at least one gene (3,789 entities) and (b) at least one HPO term (3,955 entities) were kept. Thus, the resulting 2,259 clinical entities which satisfy (a) and (b) serve as reference data (31.5% of Orphanet active clinical entities).

Phenotypic similarity algorithms

Eight similarity algorithms were compared: Resnik symmetric and asymmetric^20–22, Cosine and Cosine Weighted^21,23 Jaccard and Jaccard weighted^21,23, Bernoulli with grid²⁴and Phenodigm²⁵. Our similarity computation process employs these eight algorithms, packaged in a single JAR executable file, runSolveRD.jar, developed by Köhler et al.²⁶. It computes one-to-all ranked similarities between each case with reference RD and also amongst all cases. Similarity scores range from 0 to 1 (where 1 is the closest point): for each case, case-case and case-disorder associations are ranked by decreasing similarity score. For the purpose of this study, we selected Resnik symmetric due to its best performance²⁶ and the 50 first-ranked results were retrieved, with no limiting similarity thresholds.

Variant prioritisation methods

Programmatic reanalysis of the genomic data was performed in the RD-Connect GPAP as described in Matalonga et al.⁸ .Genomic data from all individuals included in the study was filtered by I) rare variants (MAF < 0.01 according to gnomAD and MAF < 0.02 according to RD-Connect GPAP internal frequency), ii) with a high (truncating) or moderate (amino acid change) impact at the protein level according to Variant Effect Predictor (VEP) from Ensembl and iii) falling within a gene included in the specific gene list per individual generated based on similarity results. Results were annotated with the RD-Connect GPAP annotations¹² including ClinVar and ACMG (InterVar) clinical significances. Variants were prioritized and resulting pathogenic and likely-pathogenic variants according to ACMG guidelines²⁷ were submitted to referring clinicians for final evaluation and discussion during a multidisciplinary meeting.

Work-up design

A 3-step workflow was designed using a solved case as a starting point (triggering case of a phenotypic similarity calculation cascade, represented in Fig. 2). Each step in the workflow generates a cluster around each point of interest in the cascade thus maximizing the one-to-all phenotypic similarity explorations through in-house Python 3.8 scripts.

Step A (“solved case analysis”) aims to check that the solved case phenotype corresponds clinically to the corresponding known RD caused by mutations in the same gene (A1). In case of clinical inconsistency, the similarity algorithm is run in order to detect the first 50 similar disorders; their causative genes are then re-analysed for the triggering case based on data present in RD-Connect GPAP to eventually detect further candidate variants (A2).

Step B (“unsolved around the solved case”) aims to look for candidate pathogenic variants in the solved case-related gene in the 50-first phenotypically similar cases, solved and unsolved, clustered around the triggering case (B1). Variants in genes related with the 50 more phenotypically similar ORPHAcodes clustered around each case are further examined when no variant in the candidate gene is found (B2).

Step C (“unsolved around the ORPHAcode”) aims to look for candidate pathogenic variants in the 50-first phenotypically similar unsolved cases clustered around the ORPHAcode caused by the triggering case’s gene (C1). Variants in genes related with the 50-first phenotypically similar ORPHAcodes clustered around each unsolved case are further examined when no variant in the candidate gene is found (C2).

Cluster visualisation

Cytoscape JS²⁸ was employed to visualize results obtained through the described workflow. This technology is also valuable for exploring cases and/or ORPHAcodes surrounding a case of interest and in driving case selection for diagnosis investigation, hypothesis and analysis. For each step of the workflow (Fig. 2), a specific network is built based on similarity scores between cases or ORPHAcodes. For each case of interest, Cytoscape JS constructs a related subgraph centred on this case including other cases and/or ORPHAcodes (and their associated, genes) from the case’s closest neighbourhood.

Figure 2

Example case selection

To test the relevance of the designed workflow, four solved cases were selected as representative of four different situations:

A solved case matching a known disease caused by the same gene whose phenotypic description was consistent with the latter (PX_8147689, KIF5A-related phenotype).
A solved case matching a known disease caused by the same gene but presenting an unexpected phenotypic variation compared to that of the known disease (PX_2811577, SPAST-related phenotype).
A solved case not matching the expected gene-related ORPHAcode when running similarity calculations (PX_2354306, TBL1XR1-related phenotype).
A solved case that could have matched more than one known disease based on its causative gene (PX_1162604, CASQ1-related phenotype).

