GREGoR: accelerating genomics for rare diseases

Abstract

Rare diseases are collectively common, affecting approximately 1 in 20 individuals worldwide. In recent years, rapid progress has been made in rare disease diagnostics due to advances in next-generation sequencing, development of new computational and functional genomics approaches to prioritize genes and variants and increased global sharing of clinical and genetic data. However, more than half of individuals suspected to have a rare disease lack a genetic diagnosis. The Genomics Research to Elucidate the Genetics of Rare Diseases (GREGoR) Consortium was initiated to study thousands of challenging rare disease cases and families and apply, standardize and evaluate emerging genomics technologies and analytics to accelerate their adoption in clinical practice. Furthermore, all data generated, currently representing over 7,500 individuals from over 3,000 families, are rapidly made available to researchers worldwide through the Analysis, Visualization and Informatics Lab-space (AnVIL) to catalyse global efforts to develop approaches for genetic diagnoses in rare diseases. Most of these families have undergone previous clinical genetic testing but remained unsolved, with most being exome-negative. Here we describe the collaborative research framework, datasets and discoveries comprising GREGoR that will provide foundational resources and substrates for the future of rare disease genomics.

The past decade has seen rapid progress in clinical genetics due to increased discovery of genes and variants involved in Mendelian diseases and ongoing advances in sequencing, variant analysis and data sharing^1–5. Despite this progress, most individuals who undergo clinical genetic testing for a suspected Mendelian condition remain undiagnosed^6–9. For example, in the National Human Genome Research Institute (NHGRI) Centers for Mendelian Genomics (CMG)^10–12, while over 3,800 genes were implicated in Mendelian disease, only about 11,000 out of over 28,000 families received a confirmed or potential molecular diagnosis¹. Thus, considerable challenges remain to increase the molecular diagnostic yield and explain currently unsolved rare genetic disorders (Box 1).

Box 1. Challenges in diagnosing rare genetic diseases.

(1). Undiscovered disease genes.

The pathogenic variant(s) may be located in a gene that is yet to be implicated in disease. Until now, over 5,000 protein-coding genes have been implicated in at least one disease, but it is estimated that still 10,000+ disease–gene relationships are undiscovered in just the remaining protein-coding genes¹²⁸.

(2). Understanding non-coding variation.

Candidate pathogenic variant(s) may be located in the non-coding genome, where the mechanisms for how a variant manifests a clinical phenotype are not well understood.

(3). Current technological limitations.

The variant may be in a region that is unattainable to ascertain or difficult to detect from solely short-read, exome or genome sequencing such as long repeats, inversions and CGRs. The variant may be detectable but bioinformatic algorithms may struggle to call the variant correctly such as multi-nucleotide and mosaic variants. The variant may be detectable and called correctly but asserting its functionality or pathogenicity may require unavailable, orthogonal evidence.

(4). Complex inheritance patterns.

MPV, oligogenic, polygenic, complex compound heterozygosity, variable expressivity, incomplete penetrance, allelic heterogeneity, imprinting, epimutations, maternal effect and/or mosaicism may also be confounding a diagnosis and necessitate a broader approach to understanding the disease mechanism.

(5). Lag time in curation.

A gene–disease relationship may be published or submitted to a genetic database but has yet to be reviewed and incorporated into clinical testing. This is compounded by rapidly increasing numbers of VUSs.

(6). Costs and implementation.

The costs associated with both current and new genomic technologies, as well as the expertise required for their implementation in clinical laboratories for scaled testing, present obstacles to widespread adoption in both research and clinical settings. The need for specialized training, regulatory compliance and expensive infrastructure further limits accessibility, particularly in resource-limited settings.

(7). n = 1.

Many candidate variants and genes are n = 1 regardless of best data-sharing practices.

(8). Limitations in experimentation.

Owing to the nature of novel discovery, there is not always a functional assay available to provide orthogonal evidence for or against a candidate variant or gene. Often if a candidate does meet the inclusion criteria for an existing assay, the molecular phenotype measured in the assay may not match the potential mechanism of disease or fully recapitulate the pathophysiological impact, resulting in ambiguous results or an incorrect prediction of pathogenicity.

(9). Bias.

Databases predominantly capture genetic information from individuals of European genetic ancestry potentially propagating biases in tools and reference data for variant classification for individuals of non-European genetic ancestry¹¹².

(10). Contribution of each challenge.

The relative contribution of each of these challenges to the overall diagnostic gap is difficult to quantify and based mostly on hypotheses, process of elimination and extrapolation. Understanding which challenges have the greatest impact—and in which contexts—remains key to prioritizing solutions effectively.

(11). High burden of proof.

Newer genomic technologies may offer advantages over short-read DNA sequencing, but effective and widespread use of new technologies requires clear guidance and broad demonstration of scaled efficacy.

In 2021, at the end of the CMG, the NHGRI launched GREGoR with five primary research sites and a data-coordinating centre to accelerate rare disease genetic research by harnessing the latest advances in sequencing including and especially genome sequencing and multi-omics; evaluating and prioritizing the use of functional genomics and new computational strategies including recent advances in artificial intelligence; translating advances into routine clinical testing; advancing data sharing to foster a quorum of evidence for discovery; and collaborating worldwide to continue discovery and reporting of genetic aetiologies for Mendelian diseases (https://gregorconsortium.org/data) (Fig. 1). Compared with other rare disease consortia worldwide, GREGoR is unique in its mission to generate and rapidly disseminate internationally a variety of genomic data for rare disease families and evaluate, validate and scale emerging technologies aimed at achieving molecular diagnoses in cases that remain unresolved despite conventional clinical genetic testing.

Fig. 1 |. Overview of GREGoR.

Strategic framework of the GREGoR consortium for accelerating genomics in rare-disease research, highlighting cross-cutting themes, systematic data generation, computational innovations and end points of success.

Evaluating emerging methods

Squeezing the exome

The most impactful approach to date for diagnosing rare diseases has been exome sequencing and periodic reanalysis of the protein-coding sequences in the human genome^13–17. GREGoR has led or contributed to 83 papers studying molecular diagnoses in 365 genes with more than a third being novel disease gene discoveries or phenotypic expansions (Supplementary Table 1) and provided a variety of automated pipelines for large cohort, exome and genome reanalysis, which include phenotypic integration^14–16. Reanalysis success is largely driven by new disease gene discoveries and phenotypic expansions since the original analysis¹⁵; however, new tools focused on reanalysis of well-known disease genes and loci have continued to yield diagnostic successes¹⁸.