The results of the workflow cascade triggered by these cases were discussed with a group of clinicians from participating ERNs (jamboree sessions), and further analyses were performed based on these discussions.

The results obtained following the described 3-step methodology applied to the 4 selected exemplar cases and discussed in phenotypic jamborees is summarized in Table 1.

Table 1

Step A1 was intended to understand if the solved case can receive the clinical diagnosis of the known RD caused by mutations in the gene found as causative. Similarity algorithms were able to find the candidate clinical diagnosis in 3 out of 4 cases within the first 30 results (at positions 11, 19 and 27 for KIF5A-, CASQ1- and SPAST-related cases respectively). TBL1XR1-related ORPHAcode (Pierpont syndrome, ORPHA:487825) could not be found by phenotypic similarity algorithm. After phenotype reannotation, Step A1 was re-run and the correct clinical diagnosis could be found at rank 15.

In Step A2 no other (likely) pathogenic variants were identified in the KIF5A-related case, which shown perfectly consistent with the Orphanet annotation of Autosomal dominant spastic paraplegia type 10 (ORPHA:100991)²⁹ in Step A1. As SPAST- and CASQ1- related cases were not completely consistent with the Orphanet-matching disorders based on their causative gene, we formulated the hypothesis that a double anomaly could be at the origin of the patient’s atypical phenotypic presentation. Step A2 performed for the SPAST-related case suggested that a homozygous deletion in LDHA, causing Glycogen storage disease due to lactate dehydrogenase M-subunit deficiency (ORPHA:284426)³⁰ deserved further investigation of this case that cannot be diagnosed as a Pure spastic paraplegia type 4 (ORPHA:100985).

Only one out of two possible CASQ1 corresponding disorders (ORPHA:88635) was found based on phenotypic similarity. Discussion with clinicians led to the conclusion that, because of lack of perfect consistency between the case and the possible diagnoses, other genes could be involved in the phenotype which resulted not completely solved. A2 step performed for this case retrieved another variant in the TTN gene, classified in ClinVar as “conflicting interpretations of pathogenicity”, that is currently under further investigation.

As a conclusion of Step A, running phenotypic similarity algorithms allowed to unveil phenotypic description quality issues and highlighted the need for more advanced analysis in order to provide a clinical diagnosis before considering cases as totally solved.

Step B1 was intended to use the information from solved cases to help solving the surrounding unsolved ones. Candidate variants were found for three unsolved cases related to SPAST-, TBL1XR1- and KIF5A – triggering solved cases respectively. Of those, after discussion with clinicians only the KIF5A candidate variant (c.226G > C, p.Ala76Pro, (Supplementary Table A)) was a promising non-sense, likely pathogenic, variant. It is located in the part of the gene coding for the protein motor region and it was found in a case at rank 16 of similarity, with a phenotypic description of pure spastic paraplegia. Despite the fact that parent’s DNA is not available and no functional analysis can be carried out, this finding is likely to solve the case.

The SPAST missense variant identified (c.134C > A;p.Pro45Gln, (Supplementary Table A)) in an unsolved case within the SPAST cluster bears conflicting evidence of interpretation in reference databases and it does not segregate in the other symptomatic siblings; therefore, this case remains unsolved.

The variant found in the TBL1XR1 gene (c.1184T > A (p.Tyr395Phe)) for the unsolved case in the TBL1XR1 case phenotypic cluster did not explain the typical Amyotrophic lateral sclerosis (ALS) phenotype of the patient based on the feedback from the clinical expert, although it is described as likely pathogenic, pointing out the need for variant reclassification.

Step B2 further examines step B unsolved cases for prioritised genes involved in phenotypically-related Orphanet disorders. The number of detected variants out of the total number of genes examined is given in Table 1. A variant of interest (c.136G > T (p.Asp46Tyr)) was found in the GALC gene in a both SPAST- and KIF5A- clusters, known to be causative of Krabbe disease (ORPHA:487)³¹, partially explaining patient’s phenotype, but unfortunately additional information on patient’s evolution was not available to confirm or to infirm the diagnostic hypothesis. All the other selected variants were finally discarded because either classified as likely benign or considered not explanatory of the phenotypes by the clinicians.

As a conclusion, Step B results were modest but potentially allowed for two unsolved cases to be explained and reconsider the phenotypes from a new perspective because of the unexpected variants found.