Furthermore, GREGoR has developed new computational approaches to further increase the molecular diagnostic yield from exomes. For example, difficulty in phasing short-read sequencing can confound diagnoses of pathogenic compound heterozygous variants for recessive diseases. To overcome this, GREGoR contributed to a highly accurate method for inferring phase and has calculated and released all pairwise phasing estimates and usage guidance for rare coding variants in exomes occurring in the same gene through the Genome Aggregation Database (gnomAD)^19,20. GREGoR researchers have developed tools to identify and implicate hundreds of pathogenic structural variant diagnoses from existing unsolved exomes^21–25. Combined, GREGoR’s continuous efforts to extract diagnoses from existing exomes demonstrate potential for continued innovation in genomic reanalyses.

Short-read genome sequencing

GREGoR has published a framework for genetic testing when panel or exome sequencing is inconclusive²⁶. The next step is typically short-read genome sequencing (srGS). However, most molecular diagnoses deduced by srGS are found in protein-coding genes, suggesting that they could potentially be detected by exome sequencing. By contrast, srGS offers the opportunity to profile a wide spectrum of variants within and beyond the exome, especially for structural variants, such as large deletions, duplications, insertions, inversions, translocations and complex structural variants that involve a combination of two or more variant classes^27–30. To evaluate the relative utility of srGS, a large-scale study of 822 families by GREGoR researchers reported 218 patients who received a diagnosis through srGS, where 72% of variants should have been detectable by exome sequencing³¹. The remaining 28% of cases were explained by variants that are not readily accessible on exomes such as tandem repeat expansions, deep intronic variants, structural variants and variants in difficult-to-sequence coding regions. Overall, srGS resulted in a greater than 8% increase in diagnostic yield compared with just exome sequencing and underscored the growing support for using srGS as a first-tier test.

Large-scale srGS has also provided the opportunity to evaluate selective constraint against non-coding loci. Constraint metrics, such as LOEUF³² quantify a gene’s intolerance to loss-of-function mutations and have already proven valuable tools in gene discovery and variant classification. Identifying non-coding regions intolerant of mutation has historically proven to be more difficult due to limited sample sizes, lack of precise variant effect models and heterogeneous mutation rates influenced by broader genomic features³³. A recent effort leveraged over 76,000 individuals with srGS from gnomAD and computed constraint genome-wide³⁴. Moreover, making use of large studies of structural and copy number variants has improved measures of dosage sensitivity, including identifying triplosensitive genes²⁸. As the sample size of reference populations increases, these metrics will be further refined, including increased resolution for the coding regions most intolerant to missense mutations^35,36 and non-coding regions. GREGoR is translating knowledge of these constrained genes and regions to rare disease cases, with a focus on identifying clinically significant non-coding variants.

GREGoR is developing visualization tools for structural and copy-number variants²⁵ by repurposing read-depth from srGS to mimic single-nucleotide polymorphism arrays to achieve resolution as low as 1 kb—beyond the 5 kb limit of the current standard using array comparative genomic hybridization. Thus, srGS could potentially serve as a cost-effective, first-line, unifying assay by simultaneously replacing both arrays and exomes and enable more accurate, nucleotide-resolution breakpoints of structural variants, which have historically been critical in deciphering mechanisms of genomic rearrangements. Most breakpoints including published structural variants, especially in highly repetitive or hypermutable regions of the genome, lack validation at the nucleotide resolution, which is relevant for genomic assembly and hypothesis-driven inferences of structural variant impact on gene expression. Furthermore, GREGoR investigators are showing local sequences surrounding candidate and pathogenic variants can offer insights into secondary structure mutagenesis and other mechanisms of genomic disorders^37–39. GREGoR’s work to identify diagnoses from srGS emphasizes unrealized potential of both primary analysis and reanalysis of srGS for rare disease discoveries.

Long-read sequencing

Long-read sequencing has opened new diagnostic opportunities in rare diseases. GREGoR and others have demonstrated that targeted long-read sequencing can reveal variants, particularly structural variants spanning repetitive sequences, in both known and novel disease genes that are missed or difficult to detect by short-read sequencing^40–44. Targeted sequencing panels can evolve beyond coding regions to include untranslated regions, promoters, intronic, intergenic and large expanses of non-coding regions around known disease genes. For example, GREGoR in collaboration with Twist developed the Twist Alliance Dark Genes Panel to produce phased variants across 389 medically relevant and complex autosomal genes, where short-read sequencing often fails⁴⁵.

A key focus for GREGoR is comparing the molecular diagnostic yield and cost-effectiveness of short-read versus long-read sequencing. In multiple studies, long-read sequencing uncovered novel candidate variants and genes missed by short-read sequencing, including de novo, compound heterozygous, structural and epigenetic variants^44,46,47. In comparison to short-read sequencing, long-read sequencing offers advantages such as better phasing, improved understanding of haplotype blocks and methylation analysis. GREGoR researchers have been leveraging these advantages by developing tools⁴⁸ to use methylation data for phasing and investigating the diagnostic yield improvements from genome-wide DNA methylation arrays in relation to long-read sequencing⁴⁹. Moreover, GREGoR is developing tools for improved annotation for kilobase and megabase scale variants especially using long-read sequencing with a focus on mosaic structural variants^50,51. Furthermore, innovative computational tools in long-read sequencing variant calling and analysis^51,52 are being developed, including de novo variant callers and long-read pipelines for mitochondrial variant calling.

Further long-read sequencing can offer multi-omic insights beyond traditional DNA sequencing and is increasingly being applied to find and understand molecular diagnoses as well as mechanisms in rare diseases. One such technology is fiber-seq, which uses long-read sequencing to simultaneously evaluate the primary DNA sequence with a nucleotide-resolution view of the surrounding chromatin architecture^53,54. GREGoR is developing unifying assays looking to simultaneously assay the genome, methylome, epigenome and transcriptome to identify and understand mechanisms of rare diseases⁵⁵. Such unifying assays help to explain previously elusive variants⁵⁶ that may have been visible on exome or short-read genome sequencing but may not have been nominated as candidate variants. The comparison and contrast of multiple layers of -omics in a single, unifying assay mechanistically implicates the pathogenicity of these candidates, especially for non-coding variants.

Multi-omics

In rare disease diagnosis, methylome and transcriptome data have broadly demonstrated their utility by identification of outlier events in methylation⁵⁷, splicing or gene expression implicating pathogenic variants. GREGoR has focused on complementing hundreds of cases with methylation and transcriptome data to facilitate development of standards and new computational methods. For example, recent activities in GREGoR have demonstrated how combined transcriptome and long-read genome analyses can aid in prioritizing structural variants when allele frequency information is limited⁵⁸.

A focus of GREGoR has been creating data that enable evaluation and prioritization of multi-omic assays for rare disease diagnosis. Currently, the use of multi-omics is predominantly limited to research and limited information exists to suggest which post-genome, -omics assay would yield the most useful information. To address this challenge, GREGoR has been generating a squared-off matrix for a subset of families, for whom long-read genome, methylation, chromatin-accessibility, transcriptome, proteome and metabolome data are being collected. Complementing these data, GREGoR has been supporting development and integration of reference -omics data from the Common Fund Data Ecosystem to advance outlier detection for various -omics assays by integrating larger control datasets. Multiple efforts in GREGoR are facilitating more routine use of these data such as updates to the seqr platform to expand intake of multi-omics data types to enable routine linking of outliers in these data to underlying genomic variation⁵⁹.