Step C was intended to use the solved case-Orphanet disorder similarity to expand the phenotypic clusters and therefore the genes proposed to be re-examined. In step C1, the causative gene for both the triggering solved case and its related Orphanet disorder was examined. Five SPAST pathogenic variants were found in 5 cases clustering around Autosomal spastic paraplegia type 4 (ORPHA:100985) that were solved in the meantime, and were therefore considered as positive confirmations of our analysis. Five KIF5A candidate variants were found in four unsolved cases in a SPG10 (ORPHA:100991)-centred cluster (Fig. 3). Despite their classification as likely pathogenic in ClinVar, they were discarded because not in the motor region of KIF5A gene, suggesting a variant misclassification in reference databases. In one case, already published as known to present a variant in KIF5A but still unsolved³², that was clinically consistent with SPG10 phenotype, a variant (c.1373C > T, (p.Ser458Phe) was identified and discussed. Clinicians agreed on the diagnostic hypothesis but suggested that the study of further cases is needed before certifying the variant pathogenicity. No variants were found in the CASQ1-related ORPHAcodes centric clusters and it was not possible to perform Step C starting from the TBLXR1 case as at the moment of analysis because no result was found in Step A1.

Figure 3

In Step C2 each unsolved case belonging to the reference ORPHAcode environment, and not explained by Step C1, is reanalysed for the gene causative of its top 50 most similar ORPHAcodes. In the ORPHA:100991 (SPG10) environment, a pathogenic variant in VCP, known to be associated with amyotrophic lateral sclerosis³³ was found for an unsolved case, which appeared consistent with patient’s clinical presentation. As the initial phenotypic description was limited, it was decided to perform Step A after case reannotation. ORPHA:803 (ALS) was then found at rank 24 by the similarity algorithm. A diagnostic confirmation for this case is expected after patient’s re-examination.

In conclusion, Step C identified a number of candidate variants that triggered re-investigation of patients both from a clinical and molecular point of view.

In total, 725 cases (14.5% of the study population) were found by phenotypic similarity in the top-50 rank in at least one of the pipeline steps and were further analysed for variants detection. Variants of interest (pathogenic or likely pathogenic) were found in 64 out of 725 cases (8.8%) thus leading to the formulation of diagnostic hypotheses. These hypotheses were validated for 42.1% (27/64) of those cases, considered as solved. In 7 cases (10.9%) the diagnostic hypotheses raised were considered as relevant by the clinicians, but require additional analyses.

The International Rare Disease Research Consortium (IRDiRC) roadmap 2017–2027 challenged the community with three goals, of which Goal 1: All patients coming to medical attention with a suspected rare disease will be diagnosed within one year if their disorder is known in the medical literature; all currently undiagnosable individuals will enter a globally coordinated diagnostic and research pipeline³. These “currently undiagnosable individuals” are defined in this project as those having undergone an inconclusive WES. The approach proposed in this paper consists in a combination of phenotypic similarity calculations and genomic variant prioritization for re-analysis, based on the hypothesis that using structured, standardized information on solved cases and on known rare diseases could help raising diagnostic hypotheses for unsolved cases. We proposed a 3-step methodology allowing for the exploration of all the clusters emerging from a systematic one-to-all comparison cascade triggered by solved cases. By selecting exemplar cases, and discussing the results with experts participating in the project, we aimed at proposing a systematic phenomic-genomic analytical pipeline that could be implemented at the scale in the future.

The differential diagnosis process attempts to identify candidate diseases that best explain a set of clinical features. This process can be complicated by the fact that clinical features can have varying degrees of specificity, as well as by the presence of features unrelated to the disease itself. Depending on the experience of the physician and the availability of laboratory tests, clinical abnormalities may be described in greater or lesser detail. Some similarity approaches to measure phenotypic similarity features using the Human Phenotype Ontology (HPO) and to combine it with genotypic data to make a diagnosis decision and/or variant or gene prioritization have been proposed³⁴.. For this study, we have closely collaborated with researchers who have both designed HPO¹³ and developed similarity algorithms using phenotype-driven methodology^22–24. Nevertheless, although phenotype-driven is a valuable approach, this study highlights some limitations related to its use. Firstly, HPO annotations were not complete: all cases which are not phenotypically annotated (40% in freeze 1) could not be included in similarity computation, even if the case is associated with an identified causative gene. Furthermore, not every Orphanet entity already has phenotypic annotations, as manually curated annotations are a long, ongoing process, and some could not be used by similarity algorithms.

Secondly, most of RD are multi-systemic disorders, however we have observed that phenotypic annotations are often influenced by physician’s medical specialty, thus introducing a bias as discriminating HPO terms can be missed. In the same way, we have noticed a trend to regularly use same set of HPO terms for numerous cases, which for a number of them are “coarse grain” classified in the ontology to the detriment of lower and specific HPO terms.

26,4% (1,320/5,009) of submitted cases were annotated with less than 5 HPO terms, which constitutes a major limitation, since the best Resnik’s performances are obtained for cases annotated with 10 to 40 HPO terms. This underlines the need to raise awareness on how good-quality deep phenotyping is important for improving the results of this kind of approach. Other approaches such as AI-based ones are also dependent on quality of data to feed automated reasoning. However, producing phenotype annotations is burdensome for clinicians and there is room for testing other approaches (i.e. Natural Language Processing (NLP) and entity recognition in clinical narratives).

Integration of genomic data is key for clinicians and clinical scientists to evaluate and further validate diagnosis hypotheses coming from similarity results. This is why we decided to integrate results from similarity algorithms together with a downstream analysis of the genomic data (WES or WGS) submitted to the project. To our knowledge, this is the first large-scale study linking phenotypic similarity results to genomic data. To enable this type of high-throughput analysis we used big data technologies and built on the programmatic analysis approach of the RD-Connect GPAP. This approach enables to rapidly filter genomic data from thousands of datasets by a specific gene list generated for each case and step as an output of the similarity algorithms. Resulting is a list of candidate variants based on any pre-defined filtering approach, in this case rare variants with a high or moderate impact at the protein level. The whole process is scalable and can be automated. Current limitations lie in data interpretation as the list of candidate variants can be fairly long (> 10 variants per case) in steps comparing up to 50 individuals / disease entities. The elevated number of patients assessed made it impossible for clinicians to evaluate all variants, therefore we had to restrict the submission of results to variants classified as likely pathogenic or pathogenic according to ACMG criteria, missing variants of uncertain significance that could be re-classified as pathogenic after expert evaluation.

Amongst the cases found by phenotypic similarity algorithms, variants of interest were identified in 8.8% leading to diagnostic hypotheses. Accessibility to patients and family data will in these cases be one of the major issues in the validation of the hypothesis provided and thus the final diagnosis for the patient. In 46.8% (30/64) of cases, hypotheses were formulated but a conclusion could not be raised for various reasons such as inconsistency between the phenotype patient’s phenotype and the hypothesis formulated, difficulties in recalling patients, or the detection of misclassified variants. Indeed, misclassifications in ClinVar and/or ACMG guidelines were observed for two hypotheses.

We tested our approach by using four triggering selected cases and submitting the overall methodology and results to a group of clinicians during four jamborees. While not scalable because of logistical reasons, jamborees allowed clinicians with expertise in different fields to discuss cases and to debate phenotypic associations. Indeed, the combination of a semi-automated variant prioritisation pipeline based on phenotypic similarity calculation with the expertise of clinicians from ERNs proved to be an added-value approach. These preliminary results underline the value of providing clinicians submitting standardized information on unsolved cases with an integrated phenomics-genomics pipeline at-hand, delivering diagnostic hypotheses in a user-friendly way, so as they can use it to make decisions based on their expertise.

Our current developments using Cytoscape JS are intended to drive case selection for diagnosis investigation, hypothesis and analysis, and will be presented in a future publication. In short, the service provides a large condensed similarity network which can be narrowed down around the case of interest, with several filtering options building subgraphs. This will assist experts in making diagnostic hypotheses by having at hand a network of all the HPO annotated cases and all the known RD. Therefore, based on the preliminary results presented in this article, we are devising an automated pipeline (Fig. 4), applied to the entire unsolved cases cohort in Solve-RD.

Figure 4

ACKNOWLEDGEMENTS

The authors would like to thank the data submitters of our triggered cases, Vincenzo Nigro, Rebecca Schule, Laurence Faivre, and all the clinicians who attended to the 4 jamborees for their participation and data interpretation.