Reframing rare disease analysis

Reference genomes

Adoption of new reference genomes has lagged in clinical settings. Despite the publication of the GRCh38 human reference genome over a decade ago, many clinical labs to date still use GRCh37 (ref. 60). Part of the entrenchment of GRCh37 was the lag in necessary infrastructure development to support allele frequencies, in silico scores, bioinformatic tools and clinical databases on GRCh38. Thus, to date, the vast majority of known clinical disease genes and phenotypic expansions were discovered using GRCh37. GREGoR has shown the reference genome alone impacts variant calling in around 1% of the exome, with 206 genes enriched in discordant calls, including 8 known disease genes⁶¹. GREGoR showed that these discrepancies were more pronounced at the RNA level, with 1,492 genes demonstrating reference-dependent quantification and 3,377 genes exhibiting reference-exclusive expression, affecting 512 known disease genes⁶². GREGoR investigators have focused on fixing the GRCh38 reference⁶³, benchmarking medically relevant genes for both GRCh37 and GRCh38⁶⁴, and resolving pathogenic inversions in reference genome gaps using the telomere-to-telomere (T2T) reference genome⁴⁷.

An important question for the field focuses on whether there will be development of flexible pipelines and tools capable of using the newest references, such as T2T and the pangenome⁶⁵. GREGoR, in collaboration with Illumina, has benchmarked the DRAGEN pipeline, which uses graph-based alignment among many other novel features for variant calling in srGS⁶⁶. Looking ahead, GREGoR is collaborating with the Human Pangenome Reference Consortium (HPRC) through methods development like the Pangenome Research Tool Kit to demonstrate accurate variant calling in regions of the genome previously too complex for accurate variant calling. Further, GREGoR investigators are using pangenome approaches to understand complex tandem repeats in known disease genes⁶⁷ and exploring the infrastructure necessary for widespread adoption of the pangenome in clinical settings.

The hardest molecular diagnoses

Over time, consistent themes have emerged in the life cycle of disease gene discovery. Typically, the first and easiest candidate variants implicated are de novo and/or predicted loss-of-function (pLOF) variants that segregate with the phenotype in a pedigree. These two variant classes have logical frameworks supporting their putative mechanisms. A de novo variant, found in an affected proband but not in the unaffected parents, significantly increases the probability of being causative, especially when other de novo cases show the same clinical phenotypes^68,69. pLOF variants such as nonsense single-nucleotide variants, frameshifting insertions or deletions, and splice-site altering variants imply a null effect, because many of these pLOF variants are expected to be caught by the mRNA surveillance mechanism known as nonsense-mediated decay (NMD). NMD is a highly sensitive mechanism that is present in all tissues and destroys faulty transcripts with premature stops with high efficiency and fidelity^70,71. This same mechanistic reliability explains why pLOF variants are often the first variant type to be implicated in a novel gene-to-disease discovery, because clinical geneticists can reliably infer the presence of a pLOF variant will putatively lead to destruction of the faulty RNA transcript by NMD and no protein production, which is in alignment with a potential pathogenic mechanism of loss of function. Subsequently, other single-nucleotide and structural variants in the same gene or region are often implicated after a quorum of cases is established, although different types or locations of variation in the same gene can result in distinct clinical phenotypes. Furthermore, once one gene has been implicated, it serves as a seed for other genes in the same protein complex⁷², protein pathway³⁸ or gene family^73,74 to be implicated in the same or similar clinical phenotypes.

Currently, Online Mendelian Inheritance in Man (OMIM) and the Gene Curation Coalition (GenCC) have documented over 4,500 genes as being implicated in at least one Mendelian condition^75,76. Most of these discoveries were achieved through sequencing and interpretation of primary DNA variation contextualized by a proband’s phenotypes without requiring additional -omics or integrative analyses. While many thousands more disease genes can still be discovered using these same established gene-discovery principles, GREGoR is pursuing cases that were unsolved by standard clinical genetic testing and are hypothesized to have among the rarest and most difficult to detect or interpret pathogenic variation, which often requires integration of multiple -omics to (1) discover a candidate variant refractory to traditional sequencing methods; (2) provide proper context to interpret a variant that may have been seen in the primary DNA sequence but for which there was not enough understanding to nominate the variant as a candidate; or (3) provide orthogonal validation of a candidate discovered in the primary DNA sequence but for which the interpretation was speculative at best. In Box 2, we discuss lessons learned during GREGoR, and below we discuss ten of the rarest and hardest molecular diagnoses pursued by GREGoR investigators:

Box 2. Lessons learned from GREGoR.

(1). Custom rare disease data model.

GREGoR has developed a data model emphasizing rare-disease research essentials such as accessibility, consent consistency and transparency, and use of accepted ontologies and common standards. Every variant in the GREGoR joint callset is machine-readable with both unique Global Alliance for Genomics and Health (GA4GH) Variant Representation Specification (VRS) IDs¹²⁹ and ClinGen allele IDs¹³⁰. The data model accommodates variants and output files from a wide variety of genomic, multi-omic, phenotypic and molecular data types, and is modularly designed to support the integration of future data types. GREGoR is also working with the GA4GH to contribute to global standards for the collection and sharing of rare disease data, including the transfer of rare disease phenotype data from electronic health records.

(2). Deep phenotyping and building a truth set.

Studying phenotypic heterogeneity in the context of genetic heterogeneity is critical to solving unsolved Mendelian disease. Assignment of Human Phenotype Ontology (HPO)¹³¹ terms is a mandatory requirement for GREGoR data collection to allow end users to link all possible genotypes to all possible phenotypes. Moreover, GREGoR is developing algorithms to quantitatively dissect genotypic heterogeneity and phenotypic heterogeneity especially for MPV^{39,88,132–137}. Furthermore, GREGoR is building large language models for optimal phenotypic extraction from electronic health records¹³⁸. GREGoR hopes to fulfil the unmet need for a truth set of deeply phenotyped rare disease cases linked to well described multimodal genotype data whereby solved cases can be used as positive controls for benchmarking new tools and focused challenges can be created to examine exome-negative, unsolved cases.

(3). Building reference databases.

GREGoR is using long-read sequencing and optical mapping from diverse individuals to create benchmarks and a database of structural variants for filtering and prioritizing candidates. Specifically, GREGoR researchers started the 1000 Genomes Project Oxford Nanopore Technologies (ONT) Sequencing Consortium¹³⁹ and recently released the first 100 samples of long-read data from diverse populations. Furthermore, GREGoR investigators have established a population reference of tandem repeat expansions across ancestries from over 330,000 short-read genomes in TOPMed, UK Biobank, Estonian Biobank, 1000 Genome Project and All of Us, for interpreting repeat-driven genetic diseases¹⁰¹. Also GREGoR investigators have curated over 1.1 billion unique genomic variants from over 247,000 short-read genomes in All of Us v.7.1 for population allele frequency annotation and seamless integration into clinical workflows through the mass annotators of Ensembl variant effect predictor (VEP)¹⁴⁰, dbNSFP¹⁴¹, Annovar¹⁴² and OpenCRAVAT¹⁴³.

(4). Matchmaking beyond genes.

GREGoR is leveraging federated variant-level matchmaking with GA4GH standards through VariantMatcher¹⁴⁴, MyGene2, Geno2MP and seqr⁵⁹, with the hope of accelerating disease-causing variant discovery beyond the exome. Moreover, many clinical laboratories have undiagnosed cases potentially explained by novel gene discoveries or phenotypic expansions. GREGoR has actively engaged with clinical labs to study effective strategies for exome and genome analysis without overburdening variant analysts¹⁴⁵. GREGoR has released recommendations for clinical labs to report variants in novel candidate genes and support follow-up investigations and data sharing, enabling broad discoveries and patient diagnoses¹⁴⁶.

1. Non-coding variants occur in genomic regions that do not code for proteins, such as non-coding RNAs, promoters, enhancers and untranslated regions. Pathogenic non-coding variants can disrupt mechanisms such as gene regulation, splicing, expression, translation or RNA stability. Genome sequencing is crucial for detecting variants in non-coding regions, but techniques like chromatin immunoprecipitation with sequencing (ChIP–seq), assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq), RNA sequencing (RNA-seq), fiber-seq and massively parallel reporter assays (MPRAs) can help to identify and provide mechanistic explanation for potential pathogenic non-coding variation. For example, non-coding variants such as deep intronic variants may be involved in alterations in splicing, whereby exons or introns may be incorrectly skipped or included during mRNA processing leading to changes in the final transcript potentially leading to clinical phenotypes⁷⁷. While these variants are often visible in primary DNA sequencing, their interpretation is typically speculative without orthogonal validation such as reverse transcription quantitative PCR, RNA-seq or mini-gene splicing assays to validate whether or not aberrant splicing occurred by understanding the sequence of the processed mRNA. Similarly, long-read RNA-seq is particularly useful for detecting full-length transcripts and interpreting complex splicing patterns. Work by GREGoR and others has shown that non-coding variants can be the ‘missing variant’ in trans with a pathogenic coding variant for recessive rare diseases^78,79. GREGoR is pursuing generalizable methods for understanding non-coding variant effects at scale^80,81.

2. Non-coding genes produce molecules that perform potential regulatory, structural or catalytic roles rather than encode proteins. These include rRNA, tRNA, microRNA, long non-coding RNA, small nuclear RNA and more, and perturbations in non-coding genes may cause rare genetic diseases. A flagship example is perturbations in RNU4–2, in which cases from multiple GREGoR sites contributed to rapid discovery and progress to publication, establishing perturbations in RNU4–2 as one of the most commonly mutated causes of neurodevelopmental disorders^82,83. Even though RNU4–2 is a non-coding RNA, it’s discovery timeline is analogous to the typical gene discovery life cycle. The main cases found were de novo insertions at the same site in a large cohort with phenotype-matching cases. RNU4–2 has now served as a seed for more RNU4 minor spliceosome genes being implicated in neurodevelopmental disorders^84,85. Another example from GREGoR is CHASERR, a long non-coding RNA adjacent to CHD2 which was implicated in developmental and epileptic encephalopathy. The CHASERR discovery was a strong collaboration with the father of the initial proband serving as a coauthor on the manuscript, highlighting the power of patient partnerships in accelerating rare disease genomics⁸¹. Furthermore, using fiber-seq, GREGoR investigators have identified STRTS, an intergenic locus implicated in congenital hypothyroidism⁵⁶. These findings illustrate the diagnostic potential of non-coding regions of the genome, which are not systematically included in standard variant analysis workflows. With thousands of non-coding transcripts still poorly understood, GREGoR continues to explore this untapped reservoir of genomic information, paving the way for novel disease-gene discoveries in the non-coding space.

3. Multilocus pathogenic variation (MPV) refers to the presence of multiple, independently pathogenic variants in multiple genes or loci that collectively contribute to an individual’s clinical manifestation. These variations can interact in complex ways, leading to compounded effects that may influence disease severity, onset or progression, often complicating diagnosis and treatment. GREGoR researchers have shown that as many as 5% of individuals for whom molecular diagnoses have been ascertained have MPV⁸, with this rate being even higher in rare disease families with parental consanguinity^21,86. While individual variants comprising MPV are typically routinely diagnosed from just DNA sequencing, the interpretation of MPV typically requires deeper phenotypic analyses and represents a much broader area of gene dosage models for disease causing variation. The multiple de novo copy-number variant (MdnCNV) phenotype is a form of MPV whereby four or more independent, constitutional de novo copy-number variants arise in the same person within one generation. GREGoR researchers have shown that this ultrarare phenotype occurs in around 1 in over 12,000 individuals referred for genome-wide chromosomal microarray analysis^87,88. MdnCNV typically requires integration of multiple technologies including arrays, short-read and long-read sequencing as well as quantitative phenotyping to fully characterize the genomic and clinical impact.

4. Complex genomic rearrangements (CGRs) are kilobase to megabase-scale structural changes in the genome involving multiple breakpoints, rearrangements and/or the integration of novel sequences in cis resulting in duplications, deletions, inversions and translocations, often affecting gene function and regulation. While CGRs have been catalogued and shown to be abundant across diverse populations, the accurate detection and assembly of CGRs, especially their breakpoints, typically requires more than short-read DNA sequencing^28,29,89,90. Long-read sequencing (including adaptive sampling and ultra-long reads), optical genome mapping, chromosomal microarray analysis and linked-read sequencing can help to provide nucleotide resolution for CGRs to elucidate their mechanisms of formation to better understand how they precipitate genomic disorders and contribute to clinical variability and disease severity^91–93.

5. Tandem repeats are sequences in which a nucleotide motif is repeated consecutively a varying number of times. These repetitive regions can be unstable, leading to expansions or contractions, which are associated with several genetic disorders^94,95. Owing to the potential for multi-mapped short reads, only shorter tandem repeats have typically been well detected from short-read sequencing technologies^96,97. However, long-read sequencing technologies are particularly effective at completely spanning short and longer tandem repeats, enabling accurate determination of repeat length and structure. Moreover, specialized bioinformatic tools designed for repeat analysis can help in accurately calling tandem repeats and identifying pathogenic expansions or contractions^98–100. GREGoR is collating large databases and truth sets of tandem repeats across diverse populations to enable more systematic integration of tandem repeat analysis into sequencing pipelines^101,102.