This study was supported by the European Reference Network(s) [ERNs: ERN-ITHACA, ERN-RND, ERN-Euro NMD, ERN-GENTURIS] (https://ec.europa.eu/health/ern/networks_en).

The authors are grateful to Charlotte Rodwell who contributed to the organisation of the work presented in this article as well as proof-reading.

FUNDING

The Solve-RD project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 779257.

This study makes use of data shared/provided through RD-Connect, which received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement No. 305444.

COMPETING INTERESTS

The authors declare no competing interests

ETHICAL APPROVAL

All individuals were recruited via four ERNs. Inclusion criteria were a clinical rare disease diagnosis in at least one family member by one of the associated expert centers and a not conclusive exome or genome analysis at time of submission. We did not exclude anyone based on sex, gender, ethnicity, race, age or any other socially relevant groupings.

Each patient entry was associated with its submitting investigator or clinician and linked to its corresponding ERN or UDP. The responsibility of checking the data was suitable for submission to the RD-Connect GPAP and Solve-RD lay with the data submitter as required by their Code of Conduct (institution: Fundació Centre de Regulació Genòmica) and Data Sharing Policy (institution: Solve-RD general assembly), respectively. In some cases, individuals had to be re-consented prior to data submission. This study adheres to the principles set out in the Declaration of Helsinki.

European Union. Regulation (EC) N°141/2000 of the European Parliament and of the Council of 16 December 1999 on orphan medicinal products. 2000 [Internet]. [cited 30 Nov 2022]. Disponible sur: https://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2000:018:0001:0005:en:PDF
Nguengang Wakap, S. et al. Estimating cumulative point prevalence of rare diseases: analysis of the Orphanet database. Eur. J. Hum. Genet. 28, 165–173 (2020).
Zanello, G., Chan, C. H. & Pearce, D. A. Recommendations from the IRDiRC Working Group on methodologies to assess the impact of diagnoses and therapies on rare disease patients. Orphanet J. Rare Dis. 17, (2022).
Recommendations From The Rare 2030 - Foresight Study The Future Of Rare Diseases Starts Today - February 2021 - Rare2030.Eu/Recommendations
Work Package 5 Specifications for an integrated coding application with Orphacodes. www.rd-action.eu.
https://solve-rd.eu.
Zurek, B. et al. Solve-RD: systematic pan-European data sharing and collaborative analysis to solve rare diseases. Eur. J. Hum. Genet. 29, 1325–1331 (2021).
Matalonga, L. et al. Solving patients with rare diseases through programmatic reanalysis of genome-phenome data. Eur. J. Hum. Genet. 29, 1337–1347 (2021).
Tumiene, B., Graessner, H., Mathijssen, I. M., Pereira, A. M. & Schaefer, F. 13 & Jean-Yves Blay 14,15 & Helene Dollfus 16,17 & Nicoline Hoogerbrugge. Maurizio Scarpa 12, 19.
Fujiwara, T., Yamamoto, Y., Kim, J. D., Buske, O. & Takagi, T. PubCaseFinder: A Case-Report-Based, Phenotype-Driven Differential-Diagnosis System for Rare Diseases. Am. J. Hum. Genet. 103, 389 (2018).
Hartley, T. et al. New Diagnostic Approaches for Undiagnosed Rare Genetic Diseases. (2020) doi:10.1146/annurev-genom-083118.
Laurie, S. et al. The RD-Connect Genome-Phenome Analysis Platform: Accelerating diagnosis, research, and gene discovery for rare diseases. Hum. Mutat. (2022) doi:10.1002/HUMU.24353.
Robinson, P. N. et al. The Human Phenotype Ontology: a tool for annotating and analyzing human hereditary disease. Am. J. Hum. Genet. 83, 610–615 (2008).
WHAT IS THE ORPHANET RARE DISEASE ONTOLOGY (ORDO)? www.orphadata.orgwww.orpha.net.
WHAT IS HOOM (THE HPO-ORDO ONTOLOGICAL MODULE)? www.orpha.net (2019).
Lappalainen, I. et al. The European Genome-phenome Archive of human data consented for biomedical research. Nat. Genet. 47, 692–695 (2015).
van der Velde, K. J. et al. MOLGENIS research: advanced bioinformatics data software for non-bioinformaticians. Bioinformatics 35, 1076–1078 (2019).
https://www.ga4gh.org.
Phenopackets - Concepts and Technology. http://phenopackets.org/.
Resnik, P. Using Information Content to Evaluate Semantic Similarity in a Taxonomy.
Pesquita, C., Faria, D., Falcão, A. O., Lord, P. & Couto, F. M. Semantic similarity in biomedical ontologies. PLoS Comput. Biol. 5, (2009).
Köhler, S. et al. Clinical Diagnostics in Human Genetics with Semantic Similarity Searches in Ontologies. Am. J. Hum. Genet. 85, 457–464 (2009).
Köhler, S. Improved ontology-based similarity calculations using a study-wise annotation model. Database 2018, (2018).
Bauer, S., Köhler, S., Schulz, M. H. & Robinson, P. N. Bayesian ontology querying for accurate and noise-tolerant semantic searches. Bioinformatics 28, 2502–2508 (2012).
Smedley, D. et al. PhenoDigm: analyzing curated annotations to associate animal models with human diseases. Database J. Biol. Databases Curation 2013, (2013).
Köhler, S. & Havrylenko, S. Target Dissemination Level Public1 Author(s). 7.
Richards, S. et al. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genet. Med. Off. J. Am. Coll. Med. Genet. 17, 405–424 (2015).
Franz, M. et al. Cytoscape.js: a graph theory library for visualisation and analysis. Bioinformatics 32, 309–311 (2016).
Reid, E. et al. A Kinesin Heavy Chain (KIF5A) Mutation in Hereditary Spastic Paraplegia (SPG10). Am. J. Hum. Genet. 71, 1189 (2002).
Maekawa, M., Sudo, K., Kanno, T. & Li, S. S.-L. Molecular characterization of genetic mutation in human lactate dehydrogenase-A (M) deficiency. Biochem. Biophys. Res. Commun. 168, 677–682 (1990).
Sakai, N. et al. Krabbe Disease: Isolation and Characterization of a Full-Length cDNA for Human Galactocerebrosidase. Biochem. Biophys. Res. Commun. 198, 485–491 (1994).
Daud, D. et al. Whole exome sequencing and the clinician: we need clinical skills and functional validation in variant filtering. J. Neurol. 262, 1673–1677 (2015).
Johnson, J. O. et al. Exome Sequencing Reveals VCP Mutations as a Cause of Familial ALS. Neuron 68, 857–864 (2010).
Jacobsen, J. O. B. et al. Phenotype‐driven approaches to enhance variant prioritization and diagnosis of rare disease. Hum. Mutat. 43, 1071–1081 (2022).
Gong, W., Kwak, I.-Y., Pota, P., Koyano-Nakagawa, N. & Garry, D. J. DrImpute: imputing dropout events in single cell RNA sequencing data. BMC Bioinformatics 19, 220 (2018).