6. Mosaic variation originates from post-zygotic mutations and is considered germline if confined to the germ cells or somatic if acquired during or after the first mitotic divisions. Mosaic variation can lead to variations in phenotype, depending on the proportion and distribution of the mutant cells across tissues. Detecting mosaic variation often requires more than standard DNA sequencing due to the low variant allele fraction of mosaic alleles. Reliable detection of mosaic variation typically requires high-read-depth sequencing and orthogonal techniques such as digital droplet PCR for validation and bioinformatic discrimination between mosaic variants and sequencing artifacts. GREGoR is collaborating with the Somatic Mosaicism Across Human Tissues (SMaHT) consortium to understand pathogenic mosaic variation at scale.

7. Multi-nucleotide variants (MNVs) are two or more variants within the same codon on the same haplotype with over 50 MNVs per person¹⁰³. Accurately identifying MNVs is bioinformatically challenging as a single MNV may be incorrectly interpreted as multiple independent variants. Missing MNVs can alter the interpretation of clinically pathogenic variation such as nonsense single-nucleotide variants that do not lead to a premature stop codon introduction^104,105. Many variant effect predictors and multiplex assays of variant effects systematically score every possible MNV in a target locus. GREGoR is evaluating the use of these data towards variant of uncertain significance (VUS) reclassification and novel disease gene discovery.

8. NMD-escaping variants introduce premature stop codons in mRNA that evade NMD, producing truncated proteins of unknown function that may or may not manifest clinical disease. These variants are visible on primary DNA sequencing and because extensive work has been done to determine the rules of NMD escape^106,107, many of these variants are already implicated in clinical disease and can be speculated on from just primary DNA sequencing^70,71,108. However, newer approaches such as long read RNA-seq and proteomics are opening doors into the investigation of NMD escape alleles and their downstream mechanisms.

9. Incompletely penetrant variants refer to pathogenic changes in DNA that do not always result in observable clinical disease, even in individuals carrying the variant. This variability can complicate interpretation, especially in family studies where carriers appear to be unaffected. These variants often challenge diagnostic workflows because traditional penetrance assumptions do not hold, necessitating integration of orthogonal lines of evidence such as animal models and epigenetics to understand the variant pathogenicity¹⁰⁹. These functional efforts are particularly critical for diseases where penetrance may be age-dependent, sex-influenced or modified by external factors. Furthermore, on a case-by-case analysis across all of gnomAD v4, more than 95% of incompletely penetrant pLOF variants found in severe, early-onset, highly penetrant haploinsufficient disease had explainable causes such as a downstream frame-restoring variant, predicted reinitiation by a downstream methionine, an MNV changing the interpretation of a nonsense variant to a missense or synonymous variant, or the location of the pLOF variant being in an NMD escape region¹¹⁰.

10. VUSs are most often reported in genes already established in disease pathogenesis, although, by definition, all variants found in candidate genes with insufficient evidence for disease implication are also VUSs. VUSs are accumulating rapidly over time as testing volume expands. In fact, VUSs are more often reported during panel testing compared with exome or genome sequencing due to professional practices¹¹¹ and are disproportionately called in individuals from non-European ancestries¹¹². Thus, while transitioning to consistent first-line clinical usage of exome or genome sequencing coupled with phenotypic analyses will decrease the rate of clinically reported VUSs, given most causal variants are only found in a single individual, integration of functional modelling is often required to reclassify VUSs. Multiplexed assays of variant effects (MAVEs) are high-throughput experiments orthogonal to the clinical sequencing pipeline that produce functional scores for all variant effects in a target locus and when their evidence strength is clinically calibrated and incorporated with other lines of evidence, demonstrate significant promise in massive VUS reclassification¹¹².

Recruitment and return of results

Participation in rare-disease research can be influenced by numerous factors, including but not limited to phenotypic areas of focus or exclusion, as well as institutional, socioeconomic, geographical, linguistic, cultural, educational and insurance factors^113–116. GREGoR sites, many of which are in urban centres, have implemented procedures for online enrolment, chatbots¹¹⁷, remote consent and offsite sample collection (including mobile phlebotomy in rural areas and for individuals with transportation challenges), and translated materials in multiple languages to improve accessibility. Attention to these details has enabled GREGoR to foster many international collaborations with local scientists to sequence and make available sequencing from thousands of individuals of non-European ancestry, actively seeking to address the disparities in genomic data availability across Middle Eastern, North African, Southeast Asian, South American and other under-represented groups such as African-American and Hispanic peoples. GREGoR is focused on solving unsolved cases, meaning exome-negative cases are prioritized and different sites will triage on a case-by-case basis for different -omic technologies including new exome and genome evaluations. While return of results varies per site and per case, GREGoR research centres have each developed rigorous standards and processes for return of high-confidence research results to participants and/or local clinicians who follow those participants. Attention is given to educating families on the distinction between ‘research results’ and ‘clinical testing results’, and centres have developed workflows to support participants in obtaining College of American Pathologists (CAP)/Clinical Laboratory Improvement Amendments (CLIA)-certified clinical laboratory confirmation of research findings when possible and appropriate.

Genomics for all

Diversifying genomics research through participant recruitment from under-represented populations is just one approach to fostering equity. GREGoR is pursuing orthogonal approaches to increase access to a molecular diagnosis by pursuing improved variant calling methods, applying multiplexed functional assays to improve interpretation of VUSs that are enriched in under-represented populations, and testing technologies and workflows that increase access to a genetic diagnosis for everyone. For example, GREGoR is collaborating with the HPRC to improve variant calling accuracy across all populations through the pangenome. GREGoR has analysed population biobanks to show a higher prevalence of VUSs and fewer ‘pathogenic’ or ‘likely pathogenic’ classifications in individuals of non-European genetic ancestry¹¹². These disparities were alleviated by using high throughput, multiplexed functional experiments to test every possible single variant in genes of interest to resolve VUS disparities between populations. However, this study demonstrated that allele frequency and variant effect predictors contribute to unbalanced classification of variants and more work to prevent incorrect clinical variant classification is an important future priority for GREGoR. Beyond addressing disparities, the power of special and under-represented populations to make outsized contributions to rare disease genomics is unique and well established^21,118. Distinct genetic phenomena—such as founder mutations in population isolates, large stretches of absence of heterozygosity in individuals from consanguineous populations, and private or ultrarare variants from individuals of ancestrally diverse backgrounds—often provide key genetic insights that lead to the discovery of disease–gene relationships and extrapolate to solving unsolved cases worldwide^119–123. Finally, SeqFirst¹²⁴ shows that using simple criteria to assess eligibility for rapid srGS significantly increases the proportion of non-white and Black infants who receive a precise genetic diagnosis. GREGoR and SeqFirst are conducting a comparison of long-read versus short-read sequencing in SeqFirst to better understand the relative value of these technologies within populations.

Accelerating data sharing

Data sharing is critical to advancing rare disease diagnoses^125,126. GREGoR is committed to rapid release of genomic and phenotypic data to the larger research community within AnVIL and making these data findable, accessible, interoperable and reusable (FAIR)¹²⁷ and machine readable. Inside AnVIL, GREGoR data are queryable through seqr, which integrates variant filtration, annotation and causal variant identification. Outside AnVIL, GREGoR has developed a public variant browser, which already includes over 95 million variants. Furthermore, de-identified phenotypes are being added to the public browser, enabling researchers to easily explore putative genotype–phenotype relationships in rare disease families. At the time of AnVIL submission and before analysis, deep phenotyping (Fig. 2a), potentially multiple orthogonal lines of -omics data (Fig. 2b,c) and pedigree data (Fig. 2d) are made available through the GREGoR data model for broader dissemination to the research community. Currently, DNA data on approximately 7,400 individuals from over 3,000 families are available with transcriptome data available for over 500 individuals and nearly 200 participants with both exome and short-read genome data (dbGaP: phs003047; Fig. 2b). Planned releases include additional short- and long-read genomes, short- and long-read RNA-seq, fiber-seq, ATAC-seq, metabolomics and proteomics. GREGoR has identified candidate discoveries and molecular diagnoses for over 400 families (Fig. 2e).

Fig. 2 |. Overview of publicly released GREGoR data.

Summary of third public data release (dbGaP: phs003047). a, The distribution of the top 30 phenotypes in GREGoR based on HPO descriptions. b, Table of numbers for probands and total individuals for each sequencing modality. c, The overlap across short-read genomes, RNA-seq and long-read genomes in data generation. d, Family structures comprising the overall cohort from a total of n = 3,610 families. e, Summary of current solved cases. Data are shared before analysis, but even the current diagnostic outcomes underscore the challenges and opportunities in resolving rare disease cases that are previously exome negative. WGS, whole-genome sequencing.

All GREGoR candidate genes are shared to Matchmaker Exchange through either GeneMatcher, seqr or MyGene2. Notably, for the 83 GREGoR publications (Supplementary Table 1) involving novel disease genes or phenotypic expansions, almost every project has been influenced by findings from connections made across or within Matchmaker Exchange nodes, whereas only 44 were supported by orthogonal functional experiments. Novel candidate genes and phenotypic expansions are curated for validity and publicly shared to the GenCC to accelerate access to early evidence of novel gene–disease relationships and aid in standardized clinical diagnostics and research. Analogously, candidate variants and molecular diagnoses are deposited in ClinVar.

Conclusion

Far from wrapping up the edges, these challenges represent a vast forefront in genomic research (Box 3), demanding both innovative methodologies and sustained collaboration to make meaningful progress. Alongside these challenges, advancements in genomic assays have required vetting at scale in individuals of diverse genetic ancestries and with diverse rare disease phenotypes. Such efforts are critical to establishing standards of when and how to use a specific approach and will guide expectations on their relative yields at scale and their adoption in clinical practice. Lastly, there is a palpable need to translate scientific discoveries into curation practices that align with formal clinical standards. To address these gaps, GREGoR provides data and infrastructure that will catalyse development and implementation of new approaches to advance genomics in rare disease by the broader community.

Box 3. The forefront of rare disease genomics.

(1). Finish the disease gene catalogue for protein-coding genes.

The effort to complete a comprehensive catalogue of genes underlying Mendelian conditions remains far from finished. While thousands of Mendelian conditions have been described, many of these conditions still lack a known genetic cause. Thousands of discoveries have been made using exome sequencing and hundreds of discoveries have been made in exome-negative cases using a variety of short-read or long-read genome sequencing, RNA-seq, novel analytical algorithms, new data sharing platforms and automated reanalysis strategies. The time is now to commit to a focused, unified, global, scaled effort to discover the remaining protein-coding disease genes and accelerate from a currently approximately 40% completed disease gene catalogue to a complete disease gene catalogue for protein-coding genes.

(2). Oligogenic and polygenic molecular diagnoses.

These involve the contribution of variants in two or more genes or loci that collectively lead to a clinical phenotype, a concept distinct from traditional monogenic inheritance patterns or MPV. These cases often present diagnostic challenges because the individual variants may not cause disease independently but act in concert to disrupt pathways or biological networks^147–149. Currently, we can identify oligogenic variants through DNA sequencing, but the full scope of oligogenic and polygenic molecular diagnoses may require analysis across multiple layers of multi-omics data. For example, combinations of transcriptomic, proteomic or metabolomic alterations may converge with DNA variation to create synergistic effects that contribute to disease, representing a new frontier for rare disease genomics.

(3). Integrating large- and small-effect variants.

Currently, molecular diagnoses are considered to be complete once large-effect variant(s) are identified that can explain the majority of the primary clinical phenotype. However, non-coding modifiers and other small-effect variants can contribute to variable expressivity and incomplete penetrance. Despite their influence on the clinical presentation, these variants are not incorporated into clinical molecular diagnoses. A next frontier in rare disease genomics should include discovery of systematic approaches for identifying and integrating small-effect, modifier variants into diagnostic interpretation. Just as clinical infrastructures have been developed to support the identification and return of large-effect variants, parallel infrastructures are needed to support the clinical return of relevant small-effect variants that meaningfully modify disease presentation.

(4). Building multi-omic momentum.

Several non-coding genes have now been implicated in rare diseases, including numerous small nuclear RNA genes from the RNU family, as well as long non-coding RNAs and microRNAs, among others. These discoveries emerged from cases that remained unsolved by exome sequencing and required more comprehensive approaches such as genome sequencing or transcriptomic analysis for resolution^{56,81–84,150}. Together, these studies demonstrate the quorum of investigations needed to build a compelling case for broader, scaled adoption of multi-omic methods—potentially as primary tools for rare disease diagnosis and discovery.

Supplementary Material

Additional information

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41586-025-09613-8.

GREGoR Partner Members

Aashish Adhikari⁴², Kinga M. Bujakowska⁴³, Claudia M. B. Carvalho²⁹, Ali Crawford⁴⁴, Aimée Dudley^29,45, Kelly D. Farwell Hagman⁴⁶, Yang I. Li⁴⁷, Jill E. Moore⁴⁸, Aaron R. Quinlan⁴⁹, Alex H. Wagner^25,26,27, Bo Xia⁵⁰ & S. Stephen Yi^51,52

⁴²Illumina Artificial Intelligence Laboratory, Illumina, Foster City, CA, USA. ⁴³Department of Ophthalmology, Ocular Genomics Institute, Massachusetts Eye and Ear, Harvard Medical School, Boston, MA, USA. ⁴⁴Medical Genomics Research, Illumina, San Diego, CA, USA. ⁴⁵Molecular and Cellular Biology Program, University of Washington, Seattle, WA, USA. ⁴⁶Ambry Genetics, Aliso Viejo, CA, USA. ⁴⁷Department of Medicine, University of Chicago, Chicago, IL, USA. ⁴⁸Department of Genomics and Computational Biology, University of Massachusetts Chan Medical School, Worcester, MA, USA. ⁴⁹Department of Human Genetics, University of Utah, Salt Lake City, UT, USA. ⁵⁰Gene Regulation Observatory, Broad Institute of MIT and Harvard, Cambridge, MA, USA. ⁵¹Department of Neurosurgery, Baylor Research Institute, Baylor College of Medicine, Temple, TX, USA. ⁵²Department of Medicine, School of Medicine—Temple, Baylor College of Medicine, Temple, TX, USA.

Genomics Research to Elucidate the Genetics of Rare Diseases (GREGoR) Consortium

Moez Dawood^1,2,3, Ben Heavner⁴, Marsha M. Wheeler⁴, Rachel A. Ungar^5,6,7, Jonathan LoTempio⁸, Laurens Wiel^5,6,9, Claudia M. B. Carvalho²⁹, Richard A. Gibbs^1,2, Casey A. Gifford^5,30,31,32, Susanne May⁴, Danny E. Miller^15,16,33,34, Heidi L. Rehm^20,21, Kaitlin E. Samocha^20,21, Fritz J. Sedlazeck^1,2,35, Eric Vilain⁸, Anne O’Donnell-Luria^20,21,36, Jennifer E. Posey^2,37,38, Lisa H. Chadwick³⁹, Michael J. Bamshad^14,15,40, Stephen B. Montgomery^5,6,41

U01HG011758

Jennifer E. Posey^2,37,38, Richard A. Gibbs^1,2, James R. Lupski^1,2,18, Hatoon Al Ali², Elizabeth G. Atkinson², Sairam Behera¹, Shaghayegh T. Beheshti², Eric Boerwinkle⁵³, Tugce Bozkurt-Yozgatli^53,54, Daniel G. Calame⁵⁵, Ivan Chinn⁵⁶, Zeynep H. Coban-Akdemir^2,53, Karen J. Coveler², Zain Dardas^2,46, Moez Dawood^1,2,3, Harsha Doddapaneni¹, Haowei Du², Ruizhi Duan², Iman Egab⁵³, Jawid Fatih², Mira Gandhi⁵⁵, Brandon Garcia², Nikhita Gogate², Christopher M. Grochowski², Jianhong Hu¹, Minal Jamsandekar², Shalini N. Jhangiani¹, Angad Jolly^2,57, Parneet Kaur², Ahmed K. Saad², Jesse M. Levine⁵⁵, Richard A. Lewis^2,18,58,59, Yidan Li², Pengfei Liu², Medhat Mahmoud¹, Dana Marafi^60,61,62, Tadahiro Mitani⁶³, Chloe Munderloh², Donna Muzny^1,2, Sebastian Ochoa¹⁸, Piyush Panchal¹, Shruti Pande², Davut Pehlivan^55,64,65, Archana Rai⁵³, Edgar Andres Rivera-Munoz², Aniko Sabo¹, Evette Scott¹, Fritz J. Sedlazeck^1,2,35, Vernon Reid Sutton², Kimberly Walker¹, Lauren Westerfield², Jiaoyang Xu⁵³, Bo Yuan^1,2,66 & Xinchang Zheng¹

⁵³Human Genetics Center, Department of Epidemiology, Human Genetics, and Environmental Sciences, School of Public Health, The University of Texas Health Science Center at Houston, Houston, TX, USA. ⁵⁴Department of Biostatistics and Bioinformatics, Acibadem Mehmet Ali Aydinlar University, Istanbul, Turkey. ⁵⁵Department of Pediatrics, Section of Pediatric Neurology and Developmental Neurosciences, Baylor College of Medicine, Houston, TX, USA. ⁵⁶Department of Pediatrics, Division of Immunology, Allergy, and Retrovirology, Baylor College of Medicine, Houston, TX, USA. ⁵⁷Department of Neurology, The University of Texas Dell Medical School, Austin, TX, USA. ⁵⁸Department of Ophthalmology, Baylor College of Medicine, Houston, TX, USA. ⁵⁹Department of Medicine, Baylor College of Medicine, Houston, TX, USA. ⁶⁰Department of Pediatrics, College of Medicine, Kuwait University, Safat, Kuwait. ⁶¹Section of Child Neurology, Department of Pediatrics, Adan Hospital, Ministry of Health, Hadiya, Kuwait. ⁶²Kuwait Medical Genetics Centre, Ministry of Health, Sulaibikhat, Kuwait. ⁶³Department of Pediatrics, Jichi Medical University, Shimotsuke, Tochigi, Japan. ⁶⁴Texas Children’s Hospital, Houston, TX, USA. ⁶⁵Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, TX, USA. ⁶⁶Baylor Genetics, Houston, TX, USA.

U01HG011755

Anne O’Donnell-Luria^20,21,36, Heidi L. Rehm^20,21, Michael E. Talkowski^{20,21,22,23,24}, Siwaar Abouhala²¹, K. D. Ahlquist²¹, Mutaz Amin²¹, Christina Austin-Tse^20,21, Samantha M. Baxter²¹, Benjamin Blankenmeister²¹, Philip M. Boone^20,21,36, Harrison Brand^20,21,22, Colleen Carlston^21,36, Celine de Esch^20,21, Stephanie DiTroia²¹, Michael Duyzend^20,21,36, Vijay Ganesh^21,67, Kiran Garimella²¹, Carmen Glaze²¹, Emily Groopman²¹, Sanna Gudmundsson²¹, Stacey Hall²¹, Yongqing Huang²¹, Julia Klugherz²¹, Katie Larsson²¹, Arthur S. Lee^20,21,22, Gabrielle Lemire²¹, Jialan Ma²¹, Daniel MacArthur^21,68,69, Brian Mangilog²¹, Daniel Marten^21,36, Eva Martinez²¹, Olfa Messaoud²¹, Chloe Mighton²¹, Mariana Moyses²¹, Ashana Neale²¹, Emily O’Heir²¹, Melanie C. O’Leary²¹, Ikeoluwa Osei-Owusu²¹, Lynn Pais²¹, Alicia Pham²¹, Lindsay Romo²¹, Kathryn Russell²¹, Monica Salani^20,21, Kaitlin Samocha^20,21, Alba Sanchis-Juan^20,21,22, Jillian Serrano²¹, Gulalai Shah²¹, Moriel Singer-Berk²¹, Mugdha Singh^21,36, Hana Snow²¹, Kayla Socarras²¹, Sarah L. Stenton^21,36, Jui-Cheng Tai^20,21,22, Grace VanNoy²¹, Ben Weisburd²¹, Michael Wilson²¹, Monica Wojcik^21,36,70, Isaac Wong²¹ & Rachita Yadav^20,21,22

⁶⁷Department of Neurology, Brigham and Women’s Hospital, Boston, MA, USA. ⁶⁸Centre for Population Genomics, Garvan Institute of Medical Research, Sydney, New South Wales, Australia. ⁶⁹Centre for Population Genomics, Murdoch Children’s Research Institute, Melbourne, Victoria, Australia. ⁷⁰Department of Pediatrics, Division of Newborn Medicine, Boston Children’s Hospital, Boston, MA, USA.

U01HG011762

Stephen B. Montgomery^5,6,41, Jonathan A. Bernstein¹³, Matthew T. Wheeler⁹, Emily Alsentzer⁴¹, Taylor M. Arriaga⁵, Euan A. Ashley^5,9,41, Themistocles Assimes⁹, Gill Bejerano^13,71,72, Devon Bonner¹³, Denver Bradley⁵, Jennefer Carter⁹, Clarisa Chavez Martinez⁵, Ziwei Chen⁷¹, Salil Deshpande⁷³, Sara Emami⁹, Ivy Evergreen⁵, Casey A. Gifford^5,30,31,32, Page Goddard⁶, John Gorzynski⁹, William Greenleaf⁵, Rodrigo Guarischi-Sousa⁹, Caitlin Harrington¹³, Sohaib Hassan⁷⁴, Tanner D. Jensen⁵, David Jimenez-Morales⁹, Christopher Jin⁵, Aimee Juan⁵, Jessica Kain⁵, Laura Keehan¹³, Anshul Kundaje^5,71, Soumya Kundu⁷¹, Samuel Lancaster⁵, Shruti Marwaha⁹, Dena R. Matalon¹³, Lauren Meador⁶, Hector Rodrigo Mendez⁹, Alexander Miller⁶, Matthew B. Neu³⁰, Thuy-mi P. Nguyen⁶, Jonathan Nguyen⁶, Jeren D. Olsen⁶, Evin M. Padhi⁶, Paul Petrowski⁶, Astaria D. Podesta⁶, Elizabeth Porter¹³, Wanqiong Qiao⁶, Thomas Quertermous⁹, Chloe M. Reuter⁹, Oriane Rubio¹³, Stuart A. Scott⁶, Riya Sinha⁷¹, Kevin S. Smith⁶, Michael P. Snyder⁵, Brigitte Stark⁶, Suchitra Sudarshan⁵, Raquel L. Summers⁹, Christina G. Tise¹³, Philip Tsao⁹, Rachel A. Ungar^5,6,7, Isabella Voutos⁹, Juliana M. Walrod⁹, Ziming Weng⁶, Laurens Wiel^5,6,9, Frank Wong⁵, Yao Yang⁶, Jiye Yu⁹ & Jimmy Zhen⁹

⁷¹Department of Computer Science, Stanford University, Stanford, CA, USA. ⁷²Department of Developmental Biology, School of Medicine, Stanford University, Stanford, CA, USA. ⁷³Institute of Computational and Mathematical Engineering, Stanford University, Stanford, CA, USA. ⁷⁴Department of Biomedical Data Science, School of Medicine, Stanford University, Stanford, CA, USA.

U01HG011745

Eric Vilain⁸, Seth Berger^10,11,12, Emmanuèle C. Délot⁸, Miguel Almalvez⁷⁵, Light Auriga⁸, Rebekah Barrick⁷⁶, Sami Belhadj⁴⁶, Krista Bluske⁴⁶, Leandros Boukas¹¹, Andrea J. Cohen¹⁰, Ya Cui⁷⁷, Ivan De Dios⁷⁵, Meghan Delaney⁷⁸, John Harting⁴⁶, Yun-Hua Hsiao⁴⁶, Rachid Karam⁴⁶, Charles Hadley King⁸, Arthur Ko¹¹, Wei Li⁷⁷, Bojan Losic⁴⁶, Jonathan LoTempio⁸, Georgia Pitsava⁸ & Changrui Xiao⁷⁹

⁷⁵Department of Pediatrics, University of California Irvine, Irvine, CA, USA. ⁷⁶Metabolic Disorders, Children’s Hospital Orange County (CHOC), Orange, CA, USA. ⁷⁷Department of Biological Chemistry, School of Medicine, University of California Irvine, Irvine, CA, USA. ⁷⁸Department of Pathology & Laboratory Medicine, Children’s National Hospital, Washington, DC, USA. ⁷⁹Department of Neurology, School of Medicine, University of California Irvine, Irvine, CA, USA.

U01HG011744

Michael J. Bamshad^14,15,40, Chia-Lin Wei¹⁶, Evan E. Eichler^16,17, Jessica X. Chong^14,15, Kailyn Anderson¹⁴, Peter Anderson¹⁶, Sabrina Best^15,16, Elizabeth E. Blue⁸⁰, Kati J. Buckingham¹⁴, Silvia Casadei^15,16, Yong-Han Hank Cheng¹⁶, Colleen P. Davis¹⁶, Sophia B. Gibson¹⁶, William W. Gordon¹⁴, Jonas Gustafson^14,45, William T. Harvey¹⁶, Martha Horike-Pyne⁸⁰, Gail P. Jarvik^16,80, Annelise Y. Mah-Som⁸⁰, Colby T. Marvin¹⁴, F. Kumara Mastrorosa¹⁶, Sean R. McGee¹⁶, Heather C. Mefford⁸¹, Danny E. Miller^15,16,33,34, Karynne Patterson¹⁶, Matthew Richardson¹⁶, Adriana E. Sedeño Cortés⁸⁰, Joshua D. Smith¹⁶, Olivia M. Sommerland¹⁴, Lea M. Starita^15,16, Andrew B. Stergachis^15,16,80, Elliott G. Swanson¹⁶, Jeffrey Weiss¹⁶, Qian Yi¹⁶, Christina Zakarian¹⁶ & Miranda P. Zalusky¹⁴

⁸⁰Department of Medicine, Division of Medical Genetics, University of Washington, Seattle, WA, USA. ⁸¹Cell & Molecular Biology, Center for Pediatric Neurological Disease Research, St Jude Children’s Research Hospital, Seattle, WA, USA.

U24HG011746

Susanne May⁴, Ali Shojaie^4,19, Emily Bonkowski⁴, Sarah Conner⁴, Matthew P. Conomos⁴, Stephanie M. Gogarten⁴, Ben Heavner⁴, Sarah C. Nelson⁴, Sheryl Payne⁴, Jaime Prosser⁴, Guanghao Qi⁴, Adrienne M. Stilp⁴, Catherine C. Tong⁴, Marsha M. Wheeler⁴ & Quenna Wong⁴

NHGRI Program Management

Lisa H. Chadwick³⁹, Christopher Wellington²⁸, Sara Currin³⁹ & Gabrielle C. Villard³⁹