Table 1 is available in the Supplementary Files section.

There is no duality of interest

Table1.docx
Table 1 Summarized results of the 3-step workflow starting from the 4 selected cases of interest. Step description, input data, results of the similarity algorithm analysis, and candidate variants retrieved are reported for each performed step.
Additionnalinformations.docx
Descriptionofdocument.docx
TableAAllstepsresults.xlsx
TableBFeaturesdescriptionofthevariantsanalysesinexcelfiles.xlsx

Download PDF

Journal Publication

published 06 Nov, 2023

Read the published version in European Journal of Human Genetics →

Editorial decision: revise
12 Jul, 2023
Review #1 received at journal
09 Jul, 2023
Review #2 received at journal
23 Jun, 2023
Reviewer #4 agreed at journal
09 Jun, 2023
Reviewer #3 agreed at journal
09 Jun, 2023
Reviewer #2 agreed at journal
09 Jun, 2023
Reviewer #1 agreed at journal
09 Jun, 2023
Reviewers invited by journal
09 Jun, 2023
Submission checks completed at journal
22 May, 2023
First submitted to journal
19 May, 2023
Unknown event
18 May, 2023
Editor assigned by journal
17 May, 2023

You are reading this latest preprint version

Phenotypic similarity-based approach for variant prioritization for unsolved rare disease: a preliminary methodological report

Status:

Journal Publication

Version 1

Abstract

Figures

INTRODUCTION

MATERIALS AND METHODS

a. Study population:

RESULTS

DISCUSSION

Declarations

ACKNOWLEDGEMENTS

FUNDING

COMPETING INTERESTS

ETHICAL APPROVAL

References

Table 1

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